Philosophy of Computer Science

Philosophy of Computer Science
William J. Rapaport
Department of Computer Science and Engineering,

Department of Philosophy, Department of Linguistics,
and Center for Cognitive Science
University at Buffalo, The State University of New York,
Buffalo, NY 14260-2000
rapaport@buffalo.edu
http://www.cse.buffalo.edu/rapaport/
DRAFT
c 20042015 by William J. Rapaport
November 23, 2015

2
Figure 1: 2015,
c ??
3
4
Contents
1 What Is Philosophy of Computer Science? 21

1.1 Recommended Readings: . . . . . . . . . . . . . . . . . . . . . . . . 22
1.2 What This Book Is About . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3 What This Book Is Not About . . . . . . . . . . . . . . . . . . . . . 24
1.4 . . . Further Sources of Information . . . . . . . . . . . . . . . . . . . 26
1.5 NOTES FOR NEXT DRAFT . . . . . . . . . . . . . . . . . . . . . . 29
2 What Is Philosophy? 31
2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 A Definition of Philosophy . . . . . . . . . . . . . . . . . . . . . . 33
2.4 What Is Truth? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.1 The Correspondence Theory of Truth . . . . . . . . . . . . . 35
2.4.2 The Coherence Theory of Truth . . . . . . . . . . . . . . . . 35
2.5 On Searching for the Truth vs. Finding It? . . . . . . . . . . . . . . . 36
2.6 What Is Rational? . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.6.1 Logical Rationality . . . . . . . . . . . . . . . . . . . . . . . 39
2.6.1.1 Deductive Logic . . . . . . . . . . . . . . . . . . . 39
2.6.1.2 Inductive Logic . . . . . . . . . . . . . . . . . . . 40
2.6.1.3 Abductive Logic . . . . . . . . . . . . . . . . . . . 40
2.6.1.4 Non-Monotonic Logic . . . . . . . . . . . . . . . . 41
2.6.2 Scientific Rationality . . . . . . . . . . . . . . . . . . . . . . 41
2.6.3 Is Science Philosophy? . . . . . . . . . . . . . . . . . . . . . 41
2.6.4 Could (Should?) Philosophy Be Scientific? . . . . . . . . . . 42
2.6.5 Is It Always Rational to Be Rational? . . . . . . . . . . . . . 43
2.7 What Is the Import of Personal Search? . . . . . . . . . . . . . . . 43
2.8 What Is the Import of In Any Field? . . . . . . . . . . . . . . . . . 43
2.10 Further Sources of Information . . . . . . . . . . . . . . . . . . . . . 58
3 What Is Computer Science? 63

3.1 Readings: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3 Why Ask What Computer Science Is? . . . . . . . . . . . . . . . . . 67
5
6 CONTENTS
3.3.1 Academic Motivations . . . . . . . . . . . . . . . . . . . . . 67

3.3.1.1 Academic Politics . . . . . . . . . . . . . . . . . . 67
3.3.1.2 Naming the Discipline . . . . . . . . . . . . . . . . 68
3.3.1.3 Academic Pedagogy . . . . . . . . . . . . . . . . . 68
3.3.1.4 Academic Publicity . . . . . . . . . . . . . . . . . 69
3.3.2 Intellectual or Philosophical Purposes . . . . . . . . . . . . . 69
3.4 What Does It Mean to Ask What Something Is? . . . . . . . . . . . . 70
3.5 Computer Science Studies Computers . . . . . . . . . . . . . . . . . 72
3.6 Computer Science Studies Algorithms . . . . . . . . . . . . . . . . . 75
3.7 Is Computer Science a Science? . . . . . . . . . . . . . . . . . . . . 76
3.8 Physical Computers vs. Abstract Algorithms . . . . . . . . . . . . . . 78
3.9 Computer Science Studies Information . . . . . . . . . . . . . . . . . 80
3.10 Computer Science Is a Science of Procedures . . . . . . . . . . . . . 81
3.11 Computer Science Is Not a Science . . . . . . . . . . . . . . . . . . . 81
3.12 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4 What Is Science? 149

4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.3 Early Modern Science . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.4 The Goals of Science . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.4.1 Description as the Goal of Science . . . . . . . . . . . . . . . 154
4.4.2 Explanation as the Goal of Science . . . . . . . . . . . . . . 154
4.5 Instrumentalism vs. Realism . . . . . . . . . . . . . . . . . . . . . . 155
4.6 The Scientific Method . . . . . . . . . . . . . . . . . . . . . . . . 156
4.7 Alternatives to The Scientific Method . . . . . . . . . . . . . . . . 157
4.7.1 Falsifiability . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4.7.2 Scientific Revolutions . . . . . . . . . . . . . . . . . . . . . 159
4.8 Is Computer Science a Science? . . . . . . . . . . . . . . . . . . . . 160
4.9 What Kind of Science Might Computer Science Be? . . . . . . . . . 161
5 What Is Engineering? 177

5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
5.3 A History of Engineering . . . . . . . . . . . . . . . . . . . . . . . . 180
5.4 Engineering as Design . . . . . . . . . . . . . . . . . . . . . . . . . 181
5.5 A New Kind of Engineering? . . . . . . . . . . . . . . . . . . . . . . 182
5.6 The Engineering Method . . . . . . . . . . . . . . . . . . . . . . . . 183
CONTENTS 7
6 What Is a Computer?
A Historical Perspective 197
6.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
6.3 Two Histories of Computers . . . . . . . . . . . . . . . . . . . . . . 202
6.4 The Engineering History . . . . . . . . . . . . . . . . . . . . . . . . 202
6.4.1 Ancient Greece . . . . . . . . . . . . . . . . . . . . . . . . . 202
6.4.2 17th Century Adding Machines . . . . . . . . . . . . . . . . 203
6.4.3 Babbage Machines . . . . . . . . . . . . . . . . . . . . . . . 203
6.4.4 Electronic Computers . . . . . . . . . . . . . . . . . . . . . . 204
6.5 The Logical and Mathematical History . . . . . . . . . . . . . . . . . 205
6.6 The Histories Converge . . . . . . . . . . . . . . . . . . . . . . . . . 208
7 What Is an Algorithm? 223

7.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
7.3 What Is Computation? . . . . . . . . . . . . . . . . . . . . . . . . 226
7.3.1 compute . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
7.3.2 reckon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
7.3.3 count, calculate, figure . . . . . . . . . . . . . . . . . . 227
7.3.4 computation . . . . . . . . . . . . . . . . . . . . . . . . . . 227
7.4 What Is Computation? . . . . . . . . . . . . . . . . . . . . . . . . . 227
7.4.1 What Is a Function? . . . . . . . . . . . . . . . . . . . . . . 227
7.4.1.1 Two Meanings . . . . . . . . . . . . . . . . . . . . 227
7.4.1.2 Functions Described Extensionally . . . . . . . . . 228
7.4.1.3 Functions Described as Machines . . . . . . . . . . 230
7.4.1.4 Functions Described Intensionally . . . . . . . . . 231
7.4.1.5 Computable Functions . . . . . . . . . . . . . . . . 232
7.5 Algorithms Made Precise . . . . . . . . . . . . . . . . . . . . . . . . 234
7.6 Four Great Insights of Computer Science . . . . . . . . . . . . . . . 238
7.6.1 Bacons, Leibnizs, Morses, Booles, Ramseys, Turings, and
Shannons Insight . . . . . . . . . . . . . . . . . . . . . . . . 238
7.6.2 Turings Insight . . . . . . . . . . . . . . . . . . . . . . . . . 239
7.6.3 Boehm and Jacopinis Insight . . . . . . . . . . . . . . . . . 240
7.6.4 The (Church-)Turing Thesis . . . . . . . . . . . . . . . . . . 242
7.7 Structured Programming
and Recursive Functions1 . . . . . . . . . . . . . . . . . . . . . . . 244
7.7.1 Structured Programming . . . . . . . . . . . . . . . . . . . . 244
7.7.2 Recursive Functions . . . . . . . . . . . . . . . . . . . . . . 246
7.8 The Halting Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 252
7.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
1 The material in this section is based on lectures by John Case. WHICH IN TURN ARE BASED ON
CLARK & COWELL; SO COMPARE MY PRESENTATION WITH THEIRS!!

8 CONTENTS

8 Turings Analysis of Computation 289

8.1 Required Reading: . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
8.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
8.3 The Entscheidungsproblem . . . . . . . . . . . . . . . . . . . . . . . 291
8.4 Paragraph 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
8.4.1 Sentence 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
8.4.2 Last Sentence . . . . . . . . . . . . . . . . . . . . . . . . . . 293
8.5 Paragraph 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
8.5.1 Sentence 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
8.5.2 Last Sentence . . . . . . . . . . . . . . . . . . . . . . . . . . 294
8.6 Section 1, Paragraph 1 . . . . . . . . . . . . . . . . . . . . . . . . . 295
8.7 Section 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
8.7.1 Section 9, Paragraph 1 . . . . . . . . . . . . . . . . . . . . . 295
8.7.2 Section 9, Subsection I . . . . . . . . . . . . . . . . . . . . . 296
8.7.2.1 Section 9, Subsection I, Paragraph 1 . . . . . . . . 296
8.7.2.1.1 Is it true? . . . . . . . . . . . . . . . . . 297
8.7.2.1.2 What about the paper? . . . . . . . . . . 297
8.7.2.1.3 What about the symbols? . . . . . . . . . 298
8.7.2.1.4 States of mind. . . . . . . . . . . . . . . 300
8.7.2.1.5 Operations. . . . . . . . . . . . . . . . . 301
8.7.2.1.6 Conditions. . . . . . . . . . . . . . . . . 304
8.7.2.1.7 States of mind clarified. . . . . . . . . . . 304
8.7.2.1.8 The Turing machine. . . . . . . . . . . . 305
8.7.2.1.9 Turings thesis. . . . . . . . . . . . . . . 307
8.8 Section 1, Paragraph 2 . . . . . . . . . . . . . . . . . . . . . . . . . 307
8.9 Section 2: Definitions . . . . . . . . . . . . . . . . . . . . . . . . . 312
8.9.1 Automatic Machines . . . . . . . . . . . . . . . . . . . . . 312
8.9.2 Computing Machines . . . . . . . . . . . . . . . . . . . . 313
8.9.3 Circular and Circle-Free Machines . . . . . . . . . . . . . 315
8.9.4 Section 2,
Subsection Computable Sequences and Numbers . . . . . . 317
8.10 Section 3: Examples of Computing Machines . . . . . . . . . . . . 317
8.10.1 Example I . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
8.10.2 Example II . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
8.11 Section 4: Abbreviated Tables . . . . . . . . . . . . . . . . . . . . 330
8.12 Section 5:
Enumeration of Computable Sequences . . . . . . . . . . . . . . . 331
8.13 Section 6: The Universal Computing Machine . . . . . . . . . . . . 336
8.14 The Rest of Turings Paper . . . . . . . . . . . . . . . . . . . . . . . 338
CONTENTS 9
9 What Is a Computer?
A Philosophical Perspective 359
9.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
9.3 John Searle: Anything Is a Computer . . . . . . . . . . . . . . . . . . 366
9.3.1 Computers Are Described in Terms of 0s and 1s . . . . . . . 367
9.3.2 Being a Computer Is a Syntactic Property . . . . . . . . . . . 368
9.3.3 Being a Computer Is Not an Intrinsic Property
of Physical Objects . . . . . . . . . . . . . . . . . . . . . . . 369
9.3.4 We Can Ascribe the Property of Being a Computer
to Any Object . . . . . . . . . . . . . . . . . . . . . . . . . . 373
9.3.5 Everything Is a Computer . . . . . . . . . . . . . . . . . . . 373
9.3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
9.4 Patrick Hayes: Computers as Magic Paper . . . . . . . . . . . . . . . 375
9.5 Gualtiero Piccinini:
Computers as Digital String Manipulators . . . . . . . . . . . . . . . 378
9.5.1 Definition P1 . . . . . . . . . . . . . . . . . . . . . . . . . . 379
9.5.2 Definition P2 . . . . . . . . . . . . . . . . . . . . . . . . . . 380
9.6 The Story So Far . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
9.7 What Else Might Be a Computer? . . . . . . . . . . . . . . . . . . . 381
9.7.1 Is a Brain a Computer? . . . . . . . . . . . . . . . . . . . . . 382
9.7.2 Is the Universe a Computer? . . . . . . . . . . . . . . . . . . 383
9.8 Exercises for the Reader . . . . . . . . . . . . . . . . . . . . . . . . 385
10 What Is a Procedure? 411

10.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
10.3 Preliminary Thought Experiment . . . . . . . . . . . . . . . . . . . . 416
10.4 What Is a Procedure? . . . . . . . . . . . . . . . . . . . . . . . . . . 417
10.5 Challenges to the Church-Turing Computability Thesis . . . . . . . . 417
10.5.1 Carol Cleland: Some Effective Procedures Are Not Turing
Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
10.5.2 Beth Preston: Are Recipes Algorithms? . . . . . . . . . . . . 421
10.5.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
11 What Is Hypercomputation? 449

11.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
11.3 Copelands Theory of Hypercomputation . . . . . . . . . . . . . . . 453
11.4 Trial-and-Error Machines . . . . . . . . . . . . . . . . . . . . . . . . 456
11.4.1 Does Intelligence Require Trial-and-Error Machines? . . . . . 457
11.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
10 CONTENTS

12 Software vs. Hardware 487

12.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
12.3 What Is Software and Its Relation to Hardware? . . . . . . . . . . . . 490
12.3.1 Moors Theory of the Nature of Software . . . . . . . . . . . 490
12.3.2 Subers Theory of the Nature of Software . . . . . . . . . . . 495
12.3.3 Colburns Theory of the Nature of Software . . . . . . . . . . 495
13 Copyright vs. Patent 509

13.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512
13.3 Preliminary Considerations . . . . . . . . . . . . . . . . . . . . . . . 514
13.4 Copyright . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
13.5 Patent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
13.6 Allen Newells Analysis . . . . . . . . . . . . . . . . . . . . . . . . 518
14 What Is Implementation? 537

14.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
14.3 Chalmerss Theory of Implementation . . . . . . . . . . . . . . . . . 539
14.4 Implementation as Semantic Interpretation . . . . . . . . . . . . . . . 542
15 Are Programs Theories? 563

15.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566
15.3 Three Distinctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566
15.3.1 Simulation vs. Emulation . . . . . . . . . . . . . . . . . . . . 566
15.3.2 Simulation vs. the Real Thing . . . . . . . . . . . . . . . . . 567
15.3.3 Theories vs. Models . . . . . . . . . . . . . . . . . . . . . . 567
15.4 Computer Programs as Theories . . . . . . . . . . . . . . . . . . . . 568
15.5 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 570
15.6 Computer Programs Arent Theories . . . . . . . . . . . . . . . . . . 571
CONTENTS 11
16 Can Computer Programs Be Verified? 587

16.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 591
16.3 Program Verification . . . . . . . . . . . . . . . . . . . . . . . . . . 592
16.4 The Fetzer Controversy . . . . . . . . . . . . . . . . . . . . . . . . . 594
16.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 596
17 How Do Programs Relate to the World? 613

17.1 Required Reading: . . . . . . . . . . . . . . . . . . . . . . . . . . . 614
17.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 615
17.3 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 615
17.4 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 616
17.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 619
18 Computer Ethics I: Decisions 629

18.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 631
18.3 Do Computers Make Decisions? . . . . . . . . . . . . . . . . . . . . 631
18.4 Emergency Decisions . . . . . . . . . . . . . . . . . . . . . . . . . . 634
18.5 Moral Decisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 634
19 Philosophy of
Artificial Intelligence 643
19.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 645
19.3 What Is AI? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 645
19.4 The Turing Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
19.4.1 How Computers Can Think . . . . . . . . . . . . . . . . . . 647
19.4.2 The Imitation Game . . . . . . . . . . . . . . . . . . . . . . 647
19.4.3 Thinking vs. Thinking . . . . . . . . . . . . . . . . . . . . 649
19.5 The Chinese Room Argument . . . . . . . . . . . . . . . . . . . . . 651
19.5.1 Two Chinese Room Arguments . . . . . . . . . . . . . . . . 652
19.5.2 The Argument from Biology . . . . . . . . . . . . . . . . . . 652
19.5.2.1 Causal Powers . . . . . . . . . . . . . . . . . . . . 652
19.5.2.2 The Implementation Counterargument . . . . . . . 654
19.5.3 The Argument from Semantics . . . . . . . . . . . . . . . . . 655
19.5.3.1 Which Premise Is at Fault? . . . . . . . . . . . . . 655
19.5.3.2 Semiotics . . . . . . . . . . . . . . . . . . . . . . 657
19.5.3.3 Points of View . . . . . . . . . . . . . . . . . . . . 661
19.5.3.4 A Recursive Theory of Understanding . . . . . . . 664
12 CONTENTS
19.5.4 A Better Way . . . . . . . . . . . . . . . . . . . . . . . . . . 665

19.6 Appendix: Some Definitions of AI . . . . . . . . . . . . . . . . . . . 666
20 Computer Ethics II: AI 677

20.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 679
20.3 Artificial Persons . . . . . . . . . . . . . . . . . . . . . . . . . . . . 679
20.4 Is AI Possible in Principle? . . . . . . . . . . . . . . . . . . . . . . . 679
20.5 What Is the Measure of Success? . . . . . . . . . . . . . . . . . . . . 680
20.6 Rights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682
20.7 Immoral Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 683
20.8 AIs and Morality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 685
20.10Further Sources of Information . . . . . . . . . . . . . . . . . . . . . 687
21 Summary 689
21.2 What Is Philosophy? . . . . . . . . . . . . . . . . . . . . . . . . . . 691
21.3 What Is Computer Science? . . . . . . . . . . . . . . . . . . . . . . . 691
21.4 Is Computer Science a Kind of Science or of Engineering? . . . . . . 692
21.4.1 What Is Science? . . . . . . . . . . . . . . . . . . . . . . . . 692
21.4.2 What Is Engineering? . . . . . . . . . . . . . . . . . . . . . . 692
21.5 What Does Computer Science Study? . . . . . . . . . . . . . . . . . 693
21.5.1 What Is a Computer? Historical Answer . . . . . . . . . . . . 693
21.5.2 What Is an Algorithm? Mathematical Answer . . . . . . . . . 693
21.5.3 What Is a Computer? Philosophical Answer . . . . . . . . . . 693
21.5.4 What Is an Algorithm? Philosophical Answer . . . . . . . . . 694
21.5.4.1 What Is a Procedure? . . . . . . . . . . . . . . . . 694
21.5.4.2 What Is Hypercomputation? . . . . . . . . . . . . . 694
21.5.4.3 What Is a Computer Program? . . . . . . . . . . . 694
21.5.4.3.1 What Is Implementation? . . . . . . . . . 694
21.5.4.3.2 What Is Software? . . . . . . . . . . . . . 695
21.5.4.3.3 Can (Should) Software Be Patented, or
Copyrighted? . . . . . . . . . . . . . . . 695
21.5.4.3.4 Are Programs Scientific Theories? . . . . 695
21.5.4.3.5 What Is the Relation of Programs to the
World? . . . . . . . . . . . . . . . . . . . 695
21.6 Computer Ethics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 696
21.6.1 Are There Decisions Computers Should Never Make? . . . . 696
21.6.2 Should We Build an Artificial Intelligence? . . . . . . . . . . 696
21.6.3 Philosophy of AI . . . . . . . . . . . . . . . . . . . . . . . . 696
21.7 What Is the Philosophy of Computer Science? . . . . . . . . . . . . . 697
21.7.1 The Major Question . . . . . . . . . . . . . . . . . . . . . . 697
21.7.2 9 Theses about Computation . . . . . . . . . . . . . . . . . . 697
CONTENTS 13
21.8 NOTE FOR NEXT DRAFT . . . . . . . . . . . . . . . . . . . . . . 700
22 Syntax and Semantics 701
A Argument Analysis and Evaluation 705

A.1 Recommended Readings: . . . . . . . . . . . . . . . . . . . . . . . . 706
A.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707
A.3 A Question-Answer Game . . . . . . . . . . . . . . . . . . . . . . . 707
A.4 Missing Premises . . . . . . . . . . . . . . . . . . . . . . . . . . . . 709
A.5 When Is an Argument a Good Argument? . . . . . . . . . . . . . . 712
A.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 714
A.7 NOTES FOR NEXT DRAFT . . . . . . . . . . . . . . . . . . . . . . 714
B Position-Paper Assignments 719

B.1 Recommended Readings: . . . . . . . . . . . . . . . . . . . . . . . . 720
B.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 721
B.3 Position Paper #1: What Is Computer Science? . . . . . . . . . . . . 722
B.3.1 Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . 722
B.3.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . 722
B.3.1.2 The Argument . . . . . . . . . . . . . . . . . . . . 722
B.3.2 Suggestions and Guidelines for Peer-Group Editing . . . . . . 725
B.3.3 Grading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 727
B.3.3.1 Philosophy of Grading . . . . . . . . . . . . . . . . 727
B.3.4 Grading Rubric . . . . . . . . . . . . . . . . . . . . . . . . . 727
B.4 Position Paper #2: What Is Computation? . . . . . . . . . . . . . . . 730
B.4.1 Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . 730
B.4.3 Grading Rubric . . . . . . . . . . . . . . . . . . . . . . . . . 735
B.5 Position Paper #3: Is the Brain a Computer? . . . . . . . . . . . . . . 738
B.5.1 Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . 738
B.5.3 Grading Rubric . . . . . . . . . . . . . . . . . . . . . . . . . 740
B.6 Position Paper #4: What Is a Computer Program? . . . . . . . . . . . 743
B.6.1 Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . 743
B.6.1.1 The Argument . . . . . . . . . . . . . . . . . . . . 743
B.6.1.2 Thinksheet . . . . . . . . . . . . . . . . . . . . . . 746
B.6.3 Grading Rubric . . . . . . . . . . . . . . . . . . . . . . . . . 751
B.7 Position Paper #5: Can Computers Think? . . . . . . . . . . . . . . . 754
B.7.1 Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . 754
B.7.3 Grading Rubric . . . . . . . . . . . . . . . . . . . . . . . . . 757
B.8 Optional Position Paper: A Competition . . . . . . . . . . . . . . . . 762
C Term Paper Topics 763

14 CONTENTS
D Final Exam 765

D.1 To the Instructor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 765
D.2 A Final Exam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 768
D.3 NOTES FOR NEXT DRAFT . . . . . . . . . . . . . . . . . . . . . . 770
If you begin with Computer Science, you will end with Philosophy.2
2 Clicking on the first link in the main text of a Wikipedia article, and then repeating the process for
subsequent articles, usually eventually gets you to the Philosophy article. As of May 26, 2011, 94.52% of
all articles in Wikipedia lead eventually to the article Philosophy. Wikipedia:Getting to Philosophy
http://en.wikipedia.org/wiki/Wikipedia:Getting to Philosophy
If you begin with Computer Science, you will end with Philosophy (in 12 links).
15
16 CONTENTS
Preface
Version of 23 November 2015

To readers who have found this document through Brian Leiters Leiter Reports
blog:3
Welcome!
This document is a continually-being-revised draft of a textbook on the philosophy

of computer science. It is based on a course I created for the Department of Computer
Science and Engineering and the Department of Philosophy at the State University of
New York at Buffalo.
The syllabus, readings, assignments, and website for the last version of the course
are online at:
http://www.cse.buffalo.edu/rapaport/584/
The course is described in:

Rapaport, William J. (2005), Philosophy of Computer Science: An
Introductory Course, Teaching Philosophy 28(4): 319341.
A video of my Herbert Simon Keynote Address at NACAP-2006 describing the

course can be downloaded from:
http://www.hass.rpi.edu/streaming/conferences/cap2006/nacp 8 11 2006 9 1010.asx
The current draft of the book is just that: a draft of a work in progress. Early
chapters contain Notes for the Next Draft at the end of each chapter. These represent
material that I have not yet fully incorporated into the body of the text. There are also
one or two unfinished or unwritten chapters.
Please send any comments, suggestions for improvement, etc., to me via email at:
rapaport@buffalo.edu
And thanks for looking! (Changes are listed on next page.)
3 http://leiterreports.typepad.com/blog/2013/10/a-philosophy-of-computer-science-textbook.html
17
18 CONTENTS
Changes since previous version (dated 20 November 2015):
Major changes to: Ch. 3, and Ch. 19,

Minor changes to: Ch. 7, and Bibliography
Acknowledgments
Thanks to Russ Abbott, S.V. Anbazhagan, Peter Denning, Eric Dietrich, William
Duncan, Robin K. Hill, Nurbay Irmak, Patrick McComb, Cristina Murta, Nelson Pole,
Seth David Schoen, Stuart C. Shapiro, Mark Staples, Matti Tedre, and Sen Zhang for
comments on, and suggestions for, this early draft.
To Do
1. Possibly divide into parts, each with chapters, e.g.:

I. What is computer science, including what is science, what is engineering
II. Computers and computing, including history & philosophy of computers,
algorithms & Turing
III. CTCT, including procedures, hypercompn, maybe comprog
IV.
software v hardare, including maybe comprog, cVpat, maybe implementation,
maybe theories
V. maybe implementation, maybe theories, progver, limits
VI. ethics and philai
2. Fix URLs, using BASHurl; see LaTeX Companion
3. Either include somewhere, or recommend to instructor to do, a slow reading of

(at least part of) one paper.
4. clarify which chapter is a discussion of Church-Turing Thesis, and put that in the
title
5. merge chap on progs as theories with chap on relation of progs to real world?
6. add exercises
put exercises at ends of approp chaps as per syllabus?
7. update section on reasoning to include sample valid, invalid args
8. add chap on syntactic semantics
9. write 2nd draft based on files
10. write a new closing chapter/article giving my phics.
11. add an other topics chapter, for things like the philosophy of the web, social
interaction, etc.
12. check Further Readings for soundness & completeness w.r.t. online bibs at
http://www.cse.buffalo.edu/rapaport/584/S10/directory.html
19
20 CONTENTS
13. Include all relevant emails from /web/faculty/rapaport/584/S07/EMAIL and

/web/faculty/rapaport/584/S10/EMAIL as well as anything else relevant from my
course online files.
Chapter 1
What Is Philosophy of
Computer Science?
Version of 14 April 2015
Many people know about modern electronics in general and computers in

particular. they know about code and the compilation process that turns source
code into binary executable code. Computation theory is something different. It is
an area of mathematics that overlaps with philosophy. [PolR, 2009]
21
22 CHAPTER 1. WHAT IS PHILOSOPHY OF COMPUTER SCIENCE?
Figure 1.1: 1984

c by Berkeley Breathed
1.1 Recommended Readings:

1. Strongly Recommended:
Scheutz, Matthias (2003), Computation, Philosophical Issues about,
Encyclopedia of Cognitive Science (London: Macmillan): 604610.
http://onlinelibrary.wiley.com/doi/10.1002/0470018860.s00209/abstract
(accessed 2 November 2012; requires login)
May also be accessible online at:
http://staffweb.worc.ac.uk/DrC/Courses%202006-7/Comp%204070/
Reading%20Materials/scheutz%20Comp%20Philo%20Issues.pdf
(accessed 2 November 2012)
http://home.dei.polimi.it/schiaffo/TFI/scheutz 2002.pdf
2. Recommended:
Turner, Raymond; & Eden, Amnon H. (2008), The Philosophy of Computer
Science, in Edward N. Zalta (ed.), Stanford Encyclopedia of Computer Science
http://plato.stanford.edu/entries/computer-science/
accessed 2 November 2012
1.2. WHAT THIS BOOK IS ABOUT 23
1.2 What This Book Is About

In this book, we will look at some of the central issues in the philosophy of computer
science. The book is not designed to answer all (or even any) of the philosophical
questions that can be raised about the nature of computers and computer science.
Rather, it is designed to bring you up to speed on a conversation about these issues
to give you some of the background knowledge about themso that you can read the
literature for yourself and perhaps become part of the conversation by contributing your
own views.
This book is intended for readers who might know some philosophy but no
computer science, as well as for readers who might know some computer science but
no philosophy (and there might even be some readers who know little or nothing about
both!). So, we will begin by asking what philosophy is. (Most of the rest of the book
will be concerned with what computer science is.) Thus, our first question will be:
1. What is philosophy?
(a) And, in particular: What is the philosophy of X? (where X = things like:

science, psychology, history, etc., and, of course, computer science).
Then we will ask:

2. What is computer science?
To answer this, we will need to consider a series of questions, each of which

leads to another:
(a) Is computer science a science? Or is it a branch of engineering?
i. What is science?
ii. What is engineering?
(b) If computer science is a science, what is it a science of ?
i. Is it a science of computers?
A. In that case, what is a computer?
ii. Or is computer science a science of computation?
A. In that case, what is computation?
Computations are said to be algorithms (very roughly,
procedures for doing something), so what is an algorithm?
Algorithms are said to be procedures, or recipes, so what is a
procedure? What is a recipe?
What are Churchs and Turings theses about computability?
(Briefly, these theses are claims that algorithms can be
understood as certain mathematical theories.)
What is hypercomputation?
(Hypercomputation, roughly, is the study of kinds of comp-
utation that cannot be understood by those mathematical
theories that model algorithms.)
Algorithms are said to be implemented in computer programs,

so what is a computer program?
And what is an implementation?
What is the relation of a program to that which it models or
simulates?
What is simulation?
Are programs (scientific) theories?
What is software?
Can computer programs be copyrighted, or patented?
Can computer programs be verified?
3. Next, we will turn to some of the issues in the philosophy of artificial intelligence
(AI):
(a) What is AI?

(b) What is the relation of computation to cognition?
(c) Can computers think?
(d) What are the Turing Test and the Chinese Room Argument?
(Very briefly: The Turing Test is a test proposed by one of the creators of
the field of computation to determine whether a computer can think. The
Chinese Room Argument is a thought experiment devised by a philosopher,
which is designed to show that the Turing Test wont work.)
4. Finally, we will consider two questions that fall under the heading of computer
ethics:
(a) Should we trust decisions made by computers?

(b) Should we build intelligent computers?
Along the way, we will look at how philosophers reason and evaluate logical
arguments, and there will be some suggested writing assignments designed to help
focus your thinking about these issues.
1.3 What This Book Is Not About

Have I left anything out? Most certainly! I do not wish to claim that the questions
raised above and discussed in this book exhaust the philosophy of computer science.
Rather, they are a series of questions that arise naturally from our first question: What
is computer science?
But there are many other issues. Some are included in a topic sometimes called
philosophy of computing, which might include investigations into the ubiquity of
computing (your smartphone is a computer, your car has a computer in it; perhaps
someday your refrigerator or toaster or bedroom wall will contain (or even be) a
computer): How will our notion of computing change because of this ubiquity? Will
1.3. WHAT THIS BOOK IS NOT ABOUT 25
this be a good or a bad thing?). Other topics include the role of the Internet and
social media (for instance, Tim Berners-Lee, who created the World Wide Web, has
argued that Web science should be its own discipline [Berners-Lee et al., 2006],
[Lohr, 2006]). And so on.
Other issues more properly fall under the heading of the philosophy of artificial
intelligence (AI). Confusingly, they are often called philosophy of computer
science. As noted, we will look at some of these in this book, but there are many
others that we wont cover, even though the philosophy of AI is a proper subset of the
philosophy of computer science.
And there are several books and essays on the philosophy of information. As well
see, computer science is sometimes defined as the study of how to process information,
so the philosophy of information is clearly a close cousin of the philosophy of computer
science. But I dont think that either is included in the other; they merely have a non-
empty intersection. The Further Sources of Information section below includes some
of these books and essays if this is a topic you wish to explore.
Finally, there are a number of philosophers and computer scientists who have
discussed topics related to what I am calling the philosophy of computer science whom
we will not deal with at all (such as the philosopher Martin Heidegger and the computer
scientist Terry Winograd).1 An Internet search (for example: Heidegger "computer
science") will help you track down information on these thinkers and others not
mentioned in this book.
But I think that our questions above will keep us busy for a while, as well as prepare
you for examining some of these other issues. Think of this book as an extended
infomercial to bring you up to speed on the computer-sciencerelated aspects of a
philosophical conversation that has been going on for over 2500 years, to enable you
to join in the conversation.2
So, lets begin . . .
. . . but, before we do, on the next page you will find some . . .
1 One philosopher of computer science calls them the Dark Side philosophers, because they tend not to
be sympathetic to computational views of the world.
2 In 2006, responding to a talk that I gave on the philosophy of computer science, Selmer Bringsjord (a
philosopher and cognitive scientist who has written extensively on the philosophy of computer science) said,
Philosophy of Computer Science. . . is in its infancy [Bringsjord, 2006]. This may be true as a discipline
so called, but there have been philosophical investigations of computer science and computing since at least
[Turing, 1936], and James H. Moors work goes back to the 1970s (e.g., [Moor, 1978]).
1.4 . . . Further Sources of Information

Each chapter will end with a list of essays, books, or websites that discuss some or all of the
topics coveredbut not explicitly citedin the chapter.3 They will also include items that
extend the discussion to other topics that were not covered.
1. Books and Essays:
Allo, Patrick (2010), Putting Information First: Luciano Floridi and the
Philosophy of Information, Metaphilosophy 41(3) (April): 248254.
Burkholder, Leslie (ed.) (1992), Philosophy and the Computer (Boulder, CO:
Westview Press).
Bynum, Terrell Ward (2010), Philosophy in the Information Age,
Metaphilosophy 41(3) (april): 420442.
Bynum, Terrell Ward; & Moor, James H. (eds.) (2000), The Digital Phoenix: How
Computers Are Changing Philosophy, Revised Edition (Oxford: Blackwell).
Colburn, Timothy (2000), Philosophy and Computer Science (Armonk, NY:
M.E. Sharpe).
Colburn, Timothy (2006), What Is Philosophy of Computer Science?, extended
abstract of paper presented at the European Conference on Computing and
Philosophy (ECAP06, 2224 June, Trondheim, Norway)
http://tinyurl.com/Colburn06 (accessed 2 November 2012)
Denning, Peter J.; & Bell, Tim (2012), The Information Paradox, American
Scientist 100 (November-December): 470477.
Dodig-Crnkovic, Gordana (2006), What Is Philosophy of Computer Science?
Experience from the Swedish National Course, paper presented at the European
Conference on Computing and Philosophy (ECAP06, 2224 June, Trondheim,
Norway)
http://www.cse.buffalo.edu/rapaport/584/Dodig-Crnkovic.pdf
Dunn, J. Michael (2008), Information in Computer Science, in Pieter Adriaans
& Johan van Benthem (eds.), Philosophy of Information, Vol. 8 of Dov M. Gabbay,
Paul Thagard, & John Woods (general eds.), Handbook of the Philosophy of
Science: 587614, short pre-print at
http://www.illc.uva.nl/HPI/Draft Information in Computer Science.pdf
(accessed 31 March 2015).
This is a very readable survey of the nature of information and its role in
computer science, covering many of the same topics and issues as this textbook
does.
Eden, Amnon H.; & Turner, Raymond (eds.), Special Issue on the Philosophy of
Computer Science, Minds and Machines 21(2) (Summer).
Table of contents accessed 23 July 2014 from:
http://link.springer.com/journal/11023/21/2/page/1
Floridi, Luciano (1999), Philosophy and Computing: An Introduction (London:
Routledge).
3 Though some may be cited in other chapters.
1.4. . . . FURTHER SOURCES OF INFORMATION 27
Floridi, Luciano (2002), What Is the Philosophy of Information?,

Metaphilosophy 33(12) (January): 123145.
Floridi, Luciano (ed.) (2003), The Philosophy of Information, two special issues of
Minds and Machines 13(4) & 14(1).
Floridi, Luciano (2004), Open Problems in the Philosophy of Information,
Metaphilosophy 35(4) (July): 554582.
Floridi, Luciano (ed.) (2004), The Blackwell Guide to the Philosophy of Computing
and Information (Malden, MA: Blackwell).
Floridi, Luciano (2011), The Philosophy of Information (Oxford: Oxford
University Press).
Essay Review:
Dunn, J. Michael (2013), A Guide to the Floridi Keys, Metascience 22: 93
98.
Hilbert, Martin; & Lopez, Priscila (2011), The Worlds Technological Capacity to
Store, Communicate, and Compute Information, Science 332 (1 April): 6065.
Longo, Giuseppe (ed.) (1999), Philosophy of Computer Science, special issue of
The Monist 82(1).
Machlup, Fritz; & Mansfield, Una (eds.) (1983), The Study of Information:
Interdisciplinary Messages (New York: John Wiley & Sons).
Magnani, Lorenzo (ed.) (2006), Computing and Philosophy: Proceedings of
the Second European Conference, Computing and Philosophy (ECAP2004-Italy)
(Pavia, Italy: Associated International Academic Publishers).
Moor, James H.; & Bynum, Terrell Ward (eds.) (2002), Cyberphilosophy: The
Intersection of Computing and Philosophy (Malden, MA: Blackwell).
Pylyshyn, Zenon W. (1992), Computers and the Symbolization of Knowledge,
in Ralph Morelli, W. Miller Brown, Dina Anselmi, Karl Haberlandt, & Dan
Lloyd (eds.), Minds, Brains & Computers: Perspectives in Cognitive Science and
Artificial Intelligence (Norwood, NJ: Ablex): 8294.
Preprint (1987) accessed 9 April 2014 from:
http://ruccs.rutgers.edu/ftp/pub/papers/suffolk.pdf
Rapaport, William J. (2005), Philosophy of Computer Science: An Introductory
Course, Teaching Philosophy 28(4): 319341.
Scarantino, Andrea; & Picciini, Gualtiero (2010), Information without Truth,
Metaphilosophy 41(3) (April): 313330.
Scheutz, Matthias (2003), Computation, Philosophical Issues about,
Encyclopedia of Cognitive Science (London: Macmillan): 604610.
Smith, Brian Cantwell (1996), On the Origin of Objects (Cambridge, MA: MIT
Press).
Koepsell, David R. (1998), Review of Smith 1996, Philosophical Psychology
11(3): 389391.
Loui, R.P. (1998), Review of Smith 1996, Artificial Intelligence 106: 353358.
Smith, Brian Cantwell (2002), The Foundations of Computing, in Scheutz,
Matthias (ed.), Computationalism: New Directions (Cambridge, MA: MIT Press):
2358.
DeJohn, Jerry; & Dietrich, Eric (2003), Subvert the Dominant Paradigm! A
review of Computationalism: New Directions, edited by Matthias Scheutz,
Journal of Experimental & Theoretical Artificial Intelligence 15(4) (October-
December): 375382, esp. 3.1, pp. 378379.
Tedre, Matti (2007), Know Your Discipline: Teaching the Philosophy of Computer
Science, Journal of Information Technology Education 6: 105122.
Turner, Raymond; & Eden, Amnon H. (eds.) (2007), Philosophy of Computer
Science special issue of Minds and Machines 17(2) (Summer): 129247.
Turner, Raymond; & Eden, Amnon H. (2008), The Philosophy of Computer
Science, special issue of Journal of Applied Logic 6(4) (December): 459552.
2. Websites:4
Aaronson, Scott (2011), Philosophy and Theoretical Computer Science,
http://stellar.mit.edu/S/course/6/fa11/6.893/
(accessed 23 December 2013).
Brey, Philip; & Sraker, Johnny Hartz (2008), Philosophy of Computing and
Information Technology,
http://www.idt.mdh.se/kurser/comphil/2011/PHILO-INFORM-TECHNO-20081023.pdf
(accessed 4 January 2013)
Brown, Curtis (2000, Fall), Seminar: Philosophy and Computers
http://www.trinity.edu/cbrown/philandcomp/ (accessed 2 November 2012)
Commission for the History and Philosophy of Computing, http://www.hapoc.org/
Dai, Yi; Lamberov, Lev; & Ostermann, Klaus (organizers) (2014), Discussion
Seminar: Philosophy of Programming Languages, Department of Mathematics
and Computer Science, University of Marburg (Germany)
https://github.com/plmday/DS-PhPL
(accessed 13 June 2014)
Hill, Robin K. (2014), Teaching the Philosophy of Computer Science (blog),
http://teachingphilofcs.blogspot.com/
Philosophy of Computer Science (ECAP06)
http://pcs.essex.ac.uk/ecap06/cfp.html (accessed 5 February 2013)
Philosophy of Computer Science: Online and Offline Resources
http://pcs.essex.ac.uk/ (accessed 2 November 2012)
Philosophy of Computer Science Course (CD5650, 2004)
http://www.idt.mdh.se/gdc/PI 04/PI-network-course.htm
Philosophy of Computing and Informatics Network
http://www.idt.mdh.se/gdc/pi-network.htm (accessed 2 November 2012)
Price, Colin B. (2007), Concepts and Philosophy of Computing, University of
Worcester (UK)
Accessed 21 March 2014 from:
http://staffweb.worc.ac.uk/DrC/Courses%202006-7/Comp
4 If any of these websites should disappear between the time of my writing and the time of your reading,
you may be able to find them either via a search engine such as Google or via the Wayback Machine Internet
archive at: http://archive.org/web/web.php
1.5. NOTES FOR NEXT DRAFT 29
Tedre, Matti (2007, Winter-Spring), The Philosophy of Computer Science

(175616), University of Joensuu, Finland
http://cs.joensuu.fi/mmeri/teaching/2006/philcs/ (accessed 2 November 2012)
Contains links to detailed lecture notes.
See esp.: Introduction to the Course
http://cs.joensuu.fi/mmeri/teaching/2006/philcs/files/lecture notes1.pdf
Now lets really begin . . .
1.5 NOTES FOR NEXT DRAFT

Include refs online at turing:01-WHAT IS PHILofCS
Chapter 2
What Is Philosophy?
Philosophy is the microscope of thought.

(Victor Hugo, Les Miserables [Hugo, 1862, Vol. 5, Book Two, Ch. II,
p. 1262]).
Philosophy is the peculiarly stubborn attempt to think clearly.

(attributed to William James, 18421910).
To the person with the right turn of mind,. . . all thought becomes
philosophy [Schwitzgebel, 2012].
Philosophy can be any damn thing you want!

(John Kearns, personal communication, 7 November 2013.)
Figure 2.1: 2012,

c ??
31
32 CHAPTER 2. WHAT IS PHILOSOPHY?

1. Very Strongly Recommended:
Audi, Robert (1981), Philosophy: A Brief Guide for Undergraduates (Newark,
DE: American Philosophical Assocation)
http://tinyurl.com/Audi81 (accessed 2 November 2012)
Plato, The Apology
Any edition; various versions are online: search for Plato Apology.
This is Platos explanation of what Socrates thought that philosophy was
all about; it is a good introduction to the skeptical, questioning nature of
philosophy.
3. Recommended:
Colburn, Timothy R. (2000), Philosophy and Computer Science (Armonk, NY:
M.E. Sharpe):
(a) Ch. 3: AI and the History of Philosophy (pp. 1940)
(b) Ch. 4: AI and the Rise of Contermporary Science and Philosophy (pp. 41
50)
Some of the material may be online at the Google Books
(http://books.google.com/) website for this book:
http://tinyurl.com/Colburn00
(accessed 5 February 2013)
2.2. INTRODUCTION 33
2.2 Introduction
What is philosophy? is a question that is not a proper part of the philosophy of
computer science. But, because many readers may not be familiar with philosophy, I
want to begin our exploration with a brief introduction to how I think of philosophy and
how I would like non-philosophical readers who are primarily interested in computer
science to think of it.
In this chapter, I will give you my definition of philosophy. We will also briefly
look at some of the history of philosophy that is relevant to the philosophy of computer
science, and we will examine the principal methodology of philosophy: the evaluation
of logical arguments. DO I DO THAT HERE, OR IN THE APPENDIX?
2.3 A Definition of Philosophy

The word philosophy1 has a few different meanings. When it is used informally, in
everyday conversation, it can mean an outlook, as when someone asks you what your
philosophy of life is. It can also mean something like calm, as when we say that
someone takes bad news very philosophically (that is, very calmly).
But, in this chapter, I want to explicate the technical sense of modern, analytic,
Western philosophythat is, the kind of philosophy that has been done since at least
Socratess time (though modern philosophy is itself a technical term that usually
refers to the kind of philosophy that has been done since Descartes). It is analytic
in the sense that it is primarily concerned with the logical analysis of concepts (rather
than literary, poetic, or speculative approaches), and it is Western in the sense that
it has been done by philosophers working primarily in Europe (especially in Great
Britain) and North Americathough, of course, there are very many philosophers who
do analytic philosophy in other countries.
Western philosophy began in ancient Greece. Socrates (470399 B.C.E.,2 that is,
around 2500 years ago) was opposed to the Sophists, a group of teachers who can be
caricaturized as an ancient Greek version of ambulance-chasing lawyers: purveyors
of rhetorical tricks [McGinn, 2012b]. The Sophists were willing to teach anything
(whether it was true or not) to anyone, or to argue anyones cause (whether their cause
was just or not), for a fee. Like the Sophists, Socrates also wanted to teach and argue,
1 Many philosophers have adopted a convention that single quotes are used to form the name of a word or
expression. So, when I write this:

philosophy
I am not talking about philosophy(!); rather, I am talking about the 10-letter word spelled p-h-i-l-o-s-o-p-
h-y. This use of single quotes enables us to distinguish between a thing that we are talking about and the
name or description that we use to talk about it. This is like the difference between a number (a thing
that mathematicians talk about) and a numeral (a word or symbol that we use to talk about numbers).
This is called the use-mention distinction; see the Wikipedia article at http://en.wikipedia.org/wiki/Use-
mention distinction (accessed 5 February 2013).
I will use double quotes when I am directly quoting someone. I will also sometimes use double quotes as
scare quotes, to indicate that I am using an expression in a special or perhaps unusual way (as I just did).
And I will use double quotes to indicate the meaning of a word or other expression.
2 B.C.E. is the abbreviation for before the common era; that is, B.C.E. years are the negative years
before the year 1, which is known as the year 1 C.E. (for common era).
but only to seek wisdom: truth in any field. In fact, the word philosophy comes from
Greek roots meaning love of [philo] wisdom [sophia]. The reason that Socrates only
sought wisdom rather than claiming that he had it (like the Sophists did) was that he
believed that he didnt have it: He claimed that he knew that he didnt know anything
(and that, therefore, he was actually wiser than those who claimed that they did know
things but who really didnt). As Plato (430347 B.C.E., Socratess student) put it in his
dialogue Apology, Socrates played the role of a gadfly, constantly questioning (and
annoying!) people about the justifications for, and consistency among, their beliefs, in
an effort find out the truth for himself from those who considered themselves to be wise
(but who really werent).
Plato defined philosopher (and, by extension, philosophy) in Book V of his
Republic (line 475c):
The one who feels no distaste in sampling every study, and who attacks
the task of learning gladly and cannot get enough of it, we shall justly
pronounce the lover of wisdom, the philosopher. [Plato, 1961b, p. 714,
my emphasis].
Adapting this, I define philosophy as:
the personal search for truth, in any field, by rational means.
This raises several questions:
1. Why only personal? (Why not universal?)
2. What is truth?
3. Why is philosophy only the search for truth? Cant we succeed in our search?
4. What does any field mean? Is philosophy really the study of anything and
everything?
5. What counts as being rational?
Lets look at each of these, beginning with the second.
2.4 What Is Truth?

The study of the nature of truth is one of the Big Questions of philosophy (along with
things like: What is the meaning of life? What is good? and so on).3 I cannot hope to
do justice to it here, but there are two theories of truth that will prove useful to keep in
mind on our journey through the philosophy of computer science: the correspondence
theory of truth and the coherence theory of truth.
3 See Gabriel Segals response to the question, What is it that is unique to philosophy that distinguishes
it from other disciplines?, online at:

http://www.askphilosophers.org/question/5017 (accessed 13 February 2013).
2.4. WHAT IS TRUTH? 35
2.4.1 The Correspondence Theory of Truth

According to the correspondence theory (cf. [David, 2009]), truth is a relation between
two things: On the one hand, there are beliefs (or sentences, or propositions); on
the other hand, there is reality: A belief (or a sentence, or a proposition) is true
if and only if (iff) it corresponds to reality, that is, iff it matches or accurately
characterizes or describes reality.
To take a classic example, the three-word English sentence Snow is white. is true
iff the stuff in the real world that precipitates in certain winter weather (that is, snow)
has the same color as milk (that is, iff it is white). Put somewhat paradoxically (but
correctlysee the first footnote in 2.3) Snow is white. is true iff snow is white.
How do we determine whether a sentence (or belief, or proposition) is true? In
principle, we would have to compare the parts of the sentence (its words plus its
grammatical structure, and maybe even the context in which it is thought, uttered, or
written) with parts of reality, to see if they correspond. But how do we access reality?
How can we do the pattern matching between our beliefs and reality?
One answer is by sense perception (perhaps together with our beliefs about what we
perceive). But sense perception is notoriously unreliable (think about optical illusions,
for instance). And one of the issues in deciding whether our beliefs are true is deciding
whether our perceptions are accurate (that is, whether they match reality).
So we seem to be back to square one, which gives rise to the coherence theory.
2.4.2 The Coherence Theory of Truth

According to the coherence theory of truth (cf. [Young, 2008]), a set of propositions
(or beliefs, or sentences) is true iff:
1. they are mutually consistent, and
2. they are supported by, or consistent with, all available evidence;
that is, they cohere with each other and with all evidence.
Note that observation statements (that is, descriptions of what we observe in the
world around us) are among the claims that must be mutually consistent, so this is
not (necessarily) a pie-in-the-sky theory that doesnt have to relate to the way things
really are. It just says that we dont have to have independent access to reality in
order to determine truth.
Which theory is correct? Well, for one thing, there are more than two theories:
There are several versions of each kind of theory, and there are other theories of truth
that dont fall under either category. Fortunately, the answer to which kind of theory
is correct (that is, which kind of theory is, if you will excuse the expression, true!)
is beyond our present scope! But note that the propositions that a correspondence
theory says are true must be mutually consistent (if reality is consistent!) and they
must be supported by all available evidence; that is, a correspondence theory must
cohere. (Moreover, if you include both propositions and reality in one large, highly
interconnect network, that network must also cohere.)
Lets return to the question raised above: How can we decide whether a statement
is true? One way that we can determine its truth is syntactically (that is, in terms of its
grammatical structure only, not in terms of what it means): by trying to prove it from
axioms via rules of inference. (Ill say more about what axioms and rules of inference
are in Chs. 14 and 22. For now, just think of proving theorems in geometry or logic.)
It is important to keep in mind that, when you prove a statement this way, you are not
proving that it is true! You are simply proving that it follows logically from certain
other statements, that is, that it coheres in a certain way with those statements. Now,
if the starting statementsthe axiomsare true (note that I said if they are true; I
havent told you how to determine their truth value yet), and if the rules of inference
preserve truth, then the statement that you prove by means of themthe theorem
will also be true. (Briefly, rules of inferencewhich tell you how to infer a statement
from other statementsare truth-preserving if the inferred statement cannot be false as
long as the statements from which it is inferred are true.)
Another way we can determine whether a statement is true is semantically (that is,
in terms of what it means). This, by the way, is the only way to determine whether
an axiom is true, since, by definition, an axiom cannot be inferred from any other
statements. (If it could be so inferred, then it would be those other statements that
would be the real axioms.)
But to determine the truth of a statement semantically is also to use syntax:
We semantically determine the truth value of a complex proposition by syntactic
manipulation (truth tables) of its atomic constituents. (We can use truth tables to
determine that axioms are true.) NEED TO EXPLAIN WHAT SYN AND SEM ARE,
AND HOW THEY ARE RELATED. How do we determine the truth value of an atomic
proposition? By seeing if it corresponds to reality. But how do we do that? By
comparing the proposition with reality, that is, by seeing if the proposition coheres
with reality.
2.5 On Searching for the Truth vs. Finding It?

Thinking is, or ought to be, a coolness and a calmness. . . . [Melville, 2002,
Ch. 135, p. 419]
How does one go about searching for the truth, for answering questions? As well see
below, there are basically two, complementary methods: thinking hard and empirical
investigation. I have read in many places that philosophy is just thinking really hard
about things.4 Such hard thinking requires rethinking explicitly what we already
believe implicitly [Baars, 1997, 187]. In other words, its more than just expressing
ones opinion unthinkingly.
Can we find the truth? Not necessarily; that is, we may not be able to find it.
But I also believe that finding it is not necessary; that is, we may not have to find it:
4 Several people have offered this characterization of philosophy. See the quotes at [Popova, 2012]
from Marilyn Adams (originally in [Edmonds and Warburton, 2010, xiii]) and, especially, David Papineau:
Philosophy is thinking hard about the most difficult problems that there are. And you might think scientists
do that too, but theres a certain kind of question whose difficulty cant be resolved by getting more empirical
evidence. It requires an untangling of presuppositions: figuring out that our thinking is being driven by ideas
we didnt even realize that we had. And thats what philosophy is [Edmonds and Warburton, 2010, xx].
2.5. ON SEARCHING FOR THE TRUTH VS. FINDING IT? 37
Philosophy is the search for truth.5 Albert Einstein (quoting G.E. Lessing)6 said that
the search for truth is more precious than its possession [Einstein, 1940, 492]. In a
similar vein, the mathematician Carl Friedrich Gauss said, It is not knowledge, but
the act of learning, not possession but the act of getting there, which grants the greatest
enjoyment.7
One reason that this search will never end (which is different from saying that it
will not succeed) is that you can always ask Why?; that is, you can always continue
inquiring. This is
the way philosophyand philosophersare[:] Questions beget questions,
and those questions beget another whole generation of questions. Its
questions all the way down. [Cathcart and Klein, 2007, 4]
(You can even ask why Why? is the most important question [Everett, 2012, 38]!) In
fact, as the physicist John Wheeler has pointed out, the more questions you answer, the
more questions you can ask: We live on an island of knowledge surrounded by a sea
of ignorance. As our island of knowledge grows, so does the shore of our ignorance.8
And the US economist and social philosopher Thorstein Veblen said, The outcome
of any serious research can only be to make two questions grow where only one grew
before[Veblen, 1908, 396].
If the philosophical search for truth is a never-ending process, can we ever make
any progress in philosophy? Mathematics and science, for example, are disciplines that
not only search for the truth, but seem to find it; they seem to make progress in the sense
that we know more mathematics and more science now than we did in the past. We
have well-confirmed scientific theories, and we have well-established mathematical
proofs of theorems. (The extent to which this may or may not be exactly the right
way to look at things will be considered in Ch. 4.) But philosophy doesnt seem to
be able to empirically confirm its theories or prove any theorems. So, is there any
sense of progress in philosophy? Or are the problems that philosophers investigate
unsolvable?
I think there can be, and is, progress. Solutions to problems are never as neat as
they seem to be in mathematics; in fact, theyre not even that neat in mathematics!
Solutions to problems are always conditional; they are based on certain assumptions.
In mathematics, those assumptions include axioms, but axioms can be challenged and
modified (consider the history of non-Euclidean geometry, which began by challenging
and modifying the Euclidean axiom known as the parallel postulate).9 So, solutions
5 For more on the importance of search over success, see my website on William Perrys theory of
intellectual development [http://www.cse.buffalo.edu/rapaport/perry-positions.html] and Rapaport 1982.

6 The true value of a man [sic] is not determined by his possession, supposed or real, of Truth, but rather
by his sincere exertion to get to the Truth. It is not possession of the Truth, but rather the pursuit of Truth by
which he extends his powers. . . [Lessing, 1778].
7 From Letter to Bolyai, 1808; see
http://blog.gaiam.com/quotes/authors/karl-friedrich-gauss/21863, accessed 9 September 2011.

8 Scientific American 267 (1992); see
http://www.todayinsci.com/W/Wheeler John/WheelerJohn-Quotations.htm, accessed 9 September 2011.

9 One version of the Parallel Postulate is this: For any line L, and for any point P not on L, there
is only one line L0 that P is on and that is parallel to L. For some of the history of non-Euclidean
geometries, see http://mathworld.wolfram.com/ParallelPostulate.html (accessed 13 February 2013) or
http://en.wikipedia.org/wiki/Parallel postulate (accessed 13 February 2013).
are really parts of larger theories, which include the assumptions that the solution
depends on, as well as other principles that follow from the solution. Progress can be
made in philosophy (as in other disciplines), not only by following out the implications
of your beliefs (a kind of forward-looking progress), but also by becoming aware of
the assumptions that underlie your beliefs (a kind of backward-looking progress).
Asking Why? is partperhaps the principal partof philosophys general
role of critically evaluating beliefs [Colburn, 2000, 6] and refusing to accept
any platitudes or accepted wisdom without examining it (Donna Dickenson, in
[Popova, 2012]. Critical thinking in general, and philosophy in particular, look. . .
for crack[s] in the wall of doctrinaire [beliefs]some area of surprise, uncertainty, that
might then lead to thought [Acocella, 2009, 71]. Whenever you have a question, either
because you do not understand something or because you are surprised by it or unsure
of it, you should begin to think carefully about it. And one of the best ways to do this
is to ask Why?: Why did the author say that? Why does the author believe it? Why
should I believe it? And a related question is this: What are the implications of that?
(What else must be true if that were true? And should I believe those implications?)
Philosophy is a watchdog [Colburn, 2000, 6]. This zoological metaphor is related
to Socratess view of the philosopher as gadfly, investigating the foundations or reasons
for beliefs and for the way things are, always asking What is X? and Why?.
Of course, this got him in trouble: His claims to be ignorant were thought (probably
correctly) to be somewhat disingenuous. As a result, he was tried, condemned to death,
and executed. (For the details, read Platos Apology.)
One moral is that philosophy can be dangerous. As Eric Dietrich puts it:
Thinking about the Big Questions is serious, difficult business. I tell my
philosophy students: If you like sweets and easy living and fun times and
happiness, drop this course now. Philosophers are the hazmat handlers
of the intellectual world. It is we who stare into the abyss, frequently
going down into it to great depths. This isnt a job for people who scare
easily or even have a tendency to get nervous. (Personal communication,
5 October 2006.)
But there are advantages to always asking questions and being skeptical: A
skeptical approach to life leads to advances in all areas of the human condition; while
a willingness to accept that which does not fit into the laws of our world represents a
departure from the search for knowledge [Dunning, 2007]. Being skeptical doesnt
necessarily mean refraining from having any opinions or beliefs. But it does mean
being willing to question anything and everything that you read or hear (or think!).
(Another way f putting this is this: In philosophy, the jury is always out!cf.
[Polger, 2011, 11]. But, as we saw above, this does not mean that there can be no
progress in philosophy.) Why would you want to question anything and everything?
So that you can find reasons for (or against) believing what you read or hear (or think).
And why is it important to have these reasons? For one thing, it can make you feel
more confident about your beliefs and the beliefs of others. For another, it can help you
try to convince others about your beliefsnot necessarily to convince them that they
should believe what you believe, but to help them understand why you believe what
you do.
2.6. WHAT IS RATIONAL? 39
2.6 What Is Rational?

Philosophy: the ungainly attempt to tackle questions that come naturally
to children, using methods that come naturally to lawyers [Hills, 2007].
The main concern of philosophy is to question and understand very common ideas
that all of us use every day without thinking about them [Nagel, 1987, 5]. This is
why, perhaps, the questions that children often ask (especially, Why?) are often
deeply philosophical questions.
But mere statements (that is, opinions) by themselves are not rational. Rather,
argumentsreasoned or supported statementsare capable of being rational. That
is, being rational requires logic. But there are lots of different (kinds of) logics, so
there are lots of different kinds of rationality. And there is another kind of rationality,
which depends on logics of various kinds, but goes beyond them in at least one way:
empirical, or scientific, rationality.
2.6.1 Logical Rationality

There are (at least) two basic kinds of logical rationality: deductive-logical (or
absolutely certain) rationality and scientific (or probabilistic) rationality. There is also,
I think, a third kind, which Ill call psychological or maybe economic, and which
is at the heart of knowledge representation and reasoning in AI.
2.6.1.1 Deductive Logic
Deductive logic is the main kind of logical rationality. Premises P1 , . . . , Pn

deductively support (or yield, or entail, or imply) a conclusion C iff C must
be true if all of the Pi are true; that is, C is true iff the Pi are truth preserving.
I will use the symbol `D to represent this truth-preserving relation between
premises and a conclusion that is deductively supported by them. For example,
P, P C `D C
means that the two premises P and P C (read this as: if P, then C) deductively
support (that is, they deductively imply) C.
For example, let P = Today is Wednesday. and let C = We are studying
philosophy. So the inference becomes: Today is Wednesday. If today is
Wednesday, then we are studying philosophy. Therefore (deductively), we are studying
philosophy.
Note that C can be false! It only has to be true relative to the premises (that is, true
relative to its context).
Also, any or all of the Pi can be false! A deductive argument is said to be valid iff
it is impossible for all of the premises to be true but the conclusion false. This can be
said in a slightly different way: A deductive argument is valid iff, whenever all of its
premises are true, its conclusion cannot be false (or: its conclusion must also be true).
A deductive argument is said to be sound iff it is valid and all of its premises are,
in fact, true. So, any or all of the premises of a deductively valid argument can be false,
as long asif they were true, then the conclusion would also have to be true.)
Also, the Pi can be irrelevant to C! But thats not a good idea, because it wouldnt be
a convincing argument. (Relevance logics are one way of dealing with this problem;
[Anderson and Belnap, 1975, Anderson et al., 1975].)
2.6.1.2 Inductive Logic
Inductive logic is one of the three main kinds of scientific rationality: The first is
deductive (as we just saw): Logical rationality is certainly part of science. The third is
abductive (discussed in the next section).
In inductive logic, P1 , . . . , Pn `I C iff C is probably true if all of the Pi are true. For
example, suppose that you have an urn containing over a million ping-pong balls, and
suppose that you take out one of them at random and observe that it is red. What do
you think the chances are that the next ball will also be red? They are probably not
very high. But suppose that the second ball that you examine is also red. And the third.
. . . And the 999,999th. Now how likely do you think it is that the next ball will also be
red? The chances are probably very high, so:
Red(ball1 ), . . . , Red(ball999,999 ) `I Red(ball1,000,000 ).
Unlike deductive inferences, however, inductive ones do not guarantee the truth
of their conclusion. Although it is not likely, it is quite possible that the millionth
ping-pong ball will be, say, the only blue one in the urn.
2.6.1.3 Abductive Logic
Abductive logic, sometimes also known as inference to the best explanation

[Harman, 1965, Lipton, 2004, Campos, 2011],10 is also scientific: From observation
O made at time t1 and from a theory T that deductively or inductively entails O, one
can abductively infer that T must have been the case at earlier time t0 . In other words,
T is an explanation of why you have observed O. O:f course, it is not necessarily a
good, much less the best, explanation, but the more observations that T explains, the
better a theory it is.
In another form of abduction, from observation O1 made at time t1 , and from
observation O2 made at a later time t2 , one can abductively infer that O1 might have
caused or logically entailed O2 .
Like inductive inferences, abductive ones do not guarantee the truth of their
conclusion. Moreover, abductive inferences are deductively invalid! But they are at
the heart of the scientific method for developing and confirming theories.
10 For the origin of the term in the writings of the American philosopher Charles Sanders Peirce (who
pronounced his name like the word purse), see:

http://www.helsinki.fi/science/commens/terms/abduction.html (accessed 6 February 2013).
2.6. WHAT IS RATIONAL? 41
2.6.1.4 Non-Monotonic Logic

Non-monotonic reasoning is more psychologically real than any of the others. It
also underlies what the economist and AI researcher Herbert Simon called satisficing
(or being satisfied with having a reasonable answer to your question rather than an
optimal one), for which he won the Nobel Prize in Economics. (Well return to this
topic in Ch. 5, 6.)
In monotonic logics (such as deductive logics), once you have proven that a
conclusion C follows from a premise P, then it will always so follow. But in non-
monotonic logic, you might infer conclusion C from premise P at time t0 , but, at later
time t1 , you might learn that it is not the case that C. In that case, you must revise your
beliefs.
For example, you might believe that birds fly and that Tweety is a bird, from which
you might conclude that Tweety flies. But if you then learn that Tweety is a penguin,
you will need to revise your beliefs.
2.6.2 Scientific Rationality

If philosophy is a search for truth by rational means, what is the difference between
philosophy and science? After all, science is also a search for truth by rational means!
2.6.3 Is Science Philosophy?

Is the experimental or empirical methodology of science rational? It is not deductive.
But it yields highly likely conclusions, and is often the best we can get.
I would say that science is philosophy, as long as experiments and empirical
methods are considered to be rational and yield truth. Physics and psychology,
in fact, used to be branches of philosophy: Isaac Newtons Principiathe book
that founded modern physicswas subtitled Mathematical Principles of Natural
Philosophy (italics added), not Mathematical Principles of Physics, and psychology
split off from philosophy only at the turn of the 20th century. The philosophers
Aristotle and Kant wrote physics books. The physicists Einstein and Mach wrote
philosophy. And the philosophy naturalized movement in contemporary philosophy
(championed by the philosopher Willard Van Orman Quine) sees philosophy as being
on a continuum with science.11
But, if experiments dont count as being rational and only logic counts, then science
is not philosophy. And science is also not philosophy, if philosophy is considered to be
the search for universal or necessary truths, that is, things that would be true no matter
what results science came up with or what fundamental assumptions we made.
11 [Weatherson, 2012] considers how academic disciplines such as physics and psychology were once part
of philosophy but arent anymore, and then wonders, if other parts of philosophy became separate academic
disciplines, what would be left. His conclusion, possibly though not necessarily his view on what constitutes
the core of philosophy, is: ethics, epistemology, and metaphysics. (Or, as [Schwitzgebel, 2012] put it: The
issues at the foundation are always the same: What there really is [metaphysics], how we know about it
[epistemology], and what separates the god from the bad [ethics]. Cf. [Flanagan, 2012, B4] on the nature
of comprehensive philosohphies.
There might be conflicting world views (e.g., creationism vs. evolution, perhaps).
Therefore, the best theory is one that is consistent, that is as complete as possible (that
is, that explains as much as possible), and that is best-supported by good evidence.
You cant refute a theory. You can only point out problems with it and then offer a
better theory. Suppose that you infer a prediction P from a theory T and a hypothesis
H, and then suppose that the prediction doesnt come true (your experiment fails; that
is, the experimental evidence is that P is not the case). Then, logically, either H is not
the case or T is not the case (or both!). And, since T is probably a complex conjunction
of claims A1 , . . . , An , then, if T is not the case, then at least one of the Ai is not the case.
In other words, you need not give up a theory; you only need to revise it. (This is not
to say, however, that sometimes you should give up a theory. This is what happens in
the case of scientific revolutions, such as (most famously) when Copernicuss theory
that the Earth revolves around the Sun (and not vice versa) replaced the Ptolemaic
theory, small revisions to which were making it overly complex without significantly
improving it. Ill have a bit more to say about this in Ch. 4.)
2.6.4 Could (Should?) Philosophy Be Scientific?

Could philosophy be more scientific (that is, experimental) than it is? Should it
be? Colin McGinn takes philosophy to be a science (a systematically organized
body of knowledge), in particular, what he dubs ontical science: the subject
consists of the search for the essences of things by means of a priori methods
[McGinn, 2012b, McGinn, 2012a].
There is a relatively recent movement (with some older antecedents) to have
philosophers do scientific (mostly psychological) experiments in order to find out,
among other things, what ordinary people (for example, people who are not
professional philosophers) believe about certain philosophical topics. My feeling is that
this is not really philosophy, but rather an (interesting) branch of cognitive science.12
But there is another way that philosophy can be part of a scientific worldview.
This can be done by philosophy being continuous with science, that is, by being aware
of, and making philosophical use of, scientific results. Rather than being a passive,
archair discipline that merely analyzes what others say and do, philosophy canand
probably shouldbe a more active discipline, even helping to contribute to science
(and other disciplines that it thinks about). (For a useful discussion of this, which is
sometimes called naturalistic philosophy, see [Thagard, 2012]).
(Philosophers can also be more practical in the public sphere: The philosophers
have only interpreted the world in various ways; the point is to change it [Marx, 1845].
But an opposing point of view considers that philosophers. . . are ordained as priests
to keep alive the sacred fires in the altar of impartial truth [Cohen), 1919, 19]! For
more on this, see Ch. 5, 6.))
12 For more information on this movement, see [Nahmias et al., 2006, Appiah, 2007, Appiah, 2008,
Knobe, 2009, Beebe, 2011, Nichols, 2011]; for an argument against experimental philosophy, see
[Deutsch, 2009].
2.7. WHAT IS THE IMPORT OF PERSONAL SEARCH? 43
2.6.5 Is It Always Rational to Be Rational?

Is there anything to be said in favor of not being rational?
Suppose that you are having trouble deciding between two apparently equal
choices. This is similar to a problem from mediaeval philosophy known as Buridans
Ass (see [Zupko, 2011]): According to one version, an ass (that is, a donkey) was
placed equidistant between two equally tempting bales of hay but died of starvation
because it couldnt decide between the two of them. My favorite way out of such a
quandary is to imagine throwing a coin and seeing how you feel if it lands heads up:
If you would rather that it had landed tails up, then you know what you would have
preferred, even if you had rationally decided that both choices were perfectly equally
balanced.
For another consideration, look up Andrew N. Carpenters response to the question
To what extent do philosophers/does philosophy allow for instinct, or gut feelings?
on the AskPhilosophers website.13
Finally, an interesting discussion of the roleand limitsof rationality in AI
research is [Russell, 1995].
2.7 What Is the Import of Personal Search?

My major professor, Hector-Neri Castaneda (see [Rapaport, 2005a]) used to say that
philosophy should be done in the first person, for the first person. So, philosophy is
whatever I am interested in, as long as I study it in a rational manner and aim at truth
(or, at least, aim at the best theory). As the computer scientist Richard W. Hamming
[Hamming, 1998, 650] warned, In science and mathematics we do not appeal to
authority, but rather you are responsible for what you believe.
There is another way in which philosophy must be a personal search for truth. As
one introductory book puts it, the object here is not to give answers. . . but to introduce
you to the problems in a very preliminary way so that you can worry about them
yourself [Nagel, 1987, 67].
The point is not to hope that someone else will tell you the answers to your
questions. That would be nice, of course, but why should you believe them? The
point, rather, is for you to figure out answers for yourself.14
2.8 What Is the Import of In Any Field?

[He is a philosopher, so hes interested in everything. . . . [Said by the
philosopher David Chalmers about the philosopher Andy Clark, as cited
in [Cane, 2014].]
Philosophy also studies things that are not studied by any single discipline; these
are sometimes called the Big Questions: What is truth? What is beauty? What is
good (or just, or moral, or right)? What is the meaning of life? What is the nature of
13 http://www.askphilosophers.org/question/2992.
14 For a somewhat different take on the value of first-person thinking, see [Kornblith, 2013].
mind?15 Or, as Jim Holt [Holt, 2009] put it: Broadly speaking, philosophy has three
concerns: how the world hangs together, how our beliefs can be justified, and how to
live. The first of these is metaphysics, the second is epistemology, and the third is
ethics.
The main branches of philosophy are:
1. Metaphysics (or ontology),16 which tries to answer the question What is
there?17 (and also the question Why is there anything at all?). Some of
the things that there might be include: physical objects, properties, relations,
individuals, time, God, actions, events, minds, bodies, etc.
There are major philosophical issues surrounding each of these. Here are just a
few examples:
Which physical objects really exist? Do rocks and people exist? Or
are they merely collections of molecules? But molecules are constituted
by atoms; and atoms by electrons, protons, and neutrons. And, according
to the standard model, the only really elementary particles are quarks,
leptons (which include electrons), and guage bosons; so maybe those are
the only really existing physical objects.18 Here is a computationally
relevant version of this kind of question: Do computer programs that deal
with, say, student records model students? Or are they just dealing with 0s
and 1s? And, on perhaps a more fanciful level, could a computer program
model students so well that the virtual students in the program believe
that they are real?19
Do socially constructed things like money, universities, governments,
etc., really exist (in the same way that people or rocks do)?
Do properties really exist? Or are they just collections of similar (physical)
objects; in other words, is there a propertyRednessover and above
the class of individual red things? Sometimes, this is expressed as
the problem of whether properties are intensional (like Redness) or
extensional (like the set of individual red things).20
Are there any important differences between accidental properties (such
as my property of being a professor of computer science rather than my
being a professor of philosophy) and essential properties (such as my
property of being a human rather than being a laurel tree)?
15 A Rhymes with Orange cartoon (31 August 2007), titled The Big Questions, shows two men on the
beach; one says, Does turning the air conditioning down mean you are turning it up? Does turning it
up mean you are turning it down?, and the other replies, And is there a true nature of up and down?.
The caption is: Philosophy professors, the summer months.
16 For more information on ontology, see
http://www.cse.buffalo.edu/rapaport/563S05/ontology.html, accessed 9 September 2011.

17 For a computational approach to this ontological question, see
http://www.cse.buffalo.edu/rapaport/663/F06/course-summary.html, accessed 19 September 2011.

18 For an interesting take on this, see [Unger, 1979b, Unger, 1979a].
19 If this sounds like the film The Matrix, see Ch. 20.
20 For more information on the intensional-extensional distinction, see
http://www.cse.buffalo.edu/rapaport/intensional.html, accessed 19 September 2011.

2.8. WHAT IS THE IMPORT OF IN ANY FIELD? 45
What is the mind, and how does it relate to the brain? (This is known
as the mind-body problem.) Are minds and bodies two different kinds
of substances? Or are they two different aspects of some one, underlying
substance? Or are there no minds at all, but only brains? Or are there no
independently existing physical objects, but only ideas in our minds?
Do non-existents (e.g., Santa Claus) exist? We can and do think and talk
about them. Therefore, whether or not they exist in any sense, they do
need to be dealt with.21
And so on.
2. Epistemology is the study of knowledge and belief: How do we know what

there is? How do we know that there is anything? What is knowledge? Is
it justified, true belief, as Plato thought, or are there counterexamples to that
analysis (such as those of [Gettier, 1963])? Are there other kinds of knowledge,
such as knowing how to do something, or know someone by acquaintance or
by description? What is belief, and how does it relate to knowledge? Can
a computer (or a robot) be said to have beliefs or knowledge? In fact, the
branch of AI called knowledge representation applies philosophical results
in epistemology to issues in AI and computer science in general, and it has
contributed many results to philosophy as well (see, e.g., [Buchanan, 2006] AND
OTHERS).
3. Logic is the study of good reasoning: What is truth? What is rationality? Which
arguments are good ones? Can logic be computationally automated?
4. Ethics tries to answer What is good?, What ought we to do?. Well look at
some ethical issues arising from computer science in Chs. 18 and 20.
5. Ethics is closely related to social and political philosophy, which tries to

answer What are societies?, What are the relations between societies and the
individuals who constitute them?, What is the nature of law?.
6. Aesthetics tries to answer What is beauty?, What is art?. (Can computer

programs, like mathematical theorems or proofs, be beautiful?)
7. Philosophy is one of the few disciplines (history is another) in which the history
of itself is one of its branches: The history of philosophy looks at what famous
philosophers of the past believed and tries to reinterpret their views in the light
of contemporary thinking.
8. And of central interest for the philosophy of computer science, there are
numerous philosophies of:
21 See [Quine, 1948]. See also [Hirst, 1991], for a survey of the AI approach to this. And see
[Maida and Shapiro, 1982], [Rapaport, 1986a], [Wiebe and Rapaport, 1986], [Shapiro and Rapaport, 1987],
[Shapiro and Rapaport, 1991], [Rapaport et al., 1997] for some papers on a fully intensional AI approach to
these issues.
Philosophy of language tries to answer What is language?, What is

meaning?. It has large overlaps with linguistics and with cognitive science
(including AI and computational linguistics).
Philosophy of mathematics tries to answer What is mathematics?, What are
numbers?, Why is mathematics so applicable to the real world?. (For a useful
survey, see [Pincock, 2011]
Philosophy of mind tries to answer What is the mind?, How is the mind
related to the brain?.
Philosophy of science tries to answer What is science?, What is a scientific
theory?, What is a scientific explanation?. The philosophy of computer
science is part of the philosophy of science. The philosopher Daniel C. Dennett
has written that there was a reform that turned philosophy of science from an
armchair fantasy field into a serious partnership with actual science. There came
a time when philosophers of science decided that they really had to know a lot
of current science from the inside [Dennett, 2012, 12]. Although you do not
need to know a lot about computer science (or philosophy, for that matter) to
learn something from the present book, clearly the more you know about each
topic, the more you will be able both to understand what others are saying and
to contribute to the conversation. (We will look more closely at the philosophy
of science in Ch. 4.)
And, for any X, there is a philosophy of X, which is the philosophical
investigation of the fundamental assumptions, methods, and goals of X
(including metaphysical, epistemological, and ethical issues), where X could
be: mathematics (what is a number? is math about numbers, numerals, sets,
structures?), science, physics, biology, psychology, etc., including, of course, AI
and computer science. Philosophy is 99 per cent about critical reflection on
anything you care to be interested in (Richard Bradley, in [Popova, 2012].
X, by the way, could also be. . . philosophy! The philosophy of philosophy,
also known as metaphilosophy, is exemplfied by this very chapter, which is
an investigation into what philosophy is and how it can be done. Some people
might think that the philosophy of philosophy is the height of gazing at your
navel, but its really whats involved when you think about thinking, and, after
all, isnt AI just computational thinking about thinking?
Philosophy, besides being interested in any specific topic, also has an overarching
or topic-spanning function: It asks questions that dont fall under the aegis of specific
topics and that span multiple topics: The aim of philosophy, abstractly formulated, is
to understand how things in the broadest possible sense of the term hang together in
the broadest possible sense of the term [Sellars, 1962, 35].22
22 Sellars goes on to say:
the reflective knowing ones way around in the scheme of things. . . is the aim of philosophy
What is characteristic of philosophy os not a special subject-matter, but the aim of knowing ones
way around with respect to the subject-matters of all the special disciplines
It is therefore the eye on the whole which distinguishes the philosophical enterprise.
2.8. WHAT IS THE IMPORT OF IN ANY FIELD? 47
So, for instance, while it is primarily (but not only) mathematicians who are
interested in mathematics per se and primarily (but not only) scientists who are
interested in science per se, it is primarily (but not only) philosophers who are
interested in how and why mathematics is so useful for science (cf. [Smith, 2010]).
Are there any topics that philosophy doesnt touch on? Im sure that there are some
topics that philosophy hasnt touched on. But Im equally sure that there are no topics
that philosophy couldnt touch on.23
So, lets now begin our philosophical investigation into computer science.
23 See http://www.askphilosophers.org/question/3001, accessed 19 September 2011.


1. In addition to asking the big questions (what is existence? what is knowledge?
what is good? what is beauty?), philosophy asks synoptic questions (how do
things hang together)
2. my vision of the philosophers role has been gradually reshaped, ever more
in the direction of providing the conceptual clarifications and underpinnings
for theories that are testable, empirical, scientific. Dennett, Daniel (2003?),
Autobiography (Part 2), Philosophy Now, No. 69.
This has some of the synoptic character of Sellarss def but also emphasizes
the naturalistic side of philosophy.
3. Phil is a meta-discipline, esp. philosophy of X: In X, you think about X; in
philosophy of X, you think about thinking about X. The only subjects that might
be purely philosophical are metaphysics, epist, and ethics, yet each of those has
strong liks to things like physics, philosophy of mind, and political sci, inter alia.
4. Make sure that list of philosophy topics includes philosophy of mind and
philosophy of language, philosophy of art/aesthetics.
5. Why? is what children ask, which is why philosophy tackles qns that children
ask naturally.
6. discuss why (philosophy) vs how (sci)
7. Perhaps this should merely go under Further reading, with annotation:
Floridi, Luciano (2013), What Is a Philosophical Question?, Metaphilosophy
44(3) (April): 195221.
Argues that:
Philosophical questions are questions not answerable empirically or
mathematically, with observations or calculations. They are open
questions, that is, questions that remain in principle open to informed,
rational, and honest disagreement, even after all the relevant observations
and calculations have become available and the answers have been
formulated (pp. 200201, my emphasis).
Such questions dont have right/wrong answers; theyre Contextually
Relativistic. An example is whether one should accept Alices
invitation to her party (p. 209). A Multiplist would say that theres
no correct answer, so any answer is OK. A CR says that whether
you should accept or not depends on all sorts of supplementary
assumptions, and you have to try to make the best argument you can
for an answer.
And philosophical questions are also closed under further questioning
(p. 205).
Closed in the sense that the natural numbers are closed under
addition (p. 204), that is, adding two natural numbers always and
only yields another natural number, never anything else. So, if
you question . . . [a philosophical question], you obtain one more
philosophical question (p. 204)
And philosophical questions are also ultimate questions (p. 207).
Given the network of questions that mutually pose or presuppose each
other, ultimate questions are those questions whose answers are most
influential in terms of a cascade of further questions and answers to
other related questions within that network. (p. 207)
And they require noetic resources to be answered (p. 211), where,
by noetic resources, Floridi means resources that include. . . [peoples]
beliefs, what . . . [they]read on the web, their cultural background, their
language, religion, and art, their social practices, their memories. . . , their
expectations. . . , their emotional intelligence, their past experiences, and
so forth. It is the world of mental contents, conceptual frameworks,
intellectual creations, intelligent insights, dialectical reasonings. (pp. 210
211).
Since CR questions depend on their background assumptions, this fits.
And they are not absolute questions (p. 215), where absolute questions
are questions that are formulated with no regard for the kind of LoA [Level
of Abstraction] at which their possible answers become sensible (p. 212
213)., where, in turn, an LoA is, roughly, the units of measurement or
currency (p. 213) in which an answer must be given. So, for example,
to say that a car costs 5,000 is an absolute answer that makes no
sense unless its LoA is made clear: $5,000 or 5,000 Euros. According
to Floridi, [Turing, 1950] doesnt ask whether machines can think (in
an absolute sense) but whether machines can think at the Level of
Abstraction represented by the imitation game (p. 214).
8. I said that philosophy is the search for truth in any field. Heres an interesting
variation on that theme:
Wittgenstein claims that there are no realms of phenomena whose
study is the special business of a philosopher. . . [Horwich, 2013, my
italics]24
There are two ways to interpret the italicized passage: One way is to read it as
consistent with my claim that all fields are open to philosophical investigation.
But the way that Horwich (and Wittgenstein) intended more negatively: There
is nothing unique that is the object of study for philosophy (and that is not
the object of study of any other discipline). Whether or not this is the case
(and, admittedly, it is hard to find counterexamplesmetaphysics, possible
worlds, epistemologythat are not studied by other disciplines: physics, logic,
24 For related discussions, see [Gutting, 2012], [Lynch, 2013].
psychology), surely the methods that philosophers use and their approaches to
various problems differ from those of other disciplines.
9. The American philosopher John Dewey said:
Active, persistent, and careful consideration of any belief or supposed

form of knowledge in the light of the grounds that support it and
the further conclusion to which it tends constitutes reflective thought.
[Dewey, 1993, 9]25
In other words, its not enough to merely think something; you must also
consider reasons for believing it, and you must also consider the consequences
of believing it.
10. On progress in philosophy, or possibly on the relation of philosophy to sci, or

(better) on Marxs Fichte quote:
There is no way of telling upstream how great an impact any specific

bit of research will have. . . . Who would have guessed that the arcane
research done by the small set of mathematicians and philosophers
working on formal logic a century ago would lead to the development
of computing, and ultimately to completely new industries, and to the
reconfiguring of work and life across the globe? [ONeill, 2013, 8]
11. On philosophy as a conversation across the ages:
My mind does not simply receive impressions. It talks back to the

authors, even the wisest of them, a response Im sure they would
warmly welcome. It is not possible, after all, to accept passively
everything even the greatest minds have proposed. One naturally has
profound respect for Socrates, Plato, Pascal, Augustine, Descartes,
Newton, Locke, Voltaire, Paine, and other heroes of the pantheon
of Western culture; but each made statements flatly contradicted by
views of the others. So I see the literary and philosophical tradition
of our culture not so much as a storehouse of facts and ideas but rather
as a hopefully endless Great Debate at which one may be not only a
privileged listener but even a modest participant. [Allen, 1989, 2], as
cited in [Madigan, 2014, 46].
12. One of the things about philosophy is that you dont have to give up
on any other field. Whatever field there is, theres a corresponding
field of philosophy. Philosophy of language, philosophy of politics,
philosophy of math. All the things I wanted to know about I coudl
still study within a philosophical framework. Rebecca Newberger
Goldstein, cited in [Reese, 2014b]
25 As cited in [Neuberger, 2012, 293].
13. On abduction:
Adding a new hypothesis or axiom to a theory for the purpose of

explaining already known facts is a process known as abduction.
[Sloman, 2010, slide 56]
14. In addition to the two kinds of rationality discussed in 2.6 (logical rationality
and scientific rationality, [Heng, 2014] argues that,
a third, modern way of testing and establishing scientific truthin
addition to theory and experimentis via simulations, the use of
(often large) computers to mimic nature. It is a synthetic universe in a
computer. . . . If all of the relevant physical laws are faithfully captured
[in the computer program] then one ends up with an emulationa
perfect, The Matrix-like replication of the physical world in virtual
reality. [Heng, 2014, 174]
One consideration that this raises is whether this is really a third way, or just
a version of logical rationality, perhaps extended to include computation as a
kind of logic. (This might be a good place to discuss, or to forward-reference,
Bostrums theories on virtual realities, as well as Ch. 15.)
15. Does philosophy lead to skepticism? See
http://www.askphilosophers.org/questions/5572
16. Is philosophy worth doing? Or can science answer all of our questions? See:
(a) Linker, Damon (2014), Why Neil deGrasse Tyson Is a Philistine, This
Week (6 May), accessed 12 May 2014 from:
http://theweek.com/article/index/261042/why-neil-degrasse-tyson-is-a-
philistine
(b) Powell, Lewis (2014), An Open Letter to Neil deGrasse Tyson, The
Horseless Telegraph (8 May), accessed 12 May 2014 from:
http://horselesstelegraph.wordpress.com/2014/05/08/an-open-letter-to-
neil-degrasse-tyson/
(c) Pigliucci, Massimo (2014), Neil deGrasse Tyson and the Value of
Philosophy, Scientia Salon (12 May), accessed 13 May 2014 from:
http://scientiasalon.wordpress.com/2014/05/12/neil-degrasse-tyson-and-
the-value-of-philosophy/
17. Heres another nice definition of philosophy:
[Philosophy] seek[s] to question, clarify and cohere our assumptions
and concepts behind things. [Boey, 2014]
This is consistent with my view of how there can be progress in
philosophy (it can be backwards progress forged, not by inferring new
consequences from assumptions, but by questioning those assumptions
themselves [Rapaport, 1982]. Boey also notes that
this is why there exists the potential for there to be the phi of
pretty much anythingwhenever we critically question fundamental
assumptions about our beliefs, our activities, and the world around us,
we are in the business of doing philosophy.
18. NPR reporter Scott Simon once told future-NPR-reporter Tamara Keith (when
she was in high school), Consider majoring in philosophy. I did. . . . it taught
me how to break apart arguments, how to ask the right questions [Keith, 2014]
19. A very nice, early-and-later-imitated definition and characterization of

philosophy (here called metaphysics):
Metaphysics means only an unusually obstinate attempt to think

clearly and consistently. . . . A geologists purposes fall short of
understanding Time itself. A mechanist need not know how action
and reaction are possible at all. A psychologist has enough to do
without asking how both he and the mind which he studies are able
to take cognizance of the same outer world. But it is obvious that
problems irrelevant from one standpoint may be essential fro another.
And as soon as ones purpose is the attainment of the maximum
of possible insight into the world as a whole, the metaphysical
puzzles become the most urgent ones of all. [James, 1892, Epilogue:
Psychology and Philosophy, p. 427; my boldface]
20. Need to distinguish philosophical ontology from AI ontology.
21. Nice description of an issue in metaphysics:
What is reality? Is it a game, a simulation that minds beyhond our

own are skillfully creating for study or for fun? If so, we may
be just the characters in their game, foolishly thinking that we are
autonomous, sentient beings. Is reality something out there
something we simply perceive through our senses? Or is it something
in herederivative of the inner workings of the brain, a fabrication,
so to speak, of how this small mass of some 85 billion neurons
integrates stimuli and creates a sense of the real? [Gleiser, 2014, B13]
The first option, that reality is a game, has been studied by [Bostrom, 2003]
and suggested in a story by Stanisaw Lem that well discuss in Ch. 20 (see
also [Lightman, 2012]s novel Mr g). The other two options may be a false
dichotomy: Another option is that there is a reality out there, but that what we
experience as that reality is actually a construction of our brain.
22. . . . what is the use of studying philosophy if all that it does

for you is to enable you to talk with some plausibility about
some abstruse questions of logic, etc., & if it does not improve
your thinking about the important questions of everyday life. . . .
Ludwig Wittgenstein (1944), quoted in [Malcolm, 2001, 35];
http://wings.buffalo.edu/epc/authors/perloff/witt intro.html
(accessed 12 November 2014).
23. Two years! said Dantes. Do you think I could learn all this in
two years?
In their application, no; but the principles, yes. Learning does
not make one learned: there are those who have knowledge and
those who have understanding. The first requires memory, the second
philosophy.
But cant one learn philosophy?
Philosophy cannot be taught. Philosophy is the union of all
acquired knowledge and the genius that applies it. . . [Dumas, 1844,
Ch. 17, pp. 168169]
24. On epistemology:
Epistemology is concerned with the question of how, since we live,

so to speak, inside our heads, we acquire knowledge of what there is
outside our heads. [Simon, 1996a, 162]
25. On the mind-body problem:
With a large number of programs in existence capable of many kinds

of performances that, in humans, we call thining, and with detailed
evidence that the processes some of these programs use parallel
closely the observed human processes, we have in hand a clear-cut
answer to the mind-body problem: How can matter think and how
are brains related to thoughts? [Simon, 1996a, 164]
Simons answer is that certain patternings in matter, in combination with

processes that can create and operate upon such patterns [Simon, 1996a, 164]
can do the trick. (Comapare the discussion later in 9.4 on computers as magic
paper. Cite also the book Patterns in Stone.)
26. [McGinn, 2015b] argues that philosophy is a science, just like physics or
mathematics. More precisely, he says that it is the logical science of concepts
(pp. 8788). This paper might be considered as required (or whatever) reading
for this chapter.
27. Under truth, mention Tarski!
28. Under non-mono logic: Should satisficing be included here, or is it an orthogonal

issue?
29. A self-inconsistent theory can be refuted, or can it? After all, one can always
change ones logic! If your theory is inconsistent with an observation, you can
augment the theory with an account of how the observation is misleading. But
perhaps make this clearer in the text.
30. On first-person philosophy (2.7):
I do not pretend that I can refute these two views; but I can challenge
them. . . . [Popper, 1978, 4, p. 148]
This is the heart of philosophy: not (necessarily) coming up with answers, but
challenging assumptions and forcing you to think about alternatives.
My fathers favorite admonition was: Never make assumptions. That is, never
assume that something is the case or that someone is going to do something;
rather, try to find out if it is the case, or ask the person. In other words, challenge
all assumptions.
This is one way that progress can be made in philosophy: It may be backwards-
looking progress (because instead of looking forward to implications of
your assumptions, you look backwards to see where those assumptions might
have come from). (Besides these two directions of progress, there can be a
third, which is orthogonal to these two: Considering other issues that might
not underlie (backwards or follow from (forwards) the one that you are
considering, but that are inspired or suggested by it (sideways progress?).
The desire for answers to all questions has been called the Dualistic stance
towards knowledge ([Perry, 1970, Perry, 1981],
http://www.cse.buffalo.edu/rapaport/perry.positions.html).
But the Dualist soon realizes that not all questions have answers that everyone
agrees with, and some questions dont seem to have answers at all (at least,
not yet). Rather than stagnating (my term, not Perrys!) in a middle stance of
Multiplism (because not all questions have answers, all opinionsproposed
answersare equally good), a further stance is that of Contextual Relativism:
All proposed answers/opinions can (should!) be consideredand evaluated!
relative to and in the context of assumptions/reasons/evidence that can support
them.
Eventually, you Commit to one of these answers, and you become responsible
for defending your commitment to Challenges. But that is (just) more thinking
and analysismore philosophizing.
Moreover, the commitment that you make is a personal one (one that you are
responsible for). It is in this way that philosophy is done in the first person for
the first person, as Castaneda said.
And, finally:
And among the philosophers, there are too many Platos to enumerate.
All that I can do is try to give you mine. [Goldstein, 2014, 396]
And even more finally?:
. . . my purpose is
to put my own intellectual home in order. . . . [Putnam, 2001, as cited
at http://putnamphil.blogspot.com/2015/02/rational-reconstuction-in-
1976-when.html]
And:
The philosophy of every thinker is the more or less unconscious

autobiography of its auther, Nietzsche observed. . . . [Martin, 2015]
31. Add to the list of philosophies of X: philosophy of art, philosophy of law,

philosophy of religion, philosophy of education, philosophy of history.
32. On progress in philosophy:
Progress in philosophy consists, at least in part, in constantly bringing

to light the covert presumptions that burrow their way deep down
into our thinking, too deep down for us to even be aware of them.
. . . But whatever the source of these presumptions of which we are
oblivious, they must be brought to light and subjected to questioning.
Such bringing to light is what philosophical progress often consists
of. . . ... [Goldstein, 2014, 38]
Goldstein goes on to say that Plato said as much in his myth of the cave, in that
the cave-dwellers presume/assume that the shadows are reality, but that, to get
closer to reality (if not actually to reach it), they must challenge that assumption.
33. On argumentation in philosophy:
This is why philosophy must be argumentative. It proceeds by way of

arguments, and the arguments are argued over. Everything is aired in
the bracing dialectic wind stirred by many clashing viewpoints. Only
in this way can intuitions that have their source in societal or personal
idiosyncrasies be exposed and questioned. [Goldstein, 2014, 39]
The arguments are argued over, typically, by challenging their assumptions. (It
is rare that a philosophical argument will be found to be invalid, and, when that
happens, it is usually a source of disappointmentunless the invalidity is a very
subtle one or one that reveals a flaw in the very nature of logic itself, which is an
even rarer, but not unknown, occurrence.)
34. A nice motto for the beginning of this chapter:
And what is it, according to Plato, that philosophy is supposed to do?

Nothing less than to render violence to our sense of ourselves and our
world, our sense of ourselves in the world. [Goldstein, 2014, 40]
Because it is violent to have ones assumptions challenged.

35. And another one:
M C COY: . . . Or maybe you just dont have the courage of your

convictions.
PLATO: I would prefer the courage of my questions.
[Goldstein, 2014, 338]
36. On the nature of philosophy, consistent with the view of philosophical progress
as consisting in the challenging of assumptions:
[Plato] is. . . a man keenly aware of the way that assumptions and
biases slip into our viewpoints and go unnoticed, and he devised a
field devoted to trying to expose these assumptions and biases and to
do away with any that conflict with commitments we must make in
order to render the world and our lives maximally coherent. . . . Above
all, my Plato is the philosopher who teaches us that we should never
rest assured that our view, no matter how well argued and reasoned,
amounts to the final word on any matter. [Goldstein, 2014, 396]
and
Often philosophy, from Plato on, causes me to see that a concept is

problematic that I had always felt I could just take for granted.
Hilary Putnam (2015),
http://putnamphil.blogspot.com/2015/02/rational-reconstuction-in-
1976-when.html
(accessed 5 March 2015)
37. Thinking does not guarantee that you will not make mistakes. But not
thinking guarantees that you will. [Lamport, 2015, 41]
This is relevant to the idea of philosophy as thinking hard.
38. On philosophy as dangerous (Socrates):
. . . philosophy is difficult because the questions are hard, and the

answers are not obvious. We can only arrive at satisfactory answers
by thinking as rigourously as we can with the strongest logical and
analytical tools at our disposal.
. . . I want them [my students] to care more about thinks like truth,
clear and rigorous thinking, and distinguishing the truly valuable from
the specious.
The way to accomplish these goals is not by indoctrination.
Indoctrination teaches you what to think; education teaches you how
to think. Further, the only way to teach people how to think is to
challenge them with new and often unsettling ideas and arguments.
. . . Some people fear that raising such quetions and prompting
students to think about them is a dangerous thing. They are right.
As Socrates noted, once you start asking questions and arguing out
the answers, you must follow the argument wherever it leads, and it
might lead to answers that disturb people or contradict their ideology.
[Parsons, 2015]
A propos of this, the whole point of Western philosophy since Socrates has been
to get people to think about their beliefs, to question and challenge them. It is
not (necessarily) to come up with answers to difficult questions.
39. On first-person philosophy:
. . . Im not trying to change anyones mind on this question. I gave
that up long ago. Im simply trying to say what I think is true.
[Strawson, 2012, 146]
40. On theories of truth:
recommended reading = [McGinn, 2015a, 8.1], where McGinn identifies three
classical theories of truth:
The coherence theory states that a proposition is true if and only

if that proposition coheres [that is, is consistent] with the other
propositions that one believes. . . . The correspondence theory states
that a belief is true if and only if that belief corresponds to the facts.
. . . [T]he pragmatic theory of truth. . . is that a proposition is true if
and only if it is useful to believe that proposition. . . . The problem
with the coherence theory is that a belief couldbe consistent with my
other beliefs and yet the whole lot coud be false. . . . The pragmatic
theory. . . has a very similar problem. I could have a belief about
something that is useful to me but that belief is false. . . . Most phiers
think that the correspondence theory is the best theory. It captures the
idea that truth depends on objective realitynot on us. The problem
the correspondence theory has concerns more technical issues such
as what a fact is and what the correspondence relation amounts to.
(pp. 148149)
2.10 Further Sources of Information

1. Standard Reference Works (in print and online):
(a) Craig, Edward (ed.) (1998), Routledge Encyclopedia of Philosophy (London:
Routledge).
(b) Edwards, Paul (ed.) (1967), Encyclopedia of Philosophy (New York: Macmillan).
(c) Fieser, James; & Dowden, Bradley (eds.), Internet Encyclopedia of Philosophy,
http://www.iep.utm.edu/ (accessed 6 February 2013)
(d) Zalta, Edward N. (ed.), Stanford Encyclopedia of Philosophy
http://plato.stanford.edu/ (accessed 6 February 2013)
2. What Is Philosophy?
(a) Books and Essays:
Cline, Austin (2012), Introduction to Philosophy: Defining, Studying, Doing
Philosophy Is Important (about.com)
http://atheism.about.com/od/aboutphilosoph1/a/philosophyintro.htm
McGinn, Colin (2003), Finding Philosophy, Prospect, Issue 92 (November)
http://www.prospectmagazine.co.uk/magazine/findingphilosophy/
accessed 2 November 2012 (requires login)
A brief autobiography of how a well-known contemporary philosopher
got into the field.
Nagel, Thomas (1987), What Does It All Mean? A Very Short Introduction to
Philosophy (Oxford: Oxford University Press).
My favorite introduction to philosophy.
Russell, Bertrand (1912), The Problems of Philosophy, various editions and
publishers.
One of my favorite introductions to philosophy.
Russell, Bertrand (1946), Philosophy for Laymen, in his Unpopular Essays
(London: George Allen & Unwin);
http://www.users.drew.edu/jlenz/br-lay-philosophy.html
(accessed 13 January 2014).
Not so much about what philosophy is, but about why studying it is
important for everyone, not just professional philosophers. (Warning:
The online version has many typographical errors!)
Sinnott-Armstrong, Walter (2004), What Is Philosophy?
http://ismailispirit.wordpress.com/apporved-what-is-philosophy/
Tallis, Raymond (2013), Philosophy Isnt Dead Yet, The Guardian
(26 May); online at:
http://www.guardian.co.uk/commentisfree/2013/may/27/physics-philosophy-
quantum-relativity-einstein (accessed 16 July 2013)
Green, Daryn (2014), A Philosophical Round, Philosophy Now 105
(November/December): 22
A short poem illustrating the questioning nature of philosophy.
(b) Websites:
2.10. FURTHER SOURCES OF INFORMATION 59
Dennett, Daniel; & Steglich-Petersen, Asbjrn (2008), The Philosophical

Lexicon
http://www.philosophicallexicon.com/ (accessed 2 November 2012)
A satirical dictionary of philosophical jargon.
Florian, Meghan (2013), Notes from an Employed Philosopher, Femmonite
(31 January); online at:
http://www.femmonite.com/2013/01/notes-from-employed-philosopher.html
Interesting comments about how philosophy can make you feel alive
and prepare you for the real world.
Philosophers on Their Conceptions of Philosophy (Leiter Reports: A
Philosophy Blog, June 2010)
http://leiterreports.typepad.com/blog/2010/06/philosophers-on-their-
conception-of-philosophy.html
Philosophical Fun (Department of Philosophy, University of Rochester)
http://www.rochester.edu/College/PHL/fun.html
Rosenberg, Alex (2011), What Exactly Is Philosophy of Scienceand Why
Does It Matter?,
http://www.routledge.com/philosophy/articles/
what exactly is philosophy of science and why does it matter/
(accessed 23 July 2013)
Shammas, Michael (2012), For a Better Society, Teach Philosophy in High
Schools, Huffington Post (26 December); online at:
http://www.huffingtonpost.com/mike-shammas/for-a-better-society-
teac b 2356718.html (accessed 13 February 2013)
Suber, Peter (2003), Guide to Philosophy on the Internet
http://www.earlham.edu/peters/philinks.htm
What Is Philosophy? (Department of Philosophy, University of Melbourne)
http://philosophy.unimelb.edu.au/about/what-is-philosophy.html
What Is Philosophy? (Department of Philosophy, University of Nevada, Reno)
http://www.unr.edu/philosophy/2-what is philosophy.html
Why Philosophy? (Department of Philosophy, SUNY Buffalo)
http://philosophy.buffalo.edu/undergraduate/
Why Study Philosophy? (Indiana University Department of Philosophy)
http://www.indiana.edu/phil/undergraduate/why.shtml
AskPhilosophers.org questions:
i. What do people mean when they speak of doing philosophy?
http://www.askphilosophers.org/question/2915
ii. Why are philosophers so dodgy when asked a question?

iii. Are there false or illegitimate philosophies, and if so, whos to say which
ones are valid and which are invalid?
iv. What does it take to be a philosopher?
http://www.askphilosophers.org/question/4609 (accessed 13 February
2013)
3. Can There Be Progress in Philosophy? Can Philosophy Ever Solve Any of Its Problems?
(a) Books and Essays:
Moody, Todd C. (1986), Progress in Philosophy, American Philosophical
Quarterly 23(1) (January): 3546.
Rapaport, William J. (1982), Unsolvable Problems and Philosophical
Progress, American Philosophical Quarterly 19: 289-298.
Rapaport, William J. (1984), Can Philosophy Solve Its Own Problems?, The
[SUNY] News 13(May/June): F2F3
http://www.cse.buffalo.edu/rapaport/Papers/rapaport84b.canphilsolve.pdf
Rescher, Nicholas (1985), The Strife of Systems: An Essay on the Grounds and
Implications of Philosophical Diversity (Pittsburgh: University of Pittsburgh
Press).
(b) Websites:
Have philosophers ever produced anything in the way that scientists have?
(AskPhilosophers.org, 28 July 2008)
http://www.askphilosophers.org/question/2249 (accessed 2 November 2012)
How is philosophical progress made, assuming it is made at all?
(AskPhilosophers.org, 2 February 2012)
http://www.askphilosophers.org/question/4523 (accessed 2 November 2012)
4. What Is Metaphysics?
Quine, Willard van Orman (1948), On What There Is, Review of Metaphysics
2(5): 2138; reprinted in: Quine, From a Logical Point of View: 9
Logico-Philosophical Essays, Second Edition, revised (Cambridge, MA: Harvard
University Press, 1980): 119.
Hirst, Graeme (1991), Existence Assumptions in Knowledge Representation,
Artificial Intelligence 49: 199242.
A computational perspective on non-existents and other intensional items.
5. On Epistemology: What Is Knowledge?
Gettier, Edmund L. (1963), Is Justified True Belief Knowledge?, Analysis 23:
121123; reprinted in many anthologies, and online at:
http://www.ditext.com/gettier/gettier.html (accessed 2 November 2012)
6. How to Study and Write Philosophy:
(a) Books:
Martinich, A.P. (2005), Philosophical Writing: An Introduction (Malden, MA:
Blackwell).
Woodhouse, Mark B. (2007), A Preface to Philosophy, 8th edition (Wadsworth
Publishing).
(b) Websites:
Fiala, Andrew (2004), Student Survival Guide
http://www.wadsworth.com/philosophy d/special features/popups/survival guide.html
Miller, David Marshall (2011), Writing Philosophy Papers
http://people.duke.edu/dmmiller/Classes/Writing%20Philosophy%20Papers.docx
Suber, Peter (2003), Courses, http://www.earlham.edu/ peters/courses.htm
See especially the links at the bottom of the page under the heading
Files and links that pertain to more than one course.
Do a computer search for: "writing philosophy papers" (include the quotes in
the search box)
The Philosopher within You
Theres the legend of the fish who swam around asking every sea creature hed
meet, Where is this great ocean I keep hearing about?. . .
We are very much like that fish.
For consider, its hard to look at a newborn baby without thinking: what an
incredible miracle. But when was the last time you looked at an adult and had the
same thought? But why not? Every adult was a little baby; if the latter is a miracle
then so is the former. But it never occurs to us to think this way for one simple reason:
were so used to seeing people that we stop reflecting on them.
Or you drop something, a spoon, and it falls to the floor. But why? Couldnt it, in
theory, have remained floating in air or moved upwards? And how exactly does it fall
to the floor, by gravity? There are no strings connecting the earth to the spoon. How
can the earth pull on something froma distance, that its not even attached to? Why
dont we pause every time something drops and say: what an incredible miracle!
The most ordinary things contain a whole lifetime of questions, if only we are
reminded to start asking them.
Children already know to ask these questions. Every answer you provide to one
of their Why? questions just generates the next question. Btu we were all children
once. What we need to do now is to let the child still within usthe philosopher
within usre-emerge. . . . We need to take a cold wet plungh into the great deep ocean
of thought.
Its time to start thinking.
Andrew Pessin, The 60-Second Philosopher (London: Oneworld): 67.

Chapter 3
What Is Computer Science?
Thanks to those of you who [gave their own] faculty introductions [to the
new graduate students]. For those who [werent able to attend], I described
your work and courses myself, and then explained via the Reductionist
Thesis how it all comes down to strings and Turing machines operating
on them.
Kenneth Regan, email to University at Buffalo Computer Science &
Engineering faculty (27 August 2004); italics added.
The Holy Grail of computer science is to capture the messy complexity of

the natural world and express it algorithmically.
Teresa Marrin Nakra, quoted in [Davidson, 2006, 66].
Figure 3.1: 1984,

c Washington Post Co.
63
64 CHAPTER 3. WHAT IS COMPUTER SCIENCE?
3.1 Readings:
1. Required:
(a) Newell, Allen; Perlis, Alan J.; & Simon, Herbert A. (1967), Computer Science,
Science 157(3795) (22 September): 13731374.
(b) Knuth, Donald (1974), Computer Science and Its Relation to Mathematics,
American Mathematical Monthly 81(4) (April): 323343.
required: 13
very strongly recommended: 4
strongly recommended: readers who are more mathematically inclined may
wish to read the whole essay.
(c) Newell, Allen; & Simon, Herbert A. (1976), Computer Science as Empirical
Inquiry: Symbols and Search, Communications of the ACM 19(3) (March): 113
126.
i. For the purposes of this chapter, concentrate especially on what Newell &
Simon have to say about what computer science is:
read the Introduction (pp. 113114)
read from I. Symbols and Physical Symbol Systems to the end of the
subsection Physical Symbol Systems (pp. 114117)
read the Conclusion (pp. 125126)
ii. For a detailed follow-up, see:
Newell, Allen (1980), Physical Symbol Systems, Cognitive Science 4: 135
183.
Accessed 15 March 2014 from1 http://tinyurl.com/Newell1980
(d) Hartmanis, Juris, & Lin, Herbert (1992), What Is Computer Science and
Engineering?, in Juris Hartmanis & Herbert Lin (eds.), Computing the Future:
A Broader Agenda for Computer Science and Engineering (Washington, DC:
National Academy Press), Ch. 6, pp. 163216.2
required: Computer Science & Engineering, pp. 163168.
required: Abstractions in Computer Systems, pp. 168174.
very strongly recommended: skim the rest.
(e) Brooks, Frederick P., Jr. (1996), The Computer Scientist as Toolsmith II,
Communications of the ACM 39(3) (March): 6168.
required: pp. 6164.
very strongly recommended: skim the rest.
(f) Shapiro, Stuart C. (2001), Computer Science: The Study of Procedures,
http://www.cse.buffalo.edu/shapiro/Papers/whatiscs.pdf
1 http://onlinelibrary.wiley.com/store/10.1207/s15516709cog0402
2/asset/s15516709cog0402 2.pdf?v=1&t=hst07yxo
&s=e842daedf48a175f3ee71e69bd62f2f41d07b6ae
2 This book was the subject of a petition to withdraw it from publication because we consider it
misleading and even harmful as an agenda for future research

(http://www-formal.stanford.edu/jmc/petition/whysign/whysign.html).
The petition was sponsored by Bob Boyer, John McCarthy, Jack Minker, John Mitchell, and Nils Nilsson.
Commentaries on it appeared in [Kling et al., 1993].
3.1. READINGS: 65
2. Very Strongly Recommended (arranged in chronological order):
(a) Arden, Bruce W. (1980), COSERS Overview, in Bruce W. Arden (ed.), What Can
Be Automated? The Computer Science and Engineering Research Study (COSERS)
(Cambridge, MA: MIT Press), Ch. 1, pp. 131.
(b) Krantz, Steven G. (1984), Letter to the Editor about the relation of computer science
to mathematics, American Mathematical Monthly 91(9) (November): 598600.
(c) Denning, Peter J. (1985), What Is Computer Science?, American Scientist 73
(January-February): 1619.
(d) Abrahams, Paul (1987), What Is Computer Science?, Communications of the
ACM 30(6) (June): 472473.
(e) Loui, Michael C. (1987), Computer Science Is an Engineering Discipline,
Engineering Education 78(3) (December): 175178.
(f) Bajcsy, Ruzena K.; Borodin, Allan B.; Liskov, Barbara H.; & Ullman, Jeffrey D.
(1992), Computer Science Statewide Review (unpublished report)
http://tinyurl.com/BajcsyEtAl19923
(personal copy accessed 15 February 2013; login required)
(g) Abelson, Harold, & Sussman, Gerald Jay, with Sussman, Julie (1996), Structure
and Interpretation of Computer Programs, Preface to the First Edition
http://tinyurl.com/AbelsonSussman1996Pref4
(h) Gal-Ezer, Judith, & Harel, David (1998), What (Else) Should CS Educators
Know?, Communications of the ACM 41(9) (September): 7784.
contains a section titled What Is CS?
contains a Bibliography for What Is CS?
(i) Jacob, Christian (1999), What Is Computer Science?
http://tinyurl.com/Jacob19995
(j) Shagrir, Oron (1999), What Is Computer Science About?, The Monist 82(1):
131149.
Despite its title, this paper is more about what computers are and what
computation is; Shagrir assumes that computer science is the science of
computers.
(k) Johnson, George (2001), All Science Is Computer Science, The New York Times
(25 March): WK1, WK5 http://tinyurl.com/GJohnson20016 (accessed 15 February
2013)
(l) Foley, Jim (2002), Computing > Computer Science, Computing Research News
14(4) (September): 6; http://archive.cra.org/reports/computing/index.html
3 http://www.cse.buffalo.edu/rapaport/Papers/Papers.by.Others/bajcsyetal92.pdf
4 http://mitpress.mit.edu/sicp/full-text/book/book-Z-H-7.html#% chap Temp 4
5 http://pages.cpsc.ucalgary.ca/jacob/Courses/Fall99/CPSC231/03-ComputerScience.pdf
6 http://www.nytimes.com/2001/03/25/weekinreview/the-world-in-silica-fertilization-all-science-is-
computer-science.html
See also:
Glass, Robert L. (2003), Response to Computing > Comptuer Science,
online at: http://archive.cra.org/reports/computing/glass.html
(m) Boston University Department of Computer Science (2003), What Is Computer
Science?, http://cs-www.bu.edu/AboutCS/WhatIsCS.pdf
(n) Parlante, Nick (2005), What Is Computer Science?, InroadsThe SIGCSE
Bulletin 37(2) (June): 2425.
(o) Naur, Peter (2007), Computing Versus Human Thinking, Communications of the
ACM 50(1) (January): 8594.
3.2 Introduction
In a sense, the fundamental question of this book is: What is computer science? Almost
all of the other questions we will be considering flow from this one, so a final answer
(if there is one!) will have to await the end of the book.
In this chapter, we will look at several definitions of computer science that have
been given by both philosophers and computer scientists. Each definition raises issues
that we will examine in more detail later.
3.3 Why Ask What Computer Science Is?

Before we even try to answer the question, its worth asking a preliminary question:
Why should we even bother seeking a definition? There are at least two motivations
for doing so, academic ones and intellectual (or philosophical) ones.
3.3.1 Academic Motivations

3.3.1.1 Academic Politics
First, there are academic political reasons for asking what computer science is: Where
should a computer science department (or, as it is in my case, a computer science
and engineering department) be adminstratively housed? Intellectually, this might
not matter (after all, a small school might not even have academic departments,
merely teachers of various subjects). But deciding where to place a computer science
department can have political repercussions:
In a purely intellectual sense such jurisdictional questions are sterile

and a waste of time. On the other hand, they have great importance
within the framework of institutionalized sciencee.g., the organization
of universities and of the granting arms of foundations and the Federal
Government. [Forsythe, 1967b, 455]
Some possible locations include:

a college or school7 of arts and sciences
a college or school of engineering
a school of informatics.
Another possibility is that computer science (or engineering) should not be (merely) a
department, but an entire school or college itself, with its own dean (and perhaps with
its own departments: a department of theory of computation, a department of artificial
intelligence (AI), a department of software engineering, a department of computer
architecture, etc.).
7 Colleges and universities are usually administratively divided into smaller units, variously known as
schools, colleges, orgenericallydivision, each typically headed by a dean and divided into still
smaller units, called departments.
There are examples of each, even within a single university system: My own
university (State University of New York at Buffalo) has a Department of Computer
Science and Engineering within a School of Engineering and Applied Sciences, but,
when I joined the university, there were both a Department of Computer Science in a
Faculty of Natural Sciences and Mathematics and a separate Department of Electrical
and Computer Engineering in a Faculty of Engineering and Applied Sciences. At
its sibling institution, State University of New York at Albany, the Department of
Computer Science is in the College of Computing and Information. And at my
former college, State University of New York College at Fredonia, the Department
of Computer and Information Sciences is in the College of Arts and Sciences.
3.3.1.2 Naming the Discipline

When our discipline was newborn, there was the usual perplexity as to its
proper name. [Brooks, 1996, 61]
In fact, given all of these possibilities, we can even ask the question, What should
this discipline be called?:
Computer science?
Computer engineering?
Computing science (or engineering)?
Informatics? (This name is more common in Europe.)
In this book, but only for convenience, I will call it computer science, but, by doing
so, I do not mean to exclude the other parts of the discipline, such as engineering or its
relationship to information.
(For those who are interested, [Ceruzzi, 1989, esp. pp. 265270] contains a history
of the phrase computer science. And, in a response to a letter that appeared in one of
the earliest issues of Communications of the ACM, the editor (possibly Alan J. Perlis)
listed several, admittedly facetious, names, including turingineering, turology,
applied meta-mathematics, and applied epistemology [DATA-LINK, 1958, 6]. In
1966, Turing-award-winner Peter Naur suggested datalogy [Naur, 2007, 86].)
3.3.1.3 Academic Pedagogy

Another academic purpose for asking what computer science is concerns pedagogy:
What should be taught in an introductory computer science course?
Should it be a programming course (that is, is computer science the study of
programming)?
Should it be a computer literacy course (that is, is computer science all about
how to use computers)?
Should it be a course in the mathematical theory of computation (that is, is
computer science the study of computation)?
3.3. WHY ASK WHAT COMPUTER SCIENCE IS? 69
Should it be a course that introduces students to several different branches of

computer science, including, perhaps, some of its history?
And so on.
3.3.1.4 Academic Publicity

A related part of the academic purpose concerns publicity for prospective students:
How should a computer science department advertise itself so as to attract good

students?
How should the discipline of computer science advertise itself so as to encourage

primary- or secondary-school students to consider it as something to study in
college or to consider it as an occupation? (For more motivations along these
lines, see [Denning, 2013, 35].)
How should the discipline advertise itself so as to attract more women to the
field?
How should it advertise itself to the public at large, so that ordinary citizens
might have a better understanding of what it is?
Many of the definitions of computer science that you can find on various academic
websites are designed with one or more of these purposes in mind. Here is an exercise
for the reader: Try to find some of these, and see if you can figure out whether they
were designed with any of these purposes in mind.
3.3.2 Intellectual or Philosophical Purposes

Perhaps the academic (and especially political) purposes for asking what computer
science is are ultimately of little more than practical interest. But there are deep
intellectual or philosophical issues that underlie those questions, and this will be the
focus of our investigation:
What is computer science really?
Is it like any other academic discipline? (For instance, is it more like mathematics
or more like engineering?)
Or is it sui generis (a Latin phrase meaning own kind)?8
To illustrate this difference, consider these comments by two major computer

scientists (as cited in [Gal-Ezer and Harel, 1998, 79]): Marvin Minsky, a co-founder
of artificial intelligence, once said:
8 Here is a simple analogy: A poodle and a pit bull are both kinds of dogs. But a wolf is not a dog; it is
its own kind of animal. Actually, some biologists believe that dogs are actually a kind of wolf, but others
believe that dogs are sui generis.
Computer science has such intimate relations with so many other subjects
that it is hard to see it as a thing in itself. [Minsky, 1979]
On the other hand, Juris Hartmanis, a founder of the field of computational complexity
theory, said:
Computer science differs from the known sciences so deeply that it has to
be viewed as a new species among the sciences. [Hartmanis, 1993, 1] (cf.
[Hartmanis, 1995a, 10])9
3.4 What Does It Mean to Ask What Something Is?

We will not try to give a few-line definition of computer science since
no such definition can capture the richness of this new and dynamic
intellectual process, nor can this be done very well for any other science.
[Hartmanis, 1993, 5]
It does not make much difference how you divide the Sciences, for they
are one continuous body, like the ocean. [Leibniz, 1685, 220]
Before we explore our main question further, there is a fundamental principle that
should be kept in mind whenever you ask what something is, or what kind of thing
something is: There are no sharp boundaries in nature; there are only continua. A
continuum (plural = continua) is like a line with no gaps in it, hence no natural
places to divide it up. The color spectrum is an example: Although we can identify
the colors red, orange, yellow, green, and so on, there are no sharp boundaries where
red ends and orange begins. An apparent counterexample might be biological species:
Dogs are clearly different from cats, and there are no intermediary animals that are not
clearly either dogs or else cats. But both dogs and cats evolved from earlier carnivores
(it is said that both evolved from a common ancestor some 42 million years ago).10 If
we traveled back in time, we would not be able to say whether one of those ancestors
was a cat or a dog; in fact, the question wouldnt even make sense.
Moreover, although logicians and mathematicians like to define categories in terms
of necessary and sufficient conditions for membership, this only works for abstract,
formal categories. For example, we can define a circle of radius r and center c as the
set of all and only those points that are r units distant from c.11 As philosophers,
9 [Hartmanis, 1995a] covers much of the same ground, and in many of the same words, as
[Hartmanis, 1993], but is more easily accessible, having been published in a major journal that is widely
available online, rather than in a harder-to-find conference proceedings.
10 See the GreenAnswers.com webpage at:
http://greenanswers.com/q/95599/animals-wildlife/pets/what-common-ancestor-cats-and-dogs
(which cites the Wikipedia articles Feliformia (http://en.wikipedia.org/wiki/Feliformia) and Carnivora
(http://en.wikipedia.org/wiki/Carnivora)) and the gather.com webpage Miacids at:
http://www.gather.com/viewArticle.action?articleId=281474977041872 (accessed 15 February 2013).
(According to a Ziggy cartoon (http://www.gocomics.com/ziggy/2013/02/19), however, a platypus could be
considered as an intermediary animal between birds and mammals!)
11 All such points is the sufficient condition; only such points is the necessary condition: C is a
circle of radius r at center c if and only if C = {p : p is a point that is r units distant from c}. That is, p is r
3.4. WHAT DOES IT MEAN TO ASK WHAT SOMETHING IS? 71
psychologists, and cognitive scientists have pointed out, non-abstract, non-formal

(real) categories usually dont have such precise, defining characteristics. The
most famous example is the philosopher Ludwig Wittgensteins unmet challenge to
give necessary and sufficient defining characteristics for somethings being a game
[Wittgenstein, 1958, 66ff]. And the psychologist Eleanor Rosch has pointed out
that even precisely definable, mathematical categories can have blurry edges: Most
people consider 3 to be a better example of a prime number than, say, 251 GET
CITATION. (For further discussion, see [Rosch and Mervis, 1975], [Rosch, 1978],
[Mervis and Rosch, 1981], [Lakoff, 1987], [Hofstadter and Sander, 2013, esp. Ch. 7].)
So, we do not carve nature at its joints [Plato, 1961a, 265e266a, my emphasis];
rather, we carve nature at joints that are usually of our own devising: We impose
our own categories on nature. But I would not be a good philosopher if I did not
immediately point out that this claim is controversial! After all, isnt the point of
science to describe and explain a reality that exists independently of us and of our
concepts and categoriesthat is, independently of the joints that we carve into
nature? (Well return to the topic of the goal of science in Ch. 4.) And arent there
natural kinds? Dogs and cats, after all, do seem to be kinds of things that are there in
nature, independently of us.
Is computer science similar to such a natural kind? Here, I think the answer
is that it pretty clearly is not. There would be no academic discipline of computer
science without humans, and there probably wouldnt even be any computers without
us, either (though well see some reasons to think otherwise, in Ch. 9). So, to the extent
that it is we who impose our categories on nature, there may be no good answer to
the question What is computer science? beyond something like: Computer science is
what computer scientists do. (In a similar vein, [Abrahams, 1987, 472] says, computer
science is that which is taught by computer science departments. Perhaps intended
more seriously, [Denning, 1999, 1] defines The discipline of computer science. . . [as]
the body of knowledge and practices used by computing professionals in their work.)
But then we can ask: What is it that computer scientists do? Of course, one can beg that
last question by saying that computer scientists do computer science! [Hamming, 1968,
4] suggests something like this, but goes on to point out that there is often no clear,
sharp definition of. . . [a] field.12
Here is a continuation of the exercise that I suggested earlier, in 3.3.1.4: Link to
the websites for various computer science departments (including your own schools
website!) and make a list of the different definitions or characterizations of computer
science that you find. Do you agree with them? Do any of them agree with each other?
Are any of them so different from others that you wonder if they are really trying to
describe the same discipline?
Such advertising blurbs should not, perhaps, be taken too seriously. But there are
several published essays that try to define computer science, whose authorsall well-
units from c only if p is a point on C (that is, if p is r units from c, then p is a point on C); so, being a point
that is r units from c is a sufficient condition for being on C. And if p is a point on C, then p is r units from
c; so, being a point that is r units from c is a necessary condition for being on C.
12 [Bringsjord, 2006, my emphasis] argued that any answer. . . that includes some such notion as
Whatever computer scientists actually do. is unacceptable. I would replace includes by is limited to;
surely, includes has to be included.
respected computer scientistspresumably put a lot of thought into them. They are
worth taking seriously, which we will do in the next few sections.
3.5 Computer Science Studies Computers

The first such answer that we will look at comes from three computer scientists: Allen
Newell, Alan Perlis, and Herbert Simon:13
Wherever there are phenomena, there can be a science to describe

and explain those phenomena.. . . There are computers. Ergo, computer
science is the study of computers. [Newell et al., 1967, 1373, my
emphasis]
They then reply to several objections to this definition, such as this one: Sciences
study natural phenomena, but computers arent natural; they are artifacts. Their
reply to this is to deny the premise: Sciences are not limited to studying natural
phenomena. There are sciences of artifacts; for example, botanists study hybrid corn.14
In fact, Simon wrote a book (originally in 1969) called The Sciences of the Artificial
[Simon, 1996b], and [Knuth, 2001, 167] called it an unnatural science [because]
[c]omputer science deals with artificial things, not bound by the constraints of nature.
(For a commentary on this, see the Abstruse Goose cartoon in Figure 3.4 at the end of
the chapter.)
The objector might respond that the fact that Simon had to write an entire book to
argue that there could be sciences of artifacts shows that the premisethat science only
studies natural phenomenais not obviously false. Moreover, botanists study mostly
natural plants; hybrid corn is not only not studied by all botanists, it is certainly not the
only thing that botanists study (that is, botany is not defined as the science of hybrid
corn). Are there any natural phenomena that computer scientists study? (We will see a
positive answer to this question in 3.10.)
But lets not be unfair. There certainly are sciences that study artifacts in addition to
natural phenomena: Ornithologists study both birds (which are natural) and their nests
(which are artifacts); apiologists study both bees (natural) and their hives (artifacts). On
the other hand, one might argue (a) that beehives and birds nests are not human-made
phenomena, and that (b) artifact should be used to refer, not to any manufactured
thing (as opposed to things that are grown), but only to things that are manufactured
by humans, that is, to things that are not found in nature, so to speak. The obvious
objection to this claim is that it unreasonably singles out humans as being apart from
nature.15
Moreover, there once was a science that only studied a particular artifact
microscopes!
13 Newell was also a cognitive scientist, and Simon (in addition to being a cognitive scientist) was also an
economistin fact, a winner of the Nobel prize in economics.

14 Curiously, they say that it is zoologists who study hybrid corn!
15 For more on the nature of artifacts in general, see [Dipert, 1993], [Hilpinen, 2011]; on artifacts in
computer science, see [Mizoguchi and Kitamura, 2009].

3.5. COMPUTER SCIENCE STUDIES COMPUTERS 73
. . . Marcello Malpighi (16281694), was a great scientist whose work had

no dogmatic unity.16 He was one of the first of a new breed of explorers
who defined their mission neither by the doctrine of their master nor
by the subject that they studied. They were no longer Aristotelians
or Galenists. Their eponym, their mechanical godparent, was some
device that extended their senses and widened their vistas. What gave
his researches coherence was a new instrument. Malpighi was to be
a microscopist, and his science was microscopy. . . . His scientific
career was held together not by what he was trying to confirm or to
prove, but by the vehicle which carried him on his voyages of observation
[Boorstin, 1983, 376].
In a similar fashion, surely computers are device[s] that [have] extended [our] senses
and widened [our] vistas, and the science of computer scientists is, well, computer
science. (One of the two principal professional associations is the Association for
Computing Machinery.) What holds computer scientists together. . . [is] the vehicle
which carrie[s them] on [their] voyages of observation.
But this is not necessarily a positive analogy. Arguably, microscopy no longer
exists as an independent science. Now, if you search for Department of Microscopy
on the World Wide Web, you will, indeed, find that there are some universities and
museums that have one. But, if you look closer, you will see that they are really
departments of microbiology. Non-biologists who use microscopes (such as some
geologists or physicists) are not found in departments of microscopy today. What
has happened, apparently, is that the use of this artifact by scientists studying widely
different phenomena was not sufficient to keep them in the same academic discipline.
The academic discipline of microscopy splintered into those who use microscopes to
study biology, those who use it to study geology, and so on, as well as those who
build new kinds of microscopes (who might be found in an engineering or an optics
department). A former dean who oversaw the Department of Computer Science at
my university once predicted that the same thing would happen to our department:
The computer-theory researchers would move (back) into the math department; the
AI researchers would find homes in psychology, linguistics, or philosophy; those who
built new kinds of computers would move (back) into electrical engineering; and so on.
This hasnt happened yet, and I dont forsee it happening in the near future, if at all.
But it is something to consider. ([McBride, 2007] suggests that it is already happening;
[Mander, 2007] disagrees.)
It is, in fact, close to another objection: Computer science is really just a branch
of (choose at least one): electrical engineering, mathematics, psychology, etc. Newell,
Perlis, and Simon reply that, although computer science does intersect each of these,
there is no other, single discipline that subsumes all computer-related phenomena.
This, however, begs the question of whether computer science is a single discipline.
I used to be in a philosophy department; although certain branches of philosophy
were not my specialty (ethics and history of philosophy, for instance), I was expected
to, and was able to, participate in philosophical discussions on these topics. But
16 I assume that this means that Malpighi did not study any single, natural phenomenon, but that he studied
all such phenomena that are visible with a microscope.

my colleagues in computer science often do not, nor are expected to, understand
the details of those branches of computer science that are far removed from their
own. As a computer scientist specializing in AI, I have far more in common with
colleagues in the philosophy, psychology, and linguistics departments than I do with
my computer-science colleagues down the hall who specialize in, say, computer
networks or computer security. (And this is not just an autobiographical confession
on my part; my colleagues in computer networks and computer security would be the
first to agree that they have more in common with some of their former colleagues in
electrical engineering than they do with me.)
NOTE FOR NEXT DRAFT:
The applications of computers to a discipline should be considered

properly a part of the natural evolution of the discipline. . . . The mass
spectrometer has permitted significant advances in chemistry, but there is
no mass spectrometry science devoted to the study of this instrument.
[Loui, 1987, 177]
Similarly, the microscope has permitted significant advances in biology (and many
other disciplines) but microscopy is (no longer) a science devoted to the study of
that instrument. Presumably, contributions made by AI researchers to philosophy or
psychology should not be considered to be the results of AI research?
END NOTE
If computer science does study computers, but sciences properly so-called (the
natural sciences) do not study such artifacts, perhaps computer science isnt a
(natural) science? Perhaps it is really a branch of engineering. Newell, Perlis, and
Simon reply that it is both. But this is a major topic that we will look at in greater detail
later (especially in Ch. 5).
The most interestingand tellingobjection to their view is that computer science
is really the study, not (just) of computers, but of algorithms: very roughly, the
programs and rules that tell computers what to do. (Well devote a great deal of time
later looking at what algorithms and programs are, so, at this point, I will just assume
that you already have an idea of what they are and wont try to define them further.) For
example, [Bajcsy et al., 1992, 1] explicitly mention the (incorrect) assumption that. . .
[computer science] is based solely on the study of a device. . . .
What is interesting about this objection is how Newell, Perlis, and Simon reply:
They agree! They now say:
In the definition [that is, in their definition quoted at the beginning

of this section], computers means. . . the hardware, their programs or
algorithms, and all that goes along with them. Computer science is the
study of the phenomena surrounding computers [Newell et al., 1967,
1374, my emphasis].
Readers would be forgiven if they objected that the authors have changed their
definition! But, instead of making that objection, lets turn to an interestingly different,
yet similar, definition due to another celebrated computer scientist.
3.6. COMPUTER SCIENCE STUDIES ALGORITHMS 75
3.6 Computer Science Studies Algorithms

Donald Knuth, author of a major, multi-volume work on algorithms (called The Art
of Computer Programming [Knuth, 1973]) as well as developer of the TEX computer
typesetting system, gave this answer to our question:
[C]omputer science is. . . the study of algorithms [Knuth, 1974b, 323].
He cited, approvingly, a statement by George E. Forsythe [Forsythe, 1968] that the

central question of computer science is: What can be automated? (Presumably, a
process can be automatedthat is, done automatically, by a machine, without human
interventionif it can be expressed as an algorithm.)
He even noted [Knuth, 1974b, 324] that the name computing science might be
better than computer science, because the former sounds like the discipline is the
science of computing (what you do with algorithms) as opposed to computers (the
tools that do the computing) (cf. [Foley, 2002], [Denning, 2013, 35])17 . As he pointed
out, a person does not really understand something until he teaches it to someone
else. Actually a person does not really understand something until he can teach it to a
computer, i.e., express it as an algorithm [Knuth, 1974b, 327].18 (For a commentary
on this, see the XKCD cartoon in Figure 3.5 at the end of the chapter.)
But, as Knuth then went on to point out, you need computers in order to properly
study algorithms, because human beings are not precise enough nor fast enough to
carry out any but the simplest procedures [Knuth, 1974b, 323]. But are computers
really necessary? Even if they are, does that mean that computer science is (as Newell,
Perlis, and Simon claim) the study of computers?
Lets consider some similar questions for other disciplines: Do you need a compass
and straight edge to study geometry (or can you study it just by proving theorems about
points, lines, and angles)? Do you need a calculator to study physics or mathematics
(or do they just help you perform calculations more quickly and easily)? Do you need
a microscope to study biology? Even if you do, does that make geometry the study
of compasses and straight edges, or physics and math the study of calculators, or
biology the study of microscopes? With the possible exception of geometry, I think
most people would say that these disciplines are not studies of those tools.19 As the
computer scientist Edsger W. Dijkstra is alleged to have said, Computer Science is no
more about computers than astronomy is about telescopes.20 But the tools do make it
easier to study the subject matter.
So, whereas Newell, Perlis, and Simon say that computer science is the study
of computers (and related phenomena such as algorithms), Knuth says that it is the
17 Denning says: I have encountered less skepticism to the claim that computing is science than to
computer science is science.

18 Cf. [Schagrin et al., 1985, xiii], [Rapaport and Kibby, 2010, 2.4.2]. As the celebrated cellist Janos
Starker once said, When you have to explain what you are doing, you discover what you are really doing
[Fox, 2013].
19 Geometry might be different, because Euclid may have considered geometry to be the study of which
geometrical figures were constructible using only compass and straightedge [Toussaint, 1993] (see also
[Goodman, 1987, 4]). Well come back to this in Chapter 7.
20 https://en.wikiquote.org/wiki/Computer science#Disputed
study of algorithms (and related phenomena such as computers)!21 These are what are
called extensionally equivalent descriptions: They may approach the discipline from
different viewpoints (one from the viewpoint of a physical tool, one from the viewpoint
of an abstract procedure), but the bottom line is the sameonly the emphasis is
different.22
So, we now have two (only slightly different) definitions:
1. Computer science is the study of computers (and related phenomena such as the
algorithms that they execute).
2. Computer science is the study of algorithms (and related phenomena such as the
computers that execute them).
Both characterize computer science as a study, but I think that its pretty clear from
their discussions that they think of it as a science. Is it? To fully answer that, well
need to ask a preliminary question: What is science? Well do that in Ch. 4. Here, lets
look at a slight variant of the first of our two definitions.
3.7 Is Computer Science a Science?

Debates about whether [computer science is] science or engineering
can. . . be counterproductive, since we clearly are both, neither, and more
(mathematics is so much a part of what [it is]. . . . [Freeman, 1994, 27]
In a classic paper from 1976 [Newell and Simon, 1976], Newell and Simon updated
their earlier characterization. Instead of saying that computer science is the science
of computers and algorithms, they now said that it is the empirical study of the
phenomena surrounding computers, not just the hardware, but the programmed,
living machine. (Cf. [Abrahams, 1987, 472]: Computer science focuses, not on the
properties of particular computers, but on the properties of computers in general. It
goes beyond the study of computers as objects to, at least, the study of computers as
active agents). This may not sound too different from what Newell and Simon said
when they were writing with Perlis nine years earlier, but there are some interesting
differences:
First, here they emphasize the empirical nature of the subject: Programs running
on computers are experiments, though not necessarily like experiments in other
experimental sciences. In fact, one difference between computer science and other
experimental sciences is that, in computer science, the chief objects of study (the
21 Similarly to Knuth, [Gal-Ezer and Harel, 1998, 80] say that the heart and basis of the field is
algorithmics even though this does not cover the full scope of CS.
22 Heres a more mundane example of a similar phenomenon: Large supermarkets these days not only
carry groceries, but also pharmaceuticals, greeting cards, hardware, etc. Large drugstores these days not
only carry pharmaceuticals, but also groceries, greeting cards, hardware, etc. (And large general stores
(like Walmart) also carry pretty much the same mix of products.) Although each kind of store approaches
merchandising from different viewpointswe tend to think of Walgreens as a drugstore and Wegmans as a
supermarkettheir bottom line is the same; only their emphasis is different. On extensional equivalence,
see [Rapaport, 2009].
3.7. IS COMPUTER SCIENCE A SCIENCE? 77
computers and the progams) are not black boxes, that is, things whose internal
workings we cannot see directly but must infer from experiments we perform on them.
Rather, because we (not nature) designed and built the computers and the programs, we
can understand them in a way that we cannot understand more natural things.
Second, science may be the wrong term (note that the title of their paper mentions
only empirical inquiry). Computer science may not be a science (in the classic sense)
because it doesnt always strictly follow the scientific (or experimental) method.
(Well talk more about what that method is, in 4.6.) For example, often one experiment
will suffice to answer a question in computer science, whereas in other sciences,
numerous experiments have to be run.
Third, by programmed, living machines, they mean computers that are actually
running programs, not the machines just sitting there waiting for someone to use them,
nor the programs just sitting there on a piece of paper waiting for someone to load
them into the computer, nor the algorithms just sitting there in someones mind waiting
for someone to express them in a programming language. Sometimes, a distinction
is made between a program and a process: A program might be a static piece of
text or the static way that a computer is hardwired; a process is a dynamic entitythe
program in the process of actually being executed by the computer. (Well look at
some of these distinctions in more detail later, in Chapter 12; on the program-process
distinction, see [Eden and Turner, 2007, 2.2].)
If computer science is an empirical discipline (whether its a science, a study,
or an inquiry), it should have some empirical results, or what Newell and
Simon call laws of qualitative structure. They give examples of these for
well-established empirical disciplines: the cell doctrine (biology), plate techtonics
(geology), the germ theory of disease (medicine), and the atomic theory (physics)
[Newell and Simon, 1976, 115]. What about computer science? They cite two laws
of qualitative structure, but it should be emphasized that the two that they cite are
due primarily to them and are highly controversial: the Physical Symbol System
Hypothesis (which explains the nature of symbols) and Heuristic Search (which is a
problem-solving method).
Here, well focus on the first of these,23 which they state as follows
[Newell and Simon, 1976, 116]:
A physical symbol system has the necessary and sufficient means for
general intelligent action.
Another way of stating this is:
x is a physical symbol system if and only if x has the means for general
intelligent action,
where24
x is a physical symbol system =de f

23 Well look at the secondheuristicsin Ch. 7.
24 In this book, I use the symbol =de f to mean: means by definition.
1. x is a set of physical symbols with a grammatical syntax, together

with a set of processes that manipulate the symbols (where a process
can be thought of as a function or mapping or algorithm that takes
symbols as inputs and produces (other) symbols as output), and
2. x is a physical machine that produces through time an evolving
collection of symbol structures [Newell and Simon, 1976, 116].
In short, a physical symbol system is a physical Turing machine or universal computer.

We will see in great detail what that means in Chapter 8, but for now you can think of
it as your laptop computer, or even your smartphone.25
(It is interesting to note that the use of symbols to accomplish tasks in the real,
physical world bears some similarities to the way that magic is supposed to work
[Stevens, 2001]. See the appendix to this chapter. [See NOTES FOR NEXT DRAFT,
16, below.]
Note that Newell and Simon agree with Newell, Perlis, and Simon about what
computer science studies, and they agree with Knuth about its fundamental nature. But
lets examine these definitions a bit more closely before looking at others.
3.8 Physical Computers vs. Abstract Algorithms

As we saw at the end of 3.6, we have two apparently different views. Is computer
science the (scientific) study of the physical machine, or is it the (scientific) stufy
of the abstract algorithm? Of course, as Newell, Perlis, Simon, and Knuth pointed
out, each of these characterizations includes the other, so perhaps there really is no
difference other than one of emphasis or point of view. But consider that, in order to
study a computer, it seems to be necessary to know what it does; that is, it seems to
be necessary to know what algorithm it is executing. On the other hand, in order to
study an algorithm, it does not seem to be necessary to have a computer around that
can execute it; that is, the algorithm can be studied as a mathematical object, using only
mathematical techniques, without necessarily understanding it. It may be very much
more convenient, and even useful, to have a computer handy, as Knuth notes, but it
does not seem to be necessary.
If thats so (and well see an objection in a moment), then it would seem that
algorithms are really the essential object of study of computer science: Both views
require algorithms, but only one requires computers.
But is it really the case that you cannot study computers without studying
algorithms? Compare the study of computers with neuroscience: the study of brains.
Although neuroscience studies both the anatomy of the brain (its static, physical
structure) and its physiology (its dynamic activity), it generally treats the brain as a
black box: Its parts are typically named or described, not in terms of what they
do, but in terms of where they are located. It is as if a person from the 19th century
found what we know to be a laptop computer lying in the desert and tried to figure
out what it was, how it worked, and what it did (see [Weizenbaum, 1976, Ch. 5] for
25 A more detailed exposition is given in [Newell, 1980]. Also see [Rapaport, 2012b, 67n47].
3.8. PHYSICAL COMPUTERS VS. ABSTRACT ALGORITHMS 79
the source of this kind of thought experiment). They might identify certain physical
features: a keyboard, a screen, internal wiring (and if they were from the 18th century,
they might describe these as buttons, glass, and strings), and so on. More likely, they
would describe the device as we do the brain, in terms of the locations of the parts:
an array of button-like objects on the lower half, a glass rectangle on the upper half,
and so on. But without knowledge of what the entire system and each of its parts was
supposed to dowhat their functions were-they would be stymied. Yet this seems
to be what neuroscientists study. Of course, modern neuroscience, especially modern
cognitive neuroscience, well understands that it cannot fully understand the brain
without understanding its processing (its algorithms, if indeed it executes algorithms).
But this is a topic for another branch of philosophy: the philosophy of cognitive science
(see [Fodor, 1968], [Gazzaniga, 2010], [Piccinini, 2010a]).
So, it seems that we cannot study computers without studying algorithms, but we
can study algorithms without studying computers. So, even though it is quite obvious
that computer science does, in fact, study both, it seems that a definition of computer
science that focuses on its essence, on what it really is, would be one that takes
algorithms to be central. So, let us provisionally summarize our investigation so far:
Computer science is the (scientific) study of algorithms.
Knuth is not the only person to make this claim: [Denning et al., 1989, 16, my
emphasis] define computer science as the systematic study of algorithmic processes
their theory, analysis, design, efficiency, implementation, and applicationthat
describe and transform information. Note that they avoid the word science, perhaps
for political reasons (so as not to offend engineers). It is with implementation that
real, physical computers (as well as programming languages) enter their picture. But
it is also interesting to note that Denning et al. immediately go on to cite a version of
Forsythes definition: The fundamental question underlying all of computing is, What
can be (efficiently) automated.
The background to [Denning et al., 1989]s definition-by-committee (that is, a
klunky one designed to be acceptable to a variety of competing interests)26
is that, first, no fundamental difference exists between computer science and
computer engineering in the core material; second, there are three major
paradigms[:]. . . theory, [which] is rooted in mathematics. . . [,] abstraction (modeling),
[which] is rooted in the experimental scientific method. . . [; and] design, [which]
is rooted in engineering. . . . . . . [I]n computing[,] the three processes are so
intricately intertwined that it is irrational to say that any one is fundamental
[Denning et al., 1989, 10].
But there are still some questions to consider: First, there are the questions that we
have been putting off: Is it a science? What is a science? Second, there are questions
about whether there are any alternatives to it being a science (or other kind of study) of
algorithms (or computers). Lets continue to put off the questions about science, and
focus on what else computer science might be a science of.
26 The standard joke about such definitions is that a camel is a horse designed by a committee. See
http://en.wikipedia.org/wiki/Design by committee (accessed 22 April 2013).

3.9 Computer Science Studies Information

As we just saw, Denning et al. constrain the algorithmic processes to be only those
that concern information. Although this may seem to be overly narrow (after all, the
algorithmic processes that undoubtedly underlie your use of Facebook on your laptop,
tablet, or smartphone may not seem to be related to information in any technical
sense), but others also focus on this: [Hartmanis and Lin, 1992, 164] define computer
science this way:
What is the object of study [of computer science and engineering]? For the
physicist, the object of study may be an atom or a star. For the biologist,
it may be a cell or a plant. But computer scientists and engineers focus on
information, on the ways of representing and processing information, and
on the machines and systems that perform these tasks.
Presumably, those who study the ways of representing and processing are the
scientists, and those who study the machines and systems are the engineers. And,
of course, it is not just information that is studied; there are the usual related
phenomena: Computer science studies how to represent and (algorithmically) process
information, as well as the machines and systems that do this. (Question: Are humans
included among the machines and systems? After all, we represent and process
information, too!).
Similarly, [Bajcsy et al., 1992, 1] say that computer science is a broad-based
quantitative and qualitative study of how information is represented, organized,
algorithmically transformed, and used. More precisely, Computer science is
the discipline that deals with representation, implementation, manipulation, and
communication of information [Bajcsy et al., 1992, 2]. I think their second
characterization is too broad; I can imagine other disciplines (including journalism)
that also deal with these four aspects of information. But their first definition
contains a crucial adverbalgorithmicallyand thats what makes computer science
unique. Indeed, [Hartmanis and Lin, 1992, 164] say that The key intellectual
themes in CS&E [computer science and engineering] are algorithmic thinking, the
representation of information, and computer programs. But the second of these,
although an important branch of computer science (in data structures, knowledge
representation in AI, and database theory), is also studied by logicians. And the third
computer programmingalthough clearly another important branch of computer
science (in software engineering and program verifications) is, arguably, merely the
implementation of algorithms. So, once again, algorithms come to the fore.
But the term information as many people use it informally is at least ambiguous,
if not downright vague: It could refer to Claude Shannons mathematical theory of
information [Shannon, 1948], or to Fred Dretskes philosophical theory of information
[Dretske, 1981], or some other. As Michael Rescorla observes, Lacking clarification
[of the term information], the description [of computation as information
processing ] is little more than an empty slogan [Rescorla, 2015, 6.1].27
Here is a good question to think about: Are there any other kinds of algorithmic
processes besides those that manipulate information?
27 For more on the nature of information, see [Floridi, 2010].
3.10. COMPUTER SCIENCE IS A SCIENCE OF PROCEDURES 81
3.10 Computer Science Is a Science of Procedures

Here is yet another definition, though it is one that slightly modifies the algorithmic
point of view:
Computer Science is a natural science that studies procedures.

[Shapiro, 2001]
The procedures are, or could be, carried out in the real world by physical (biological,
mechanical, electronic, etc.) agents. Procedures are not natural objects, but they are
measurable natural phenomena, in the same way that events are not natural objects
but are natural phenomena. Procedures include, but are not limited to, algorithms.
Whereas algorithms are typically considered to halt and to produce correct solutions,
the more general notion of procedure allows for an algorithm-like entity that may be
vague (e.g., a recipe), that need not halt (e.g., that goes into an infinite loop either
by accident or on purpose, as in an operating system or a program that computes the
infinite decimal expansion of ), or that does not necessarily produce a correct solution
(e.g., a chess procedure does not always play optimally). (We will return to these issues
in Chs. 7 and 11.) Does computer science really study things like recipes? According
to Shapiro (personal communication), the answer is yes: An education in computer
science should help you write a better cookbook (because it will help you understand
the nature of procedures better)! (For a nice discussion of the difficulties of writing
recipes, see [Sheraton, 1981].)
And computer science is a science, which, like any science, has both theoreticians
(who study the limitations on, and kinds of, possible procedures) as well as
experimentalists. And, as Newell and Simon suggested in their discussion of empirical
results, there are fundamental principles of computer science as a science. Shapiro
cites two: The Church-Turing Thesis to the effect that any algorithm can be expressed
as a Turing-machine program (which we will discuss in Chs. 7, 8, and 10) and
the Boehm-Jacopini Theorem that codifies structured programming (which we will
discuss in Ch. 7).
Where do computers come in? A computer is simply a general-purpose procedure-
following machine. But does a computer follow a procedure, or merely execute
it? (Well come back to this in Ch. 10.)
How about the role of information? According to [Shapiro, 2001], computer
science is not just concerned with abstract information, but also with procedures that
are linked to sensors and effectors that allow computers to operate in the real world.
(Well examine the relationship between computers and the world in Ch. 17.)
3.11 Computer Science Is Not a Science

We have a number of open questions: Insofar as computer science studies both
algorithms and computers, we need to look further into what, exactly, algorithms are
(and how they are related to the more general notion of procedure), what kinds of
things they manipulate (information? symbols? real-world entities?), what computers
are, and how computers and algorithms are related to each other. All of this in due
time!
Another open question that we should look at is whether computer science is
a science. Dont forget, we are calling the subject computer science only for
convenience; it is not a tautology to say that computer science is a science nor is it a
self-contradiction to say that it computer science is not a science. Think of the subject
as being called by the 16-letter word computerscience that has as much to do with the
words computer and science as cattle has to do with cat.
[Brooks, 1996] says that it isnt an analytic science, presumably because, according
to him, it is not concerned with the discovery of facts and laws. Whether or not
Brookss notion of what science is is accurate will be our focus in Chapter 4. For
now, it will have to suffice to point out (a) that computer scientists who study the
mathematical theory of computation certainly seem to be studying scientific laws and
(b) that computer scientists like Newell, Simon, and Shapiro have pointed out that the
physical symbol system hypothesis or the Boehm-Jacopini theorem certainly seem to
be scientific theories, facts, or laws: Computer programming is an exact science in
that all the properties of a program and all the consequences of executing it in any
given environment can, in prinicple, be found out from the text of the program itself by
means of purely deductive reasoning [Hoare, 1969, 576, my italics].28 So, it certainly
seems that at least part of computer science is a science. But perhaps there are parts
that arent.
In particular, Brooks argues that computer science is an engineering discipline;
he views the computer scientist as a toolmaker. Computer scientists are concerned
with making things like physical computers and abstract tools like algorithms,
programs, and software systems for others to use. He uses J.R.R. Tolkiens phrase the
gift of subcreation to describe this concern. Computer science, he says, is concerned
with the usefulness and efficiency of the tools it makes; it is not, he says, concerned
with newness for its own sake (as scientists are). And the purpose of the tools is to
enable us to manage complexity. So, the discipline we call computer science is
really the synthetici.e., engineeringdiscipline that is concerned with computers.
Of course, before we can seriously consider whether computer science is really
an engineering discipline as opposed to a science, well need to look into what
engineering is, just as we will have to look into what science is. But here is an
argument for your consideration: Perhaps computer science is both a science and an
engineering discipline. First, it makes no sense to have a computer without a program:
The computer wouldnt be able to do anything. (It doesnt matter whether the program
is hardwired (in the way that a Turing machine is; see 8.12) or is a piece of software
(like a program inscribed on a universal Turing machines tape; see 8.13). But, second,
it also makes no (or at least very little) sense to have a program without a computer to
run it on. Yes, you can study the program mathematically (e.g., try to verify it (see
Ch. 16), study its computational complexity [Loui, 1996, Aaronson, 2013], etc.). But
what good (what fun) would it be to have, say, a program for passing the Turing test
that never had an opportunity to pass it? So, computers require programs in order for
them to do anything, and programs require computers in order for the program to be
28 Well look into this claim in more detail in Ch. 16.
3.12. SUMMARY 83
able to do anything. The computer scientist R.W. Hamming said:

Without the [computing] machine almost all of what we [computer
scientists] do would become idle speculation, hardly different from that
of the notorious Scholastics of the Middle Ages. [Hamming, 1968, 5]
This is reminiscent of the 18th-century philosopher Immanuel Kants slogan that
Thoughts without content are empty, intuitions without concepts are
blind. . . . The understanding can intuit nothing, the senses can think
nothing. Only through their union can knowledge arise. [Kant, 1787, 93
(A51/B75)]29
Similarly, we can say: Computers without programs are empty; programs without
computers are blind.30 Only through their union can computational processing
arise. So, computer science must be both a science (that studies algorithms) and an
engineering discipline (that builds computers).
But now, I think, were ready to look into all of these issues in more detail, so that
well be able to have a better grasp of exactly what computer science is. Before doing
that, lets sum up to see where we are.
3.12 Summary
What is computer science? As we have seen, there are several different, but related,
answers to this question.
The answers differ on several dimensions:
Category (or genus):31
Computer science is. . .

1. a science
2. a (systematic) study
3. a branch of engineering
Qualification (kind of science or study; i.e., differentia):
Computer science is. . .

1. (a) mathematical
29 For an informal presentation of some of Kants ideas, see [Cannon, 2013]. For a more philosophically
sophisticated introduction, see [Rohlf, 2010] and other articles on Kant in the online Stanford Encyclopedia
of Philosophy at: http://plato.stanford.edu/search/searcher.py?query=kant (accessed 15 November 2013).
30 Or is it the other way round?: Programs without computers are empty; computers without programs are
blind?
31 Here, I am using the word genus in the sense of Aristotles notion of a definition by genus and
differentia, where a term is defined by reference to its genus (what kind of thing it is) and its differentia
(how it differs from other things of that kind); see the Wikipedia article Genus-differentia definition,
http://en.wikipedia.org/wiki/Genus-differentia definition, accessed 30 September 2011.
(b) empirical
2. (a) natural
(b) artificial
(These cross-cut each other: That is, there are 4 possibilities.)
Subject matter:
1. computers (and related phenomena)
empirical, artificial?
2. (a) algorithms (and related phenomena)
(b) procedures (and related phenomena)
natural?, mathematical?, empirical/experimental?
3. information (and related phenomena)
natural?, mathematical?, empirical/experimental?
Finally, these may all turn out to be extensionally equivalentthat is, it may be the
case that computer science is an engineering discipline of computers if and only if it is
also science of algorithms, etc. The only differences may be the focus or point of
view that depend on what the related phenomena are in each case.
A more complete answer is going to depend on answers to many other questions,
including (but not limited to):
What is science?
What is engineering?
What is a computer?
What is an algorithm?
We will discuss some, but not all, of these topics later. For example, we wont
discuss mathematics vs. empirical sciences or natural vs. artificial sciences; if you are
interested in these questions, you should read books on the philosophy of science or
the philosophy of mathematics.
But well discuss science vs. engineering (Ch. 4 and Ch. 5), computers (Chs. 6 and
9), and algorithms and procedures (Ch. 7). And well begin by asking, If computer
science is a science, what is science?.

1. Need to split the question of what CS is into two parts:
(a) What is its object? (What does it study/investigate?)

Might want to put part of this off to Ch. 9 on what a computer is.
[Barwise, 1989] identifies three ways to understand what a computer
and computation is, in terms of what they traffic in: numbers,
symbol strings, information.
(b) What is its methodology? (Is it a sci, a branch of engg, math?...)
(c) see spreadsheet in Dropbox:WhatIsCS.xlsx
Actually, theres a prior division: CS is nothing but something else (the John
Ho/CS is math/CS is EE arguments; bring in microscopy here) vs. CS is its own
discipline; in the latter case, it has a methodology and an object (see above).
Still, it might be the case that its methodology and/or its object are (also) those
of some other discipline. Finally, it may have a new, unique methodology and/or
object (sui generis).
2. Then: Need to make a complete list of all published alternatives:
(a) object: computers, computing/algorithms/procedures, info, ...

method: sci, not a sci, engg, math, study...
Possibly do all of this chronologically, rather than logically:

1967: newell/simon/perlis: sci of comprs (etc.)
1974: knuth: study of algs (etc.)
1976: newell/simon: empl study/artif sci of comprs etc.
1987: loui: new kind of engg of algs for procg info
1989: denning: systematic study of algic procs over info
1992: hartmanis/lin: sci & engg of info
1996: brooks: engg of comprs
2001: shapiro: nat sci of procedures
3. Relevant to this might be Parlantes suggestion that computer science is a sort

of spectrum. . . with science on the one end and engineering on the other.
[Parlante, 2005, 24]
4. [Shannon, 1950], writing way before AI was created, wrote a

paper. . . concerned with the problem of constructing a computing routine or
program for a modern general purpose computer which will enable it to play
chess. That is, his pricipal question was whether chess can be mathematically
analyzed and, in particular, whether the kind of mathematical analysis to be
given was a computational one, presumably, a procedural one. (Another way
of phrasing the concern of his paper is: Can chess be played rationally?) This
is clearly a computer science problem from our modern perspective. So, the
methodology he advocates is that of mathematical analysis, and, in particular,
computational analysis. And the object of the analysis is a procedurethe

procedure of playing chess.
5. My def: Computer science is centrally concerned with the question: Which

functions are computable? Various branches of CS are concerned with
identifying which problems can be expressed by computable functions. For
instance, the central question of AI is whether the functions that describe
cognitive processes are computable. And since the only (or the best) way to
decide whether a computable function really does what it claims to do is to
execute it on a computer, computers become an integral part of CS. (This is
closest to Knuths definition, I think.)
E.g., devising computational methods of fingerprint identification is CS, because

its computer scientists who are trained in procedural methods for solving
problems. Though it is applied CS, perhaps.
E.g., Shannons 1950 paper on chess: The principal question is: Can we
mathematically analyze chess? In partic, can we computationally analyze
it (suggesting that computational analysis is a branch/kind of mathematical
analysis)i.e., can we analyze it procedurally? I.e., can we play chess
rationally?
E.g., is the weather computable? See [Hayes, 2007a].
E.g., faculty members in my department recently debated whether it was possible

to write a computer program that would schedule final exams with no time
conflicts and in rooms that were of the proper size for the class. Some thought
that this was a trivial problem; others thought that there was no such algorithm
(on the grounds that no one in the university administration had ever been able
to produce such a schedule). So, is final-exam-scheduling computable?
6. Discussion of Arden 1980:
(a) (Untitled) preface on the (unnumbered) pages immediately preceding p. 1:

A riff on Forsythes 1968 quote.
Theres an interestingly subtle difference between my What can be
computed? and Forsythes What can be automated?: The latter
allows, a bit more clearly, the possibility of including computers, and
not just computing, in the subject matter, because automation has a
machine-oriented connotation that computation lacks.
(b) p 4: Distinguishes between computer science research and computer
engineering research, suggesting that he (or they) think its easier/simpler
to keep these separate, rather than trying to fit both under a single label.
(c) Sections About Names and Labels (pp. 57):
Has a useful discussion about the meanings of such terms as

computation, computer, information, etc., and about science vs.
engineering.
p. 6: Argues that neither math nor computer science are sciences,
because their objects are not natural phenomena. In the case of
math, the object is human-produced systems. . . concerned with the
development of deductive structures. But this gets us too deeply
into the controversial philosophy of math (Platonism vs. logicism vs.
syntacticism vs. intuitionism. In the case of computer science, the
object is man-made. But what is the object? Computers? Yes,
theyre clearly man-made, and this leads us back to the Perlis et al.
arguments. Algorithms? Theyre only man-made in whatever sense
mathematical structures are.
pp. 67: Argues that engineering is concerned with implementation,
rather than understanding (6), which is the best distinction between
engineering and science (7). Because of multiple implementations of
a single theory, questions of efficiency come to the fore in engineering
(7), and much of computer science is concerned with. . . efficiency
(7). But surely complexity theory is on the mathematical/scientific
side of the science/engineering border.
p 7: But then suggests that implementations need not be
physically realized (cf. [Rapaport, 1999, Rapaport, 2005b]), so that the
implementation/understanding distinction doesnt really capture the
engineering/science distinction. Suggests that the engineering/science
spectrum matches the physical/theoretical spectrum, with no clear
dividing line, and chalks this up to the lack of a natural phenomenon
that is the object of study. But [Shapiro, 2001] argues that procedures
are the natural phenomena under study.
(d) Section Some Definitions, pp. 5-10:
Accepts [Newell et al., 1967]s the phenomena related to computers
(Arden, p.7) as the subject of study.
pp. 78: Points out that the non-natural nature of this subject matter is
mitigated by its complexity.
p. 8: Suggests a dilemma: Natural systems are ipso facto interesting
to study (presumably because theyre out there, so we should be
curious about them). But why study a synthetic system if it has no
rationale? On the other hand, . . . if there is external motivation,
why not study that external motivation directly?
i. But the first horn of the dilemma is countered by pure
mathematics.
ii. Ant the second horn can be countered by taking [Shapiro, 2001]s
position.
Contra both [Newell et al., 1967] and, in particular, [Knuth, 1974b],
the study of algorithms and the phenomena related to computers are
not coextensive, since there are important organizational, policy, and

nondeterministic aspects of computing that do not fit the algorithmic
mold (p. 9).
i. I dont think that either [Newell et al., 1967] or [Knuth, 1974b]
had those things in mind.
ii. And if phenomena related to computers is taken that widely,
then it encompasses pretty much everything, thus making any
definition based on such a wide notion virtually useless. The
classical sciences (physics, chemistry, biology, etc.) also have
important organizational, policy, and nondeterministic aspects,
but those arent used in trying to define what those sciences are
about.
Suggests, but does not endorse, a committee-produced, all-purpose
definition: computer science is the study of the design, analysis, and
execution of algorithms, in order to better understand and extend the
applicability of computer systems (p. 9). Note that this avoids the
science/engineering quandary, by its use of study, and tries to cover
all the ground. Suggests, however, that the entire book of which this is
the overview should be taken as the elaboration of this definition,
which is a bit like saying that computer science is what computer
scientist do.
(e) Contains the following sample questions that are more specific than
What can be automated? (p. 10):
i. Are there useful, radically different computer organizations yet to be
developed?
ii. What aspects of programming languages make them well suited for
the clear representation of algorithms in specific subject areas?
iii. What principles govern the efficient organization of large amounts of
data?
iv. To what extent can computers exhibit intelligent behavior?
v. What are the fundamental limits on the complexity of computational
processes?
vi. What are the limits on the simulation of physical systems?
vii. How can processes be distributed in a communicating network to the
advantage of microtechnology?
viii. To what extent can numerical computation be replaced by the
computer manipulation of symbolic expressions?
ix. Is it possible to replace a major part of system observation and testing
with automated proofs of the correctness of the constituent algorithms?
(f) Also contains the following list of subject areas of CS:
i. Numerical computation
ii. Theory of computation
iii. Hardware systems

iv. Artificial intelligence
v. Programming languages
vi. Operating systems
vii. Database systems
viii. Software methodology
From this, he concludes that What can be automated? means: What
are the limits, imposed by complexity, on the application of computers?
NOTE: I cant find this quote in the Overview. In any case, asking what
can be automated logically entails the possibility that some things cannot
be automated, hence that there are limits. But these limits may have nothing
to do with complexity: They could be entirely theoretical (as in the Halting
Problem), or practical (time and space limitations, some of which might be
theoretically interesting, as in computational complexity theory, but some
of which might be uninteresting, such as if a research group runs out of
funds before being able to automate something), or ethical.)
(g) Contains a good, short history of modern computing (A Brief History,
pp. 1013.
(h) the basic question[:] . . . what can and what should be automated (p. 29).
Note that should can have both an efficiency answer (X shouldnt
be automated, because its more efficient not to do so) and an ethical
answer (X shouldnt be automated because its unethical/immoral to do so;
[Tedre, 2015, 167168]).
7. From Abelson & Sussman 1996:
Underlying our approach to this subject is our conviction that

computer science is not a science and that its significance has little
to do with computers. The computer revolution is a revolution in
the way we think and in the way we express what we think. The
essence of this change is the emergence of what might best be called
procedural epistemology the study of the structure of knowledge
from an imperative point of view, as opposed to the more declarative
point of view taken by classical mathematical subjects. Mathematics
provides a framework for dealing precisely with notions of what is.
Computation provides a framework for dealing precisely with notions
of how to.
How to is certainly important, and interestingly distinct from what is. But
consider Euclids Elements; it was originally written in how to form: To
construct an equilateral triangle, do such-and-such
(http://tinyurl.com/kta4aqh32 [Toussaint, 1993]). But today it is expressed in
what is form: The triangle that is constructed by doing such-and-such is
32 http://www.perseus.tufts.edu/hopper/text?doc=Perseus%3Atext%3A1999.01.0086%3Abook%3D1%3Atype%3DProp%3Anumber%3D1
equilateral. So, is there really a difference between how to and what is?
(See also [Goodman, 1987, 4].)
8. On Malpighi:
The debate over the appropriate place of computing in grade schools
and high schools echoes the debate in universities decades ago,
when computers and software were initially seen as mere plumbing.
And certainly not something worthy of study in its own right. A
department of computer science? Why not a department of slide
rules? From: Lohr, Steve (2008) Does Computing Add Up in the
Classroom?, New York Times (1 April).
9. include various taxonomies of subfields of CS; see

http://www.cse.buffalo.edu/rapaport/584/whatiscs.html 9b
10. discuss sui generis; include this quote
Trying to categorize computing as engineering, science, or math is
fruitless; we have our own paradigm. Denning & Freeman 2009
11. Fant 1993 argues that the notion of process is more central to CS than
the notion of algorithm. His main argument seems to be that the notion
of algorithm is too narrow, requiring halting, disallowing randomization or
probabilisitic methods, etc. He is also more concerned with the expression of
an algorithm than with the algorithm itself, the latter of which he considers to be
an object of mathematical study.
See also Guzdial 2008, esp citation to Alan Perlis; Also on the nature of both
procedural thinking and computational thinking more generally, see: Carey
2010.
12. Two quotes from Bernard Chazelle, Prof. of CS, Princeton:

CS is the new new math, and people are beginning to realize that
CS, like math, is unique in the sense that many other disciplineswill
have to adopt that way of thinking. It offers a sort of conceptual
framework for other disciplines, and thats fairly new.
Were at the tail end of Moores Law....But this will be the best thing
that can happen to CS. Moores Law has been tremendously beneficial
to society. At the same time, its been so damn powerful that it has set
back the development of algorithms. But thats about to change. Any
student interested in science and technology needs to learn to think
algorithmically. Thats the next big thing.
From Anthes 2006
13. On difference between CS and science:
Where physical science is commonly regarded as an analytic

discipline that aims to find laws that generate or explain observed
phenomena, CS is predominantly (though not exclusively) synthetic,
in that formalisms and algorithms are created in order to support
specific desired behaviors. Hendler et al. 2008:63
14. On CS as math:
That computer science is somehow a mathematical activity was a
view held by many of the pioneers of the subject, especially those who
were concerned with its foundations. At face value it might mean that
the actual activity of programming is a mathematical one. Indeed, at
least in some form, this has been held. But here we explore a different
gloss on it. We explore the claim that programming languages are
(semantically) mathematical theories. This will force us to discuss the
normative nature of semantics, the nature of mathematical theories,
the role of theoretical computer science and the relationship between
semantic theory and language design. Turner 2010
15. On the nature of defs, discuss the def of planet. See, e.g., [Lemonick, 2015].
Exact definitions are undoubtedly necessary but are rarely perfect

reflections of reality. Any classification or categorization of reality
imposes arbitrary separations on spectra of experience or objects.
Matt Craver, letter to the editor of SciAm, May 2007, p. 16. also
see reply by Soter on same page.
Cf. the def of alg, culminating in the Ch-T thesis.
Typically, when formalizing an informal notion, you exclude some old

thingsi.e., things that were informally considered to fall under the notion
and you include some new thingsi.e., things that you hadnt previously
considered to fall under the notion. The attitude towards the former is sour
grapes: Our intuitions were wrong; those things really werent Xs after all. The
attitude towards the latter is: Wow! Thats right! Those things really are Xs! Cf.
not only the Ch-T thesis, but also the def of thinking as a kind of computation.
16. On magic:
Figure 3.2: 2014,

c John L. Hart FLP
The entities that populate the referent node [of the meaning triangle
consisting of words, concepts (abstract, or in the head), and referents
(concrete, and part of the external world)] are usually construed
as objects out there. . . because they are independent of what any
person thinks or says. The qualifier out there is to capture the
intution that words can be uttered, and thought can be generated,
but no amount of thinking or talking is going to bring a table or cat
or chair into being. (p. 507) Regoczei, Stephen; & Hirst, Graeme
(1990), The Meaning Triangle as a Tool for the Acquisition of
Abstract, Conceptual Knowledge, International Journal of Man-
Machine Studies 33: 505520.
Thats exactly what magic does: spells bring physical objects and events into
being. (syntax produces semantics!)
But
There are socially constructed abstract entities, especially in the

fields of law, culture, interpersonal interactions, organizations, and
mathematical sciences, which are brought about through talking and
thinking by individuals within a group. (p. 507)
I.e., they are magical!
The programmer, like the poet, works only slightly removed from
pure thought-stuff. He [sic] builds castles in the air, creating by the
exertion of the imagination. . . . Yet the program construct, unlike
the poets words [or the magicians spells?], is real in the sense
that it moves and works, producing visible outputs separate from the
construct itself. . . . The magic of myth and legend has come true in our
time. One types the correct incantation on a keyboard, and a display

screen comes to life, showing things that never were nor could be.
[Brooks, 1975, 78, my emphasis].33
Of course, the main difference between the magic of myth and legend and
how computers work is that the former lacks (or at least fails to specify)
any causal connection between incantation and result. As Arthur C. Clarke
said, Any sufficiently advanced technology is indistinguishable from magic
(http://en.wikipedia.org/wiki/Clarkes three laws).
33 For a nice illustration of this, see Fig. 3.3.

Since classical times sophisticated forms of mysticism that relied on

written notations and formulas have been called hermetic magic, after
Hermes Trismegistus, Greek variant of the Egyptian Thoth, regarded
as the god or principle of wisdom and the originator of writing.
[Stevens, 1996, 721, col. 1]
Clearly, computer programming relies on written notations and formulas. (Of

course, its not clear whether its mysticism!)
In anthropology, magic generally means beliefs in the use of

symbols to control forces in nature. . . . [Stevens, 1996, 721, col. 1]
A definition of magic can be constructed to say something like the
following: Magic involves the human effort to manipulate the forces
of nature directly, through symbolic communication and without
spiritual assistance. [Stevens, 1996, 723, col. 2].
The first quote is even better, because it has no mystical overtones. Clearly,
programming involves exactly that kind of use of symbols (which well look at
in more detail when we investigate the relation of computers to the world, in
Ch. 17). The second quote either clearly includes the use of computers, or else
is too broad a definition of magic!
[James G.] Frazer [[Frazer, 1915]] had suggested that primitive

people imagine magical impulses traveling over distance through a
kind of invisible ether. [Stevens, 1996, 722, col. 1]
That sounds like a description of electricity: Think of electrical currents running

from a keyboard to a CPU, or even of information traveling across the Internet.
The magical act involves three components: the formula, the rite,
and the condition of the performer. The rite consists of three
essential features: the dramatic expression of emotion through gesture
and physical attitude, the use of objects and substances that are
imbued with power by spoken words, and, most important, the
words themselves. [Stevens, 1996, 722, col. 2, citing Bronislaw
Malinowski]
Ignoring the dramatic expression of emotion, use of a compuer involves

gestures with a mouse, a trackpad, or a touchscreen; the computer itself can be
thought of as imbued with power when we issue a command, either spoken or
typed; and the words used by the programmer or user are surely importantso
the rite criterion is satisfied. Its not clear what Stevens (or Malinowski) have
in mind by formula or by condition of the performer, but computer programs
can be thought of as formulas, and only those programmers who know how to
use appropriate programming languages, or those users who have accounts on a
computer, might be considered to be in the right condition.
Figure 3.3: Computers can show. . . things that never were nor could be.
[A symbol] can take on the qualities of the thing it represents, and

it can take the place of its referent; indeed, as is evident in religion
and magic, the symbol can become the thing it represents, and in so
doing, the symbol takes on the power of its referent. [Stevens, 1996,
724, col. 1, my italics]
First, we see this happening in computers when we treat icons (which are a kind
of symbol) on a desktop or a WYSIWYG word processor as if they were the
very things they represent. But, perhaps more significantly, we see this in the
case of those computer simulations in which the simulation of something is that
(kind of) thing, as in online banking, where the simulation of transferring funds
between accounts is the transferring of funds. (See [Rapaport, 2012b, 8] for
further discussion.) Computers are not just metaphorically magic (as Arthur C.
Clarke might have said); they are magic!
Everything going on in the software [of a computer] has to be

physically supported by something going on in the hardware.
Otherwise the computer couldnt do what it does from the software
perspectiveit doesnt work by magic. But usually we dont have to
know how the hardware worksonly the engineer and the repairman
do. We can act as though the computer just carries out the software
instructions, period. For all we care, as long as it works, it might
as well be magic. [Jackendoff, 2012, 99]
I am tempted to lay claim to the term magic. A lot of what

computer scientists do is now seen by the lay public as magical.
Programmers (especially systems engineers) are often referred to as
gurus, sorcerers, and wizards. Given the lofty goals of white magic,
understanding the power of names and the value of pure thought,
the power of labels is indeed attractive. However, many historically
compelling reasons argue against this connotation, including the sad
history of Isaac Newtons alchemical pursuit of the philosophers
stone and eternal life, and the religiously driven 17th century witch
trials in Salem, MA, and other seemingly rational explanations for
irrational behaviors. [Crowcroft, 2005, 19n2]
This would make a nice intro and closing quote for a discussion of computer
science as magic.
[Newell, 1980, 156] says some things about the nature of physical symbol
systems (that is, computers) that have magical overtones. The symbols of
such a system stand for some entity, that is:
An entity X designates an entity Y relative to a process P, if, when P

takes X as input, its behavior depends on Y.
Here, I take it that what Newell means is that Ps behavior really depends on
Y instead of on X. But that seems to be the essence of magic; it is action at a
distance: The process behaves as if inputs, remote from those it in fact has, effect
it. Process P behaves as it does because of a symbolic spell cast at a distance
from P itself.
[Samuelson et al., 1994, 2324n44,n46; 2325n47] cite an NRC report that talks
about user interfaces as illusions.
Unlike physical objects, the virtual objects created in software are

not constrained to obey the laws of physics. . . . In the desktop
metaphor, for example, the electronic version of file folders can
expand, contract, or reorganize their contents on demand, quite unlike
their physical counterparts. [Samuelson et al., 1994, 2334]
If that isnt magic, I dont know what is!

17. I consider computer science, in general, to be the art and science

of representing and processing information and, in particular,
processing information with the logical engines called automatic
digital computers. [Forsythe, 1967a, 3]
(a) both art & sci; in what sense art?he never says. (but see below)
(b) And why info? Possibly because of this earlier (1963) statement:
Machine-held strings of binary digits can simulate a great many
kinds of things, of which numbers are just one kind. For
example, they can simulate automobiles on a freeway, chess
pieces, electrons in a box, musical notes, Russian words, patterns
on a paper, human cells, colors, electrical circuits, and so on.
(Quoted in [Knuth, 972b, 722].)
This is one of the Great Insights. Whats common to all of these,
encoded/abstracted as bit strings, is the information contained in them.
And perhaps because some of these are not sciences, if CS is going
to study them, it must be an art.
(c) Forsythe goes on to say:
[A] central theme of computer science is analogous to a central
theme of engineering sciencenamely, the design of complex
systems to optimize the value of resources. [Forsythe, 1967a, 3].
Note that he doesnt say that CS is engineering.
And note that he speaks of engineering science, suggesting that he thinks
of engineering as a kind of science; certainly, engineering and science per
se are on the same side as opposed to art.
(d) That Forsythe sees CS as more of a science than a branch of engineering
is emphasized when he says, Probably a department of computer science
belongs in the school of letters and sciences, because of it close ties with
departments of mathematics, philosophy, and psychology. But its relations
with engineering departments. . . should be close. [Forsythe, 1967a, 6].
18. In order to teach nontechnical students about computers and computational

thinking, and specialists in other technical fields about how to use computers
as a tool (alongside mathematics, English, statistics), and computer science
specialists about how to lead the future development of the subject,
The first major step. . . is to create a department of computer
science. . . . . . . Without a department, a university may well acquire
a number of computer scientists, but they will be scattered and
relatively ineffective in dealing with computer science as a whole.
[Forsythe, 1967a, 5]
Compare the history of microscopy.
19. Outline of revised version of Newell, Simon, Perlis:
(a) Although they say initially that computer science is the science of
computers, they almost immediately modify that in their Obejction 3,
where they say that by computers they mean living computers
the hardware, their programs or algorithms, and all that goes with them.
Computer science is the study of the phenomena surrounding computers
[Newell et al., 1967, 1374, my emphasis]. And, as they point out in their
reply to Objection 5, Phenomena define the focus of a science, not its
boundaries [Newell et al., 1967, 1374].
(b) They also say that astronomy is the science of stars [Newell et al., 1967,
1373]. Edsger W. Dijkstra once noted that astronomy was not the science
of telescopes CITE; telescopes are used to study the stars.34 Similarly, one
might say that computers are used to study algorithms (and that algorithms,
in turn, are used to study various phenomena in mathematics, other
sciences, etc.). So, Dijkstra would say that computer science should not
be considered a science of the instrument that is used to study algorithms,
any more than astronomy is considered the science of the instrument that
is used to study stars. Newell et al. address this in their Objection 4:
The computer is such a novel and complex instrument that its behavior
is subsumed under no other science [Newell et al., 1967, 1374]. But the
same was once said about microscopes. And we know what happened to
the science of microsopy. DISCUSS MICROSCOPY.
(c) Moreover, that there is no other single science that studies computer
behavior may be irrelevant. Newell et al. deal with this in their Objection 5
FORMER DEAN STORY FROM ABOVE.
(d) At the end, they even allow that the study of computers may also be
an engineering discipline [Newell et al., 1967, 1374]. So, they ultimately
water down their definition to something like this: Computer science is
the science and engineering of computers, algorithms, and other related
phenomena.
20. When the discipline was first getting started, it emerged from various other
disciplines: electrical engineering, physics, mathematics, or even business
[Hamming, 1968, 4]. (In fact, the first academic computer programming course
I took (in Fortran)the only one offered at my university in the late 1960swas
given by the School of Business.)
21. CS questions:
(a) What can be computed? (computability vs. non-computability)
The question What can be automated? is one of the most
inspiring philosophical and practical questions of contemporary
civilization. (George Forsythe, 1968; quoted in [Knuth, 972b,
722723].)
34 Dikjstra 1987, cited in Tedre & Sutinen 2008, said that calling the discipline computer science is like
referring to surgery as knife science. This may be true, but the problem, of course, is that we have no term
corresponding to surgery.
(b) What can be computed efficiently? (complexity theory)

(c) What can be computed practically? (SW engineering considerations) (Cf.
[Hamming, 1968]
(d) What can be computed physically? How can computations be performed
practically? (engineering considerations)
(e) What should be computed? (ethical considerations)
22. [I]n the long run the solution of problems in field X on a computer
should belong to field X, and computer science should concentrate on
finding and explaining the principles [the methodology] of problem
solving [with computers]. [Forsythe, 1967b, 454]
(a) So, the kind of work that we did for QIKR was really philosophy, not CS.
(b) As an example of such methodological research, he cites interactive
systems, in which a man [sic] and a computer are appropriately
coupled. . . for the solution of scientific problems [Forsythe, 1967b, 454].
This is a kind of oracle computation.
(c) Presumably, the idea is for CS proper to abstract away from solving specific
problems in X and to concentrate on general methods of problem solving
using computers. But what does that come down to besides how to write
algorithms? (It does ignore how to make computers.)
23. Computer science as the mathematics of computing:

E.W. Dijkstra [Dijkstra, 1974] argues that programming [i]s a mathematical
activity. He doesnt explicitly say that (all) of computer science is a branch
of mathematics, but I think that he would agree, and that it is quite clear, that
large portions of computer science can be considered to be branches of math
not only programming, but also all of the theory of computability and the theory
of complexity (and no doubt other aspect of computer science, too).
His argument for programming to be considered as math is, roughly, this

[Dijkstra, 1974, 608]:
(a) A discipline Ds assertions are:

i. unusually precise,
ii. general in the sense that they are applicable to a large (often infinite)
class of instances, and
iii. capable of being reasoned about with an unusually high confidence
level
iff D is a mathematical discipline.
(b) Programming satisfies characteristics 23(a)i23(a)iii.
(c) Programming is a mathematical discipline.
He does not argue for premise 23a, taking the only if half (mathematical
disciplines satisfy 23(a)i23(a)iii) as something that most of us can agree
upon, and implicitly justifying the if half (disciplines that satisfy 23(a)i
23(a)iii are mathematical) on the grounds that the objects of mathematical
investigation need not be such usual suspects as numbers, shapes, etc., because
what matters is how objects are studied, not what they are.
And he argues for premise 23b on the grounds that programming clearly requires
extraordinary precision, that programs can accept a wide range of inputs (and
thus are general), and that contemporary program-verification techniques are
based on logical reasoning. We will look into those techniques in Ch. 16; for
now, lets assume that this is true.
Are programs really general in the sense that mathematical assertions are? A
typically general mathematical assertion might be something like this: For any
triangle, the sum of its angles is 180 degrees. In other words, the generality
of mathematical assertions comes from their being universally quantified (for
any x. . . ). Is that the kind of generality that programs exhibit? A program
(as we will see more clearly in Ch. 7) computes a (mathematical) function. So,
insofar as mathematical functions are general, so are programs. Consider a
simple mathematical function: f (x) = 2x. If we universally quantify this, we
get: x[ f (x) = 2x]. This seems to be general in the same way that our assertion
about triangles was, so I think we can agree with Dijkstra about programming
being mathematical.
24. More on Knuth:
(a) Why should we give so much emphasis to teaching how to use
computers, if they are merely valuable tools like (say) electron
microscopes? [Knuth, 1974b, 324]
Nice lead-in to the microscopy section.
His response is that computers are more than this, that [t]he
phenomena surrounding computers are immensely varied and
complex, requiring description and explanation [Knuth, 1974b, 324].
But that can be taken two ways: Algorithms are the phenomena
surrounding computers, and surely they are an appropriate object of
study. Alternatively, because the phenomena that surround computers
are an appropriate object of study, that makes their focus, namely,
computers, also appropriate (unlike microscopes, the construction of
which doesnt illuminate biology directly).
(b) [L]ike electricity, these phenomena [surrounding computers]
belong both to engineering and to science. [Knuth, 1974b, 324]
So maybe when Knuth says that computer science is the study of
algorithms, by study he means both science and engineering. In
what sense does the study of electricity belong both to engineering
and to science? Certainly, the science of physics studies electricity

as a physical phenomenon. And, certainly, electrical engineering
studies electricity from an engineering perspective. But physics and
electrical engineering are typically considered to be separate (albeit
related) disciplines. So, perhaps the same should be said for computer
science (which would study algorithms) and computer engineering
(which would study computers and, perhaps, software engineering).
Cf. [Abrahams, 1987, 472]: Computer science is both a scientific
discipline and an engineering discipline. . . . The boundary [between
the division of computer science into theory (that is, science) and
practice (that is, engineering)] is a fuzzy one.
(c) Knuth is quite clear that he does not view computer science as a branch
of math, or vice versa [Knuth, 1974b, 2], primarily because math
allows for infinite searches and infinite sets, whereas computer science
presumably does not. But there is no reason in principle why one
couldnt write an algorithm to perform such an infinite search. The
algorith could not be executed, perhaps, but that is a physical limitation,
not a theoretical one. I think a better contrast is that mathematics
makes declarative assertions, whereas computer science is concerned with
procedural statements ([Loui, 1987, 177] makes a similar point, quoting
Abelson & Sussman, in arguing that computer science is not mathematics).
And Knuth suggests this in 3. (But this distinction is hard to make
precise. First, consider programming languages such as Prolog, where each
statement can be interpreted either declaratively (p if q) or imperatively
(to make p true, first make q true). Second, any imperative statement can
be converted to a declarative one (consider Euclidean geometry, which was
originally expressed in imperative language: To construct an angle bisector,
do this. . . ; but most modern treatments are declarative: The angles created
by doing this are equal. [Toussaint, 1993]))
25. Is computer science an art or a science?
(a) Only Forsythe suggests that it is, at least in part, an art.

(b) But [Knuth, 1974a] defends his use of art in the title of his multi-volume
classic The Art of Computer Programming [Knuth, 1973]. He does not say
that all of computer science is an art, but he does argue that art can be
applied to, at least, computer programming, not in opposition to science,
but alongside it.
(c) He gives a useful survey of the history of the term, and concludes that
Science is knowledge which we understand so well that we can teach it
to a computer; and if we dont fully understand something, it is an art to
deal with it. . . . [T]he process of going from an art to a science means
that we learn how to automate something [Knuth, 1974a, 668]. This
suggests that being an art is a possibly temporary stage in the development
of our understanding of something, and that our understanding of computer
programming is (or at least was at the time of his writing) not yet fully
scientific. If some subject is never fully scientifically understood, then what
is left over remains an art:
Artificial intelligence has been making significant progress, yet
there is a huge gap between what computers can do in the
foreseeable future and what ordinary people can do. . . . [N]early
everthing we do is still an art. [Knuth, 1974a, 668669].
(d) But he then goes on to give a slightly different meaning to art, one more
in line with the sense it has in the fine arts, namely, as an art form, in an
aesthetic sense [Knuth, 1974a, 670]. In this sense, programming can be
beautiful and it can be like composing poetry or music [Knuth, 1974a,
670]. This not inconsistent with programming being considered a science
or a branch of mathematics (which well discuss in Ch. 16); indeed, many
scientists and mathematicians speak of theories, theorems, demonstrations,
or proofs as being beautiful. One can, presumably, scientifically (or
mathematically) construct a logically verifible program that is ugly (e.g.,
difficult to read or understand) as well as one that is beautiful (e.g., a
pleasure to read or easy to understand).
(e) So, whether computer science is an art is probably not to be set alongside
the question of whether it is a science or a branch of engineering.
26. Our (first order) definition is that computer science is the

study of the theory and practice of programming computers.
[Khalil and Levy, 1978, 31]
A study, not necessarily a science; possibly an art.

Focus on programming; presumably, the theory includes the study
of algorithms, and the practice includes executing the programs on
computers.
27. Computer science as a branch of math:
[Krantz, 1984] offers a sketch of an argument to the effect that computer

science is (just) a branch of mathematics. He doesnt really say that; rather,
his concern is that computer science as an academic discipline is young and
unlikely to last as long as math: Computer Science did not exist twenty-
five years ago [i.e., in 1959]; will it exist twenty-five years from now?
[I.e., in 2009] [Krantz, 1984, 600]. Now, as an academic department in
a university, computer science probably did not exist in 1959, although,
arguably, it did two years later [Knuth, 972b, 722]; it certainly continued
to exist, quite strongly, in 2009. So, if anything, computer science is
becoming much stronger as an academic discipline, not weaker. (But
consider John Hos observations, made around the same time as Krantzs,
as well as the history of microscopy.)
The closest Krantz comes to making an argument is in this passage:

Computer scientists, physicists, and engineers frequently do not realize
that the technical problems with which they struggle on a daiy basis
are mathematics, pure and simple [Krantz, 1984, 599]. As a premise
for an argument to the conclusion that computer science is nothing but
mathematics, this is obviously weak; after all, one could also conclude from
it that physics and engineering are nothing but mathematics, a conclusion
that I doubt Krantz would accept and that I am certain that no physcist or
engineer would accept.
But lets see if we can strengthen it a bit: Suppose, for the sake of
argument, that all of the problems that a discipline D1 is concerned with
come from discipline D2. Does it follow that D1 is nothing but D2?
Heres a (perhaps weak) analogy, which I will strengthen in a moment:
Suppose that you want to express some literary idea; you could write a
short story, a novel, or a poem perhaps. Does it follow that prose fiction and
poetry are the same thing? Probably not; rather, poetry and prose are two
different ways of solving the same problem (in our example, the problem of
expressing a certain literary idea). Similarly, even if both computer science
and mathematics study the same problems, they do so in different ways:
Mathematicians prove (declarative) theorems; computer scientists express
their solutions algorithmically.
So, the argument is probably invalid. Worse, its premise is most likely
false: Surely, there are many problems dealt with in computer science that
are not purely mathematical in nature. Even if we eliminate hardware issues
(the purely computer engineering side of the discipline) from consideration,
surely topics such as operating systems and artificial intelligence are not
purely mathematical in nature.
28. CS as the study of information:
In addition to others noted above, [Denning, 1985, 16] defines it as the

body of knowledge dealing with the design, analysis, implementation,
efficiency, and application of processes that transform information.
Because of the long list of techniques (design [etc.]), I will assume
that, by body of knowledge, he means the neutral study (rather than
science). Moreover, he says that Computer science includes in one
discipline its own theory, experimental method, and engineering. This
contrasts with most physical sciences. . . . . . . I do not think the science and
the engineering can be separated within computer science because of the
fundamental emphasis on efficiency. But I do believe [in contrast with
[Krantz, 1984], John Ho, et al.] that the discipline will endure, because
the fundamental question underlying all of computer sciencewhat can
be automated?will not soon be answered [Denning, 1985, 19]. So, he
seems to be on the side of sui generis. He takes the object to be processes
that transform information. So, one could either take him as seeing
algorithms (processes, procedures) as being the focus, or else information.
Interestingly, he twice cites with great approval Forsythes What can be
automated? [Denning, 1985, 16, 19] without clearly stating its relation to
processes that tranform information.
[Barwise, 1989, 386387] says that computers are best thought of as
information processors, rather than as numerical calculators or as
devices which traffic in formal strings. . . of meaningless symbols.
Although he doesnt argue that computer science is the study of information
or information processing, I think he might side with that characterization.
His principal reason seems to be that the. . . view of computers as
informational engines. . . makes sense of the battle for computational
resources and enables us to think about them so as to make the best
decisions about their acquisition and use. And why is that? One
(abductive) reason is that this view enables us to understand the impact
of computers along the same lines as we understand the impact of books
and printing [and] . . . movable type. . . . [C]omputers are not just super
calculators. They make available a new informational medium. . . just as
with printing. This may seem obvious or not, but Barwise was writing
in 1989, way before the advent of the World Wide Web, Kindles, or
iPads, and his prediction certainly seems to be coming true. But why is
information processing the key, rather than, say, symbol manipulation?
Arguably, information processing is nothing but symbol manipulation.
Arguably, however, it is interpreted symbol manipulation; moreover, not
all symbol manipulation is necessarily information in some sense. So,
perhaps, although computers are nothing but symbol manipulators, it is
as information processors that they will have (are having) an impact.
29. Computer science is a branch of engineering:
Just as with the claim that computer science is a science, there are two
versions of this, a strong and a weak version: The weak version is
that computer science is at least a branch of engineering, or that part
of computer science is a branch of engineering. This allows for the
possibility that computer science might be engineering together with being
something else (such as a science) or that computer science can be divided
into subdisciplines (such as a science subdiscipline and an engineering
subdiscripline).
The strong version is that computer science is nothing but engineering (and
is not also a science).
[Loui, 1987, 175] argues against [Krantz, 1984] that computer science is a
legitimate academic discipline (not another discipline, such as math, and
not something that will disappear or dissolve into other disciplines, like
microscopy).
And, apparently against people like [Newell et al., 1967], he argues that it
is not the study of the use of computers [Loui, 1987, 175]. But that word
use is important. What Loui has in mind is certainly includes the study of
hardware [Loui, 1987, 176]; what he is rejecting is that computer science,
as an academic discipline, is the study of how to use computers (in the way
that driving schools teach how to drive cars).
His definition is this: Computer science is the theory, design, and analysis
of algorithms for processing [that is, for storing transforming, retrieving,
and trasmitting] information, and the implementations of these algorithms
in hardware and in software [Loui, 1987, 176]. He thus seems to side with
[Denning, 1985] and, perhaps, [Knuth, 1974b].
He then goes on to argue that computer science is an engineering discipline
[Loui, 1987, 176]:
(a) Engineering. . .
i. is concerned with what can exist (as opposed to what does exist.
ii. has a scientific basis
iii. applies principles in design
iv. analyzes trade-offs
v. has heuristics and techniques
(b) Computer science has all the[se] significant attributes of engineering
(c) , computer science is a branch of engineering.
We will delay investigation of the first premise to Ch. 5.
His justification that computer science is not concerned with what exists is
related to the claim that others have made CITE that computer science is
not a natural science, but a science of human-made artifacts. There are
two possible objections to this: First, insofar as procedures are natural
entities, computer scienceas the study of procedurescan be considered
a natural science. Second, insofar as some artifactssuch as birds nests,
beehives, etc.are natural entities, studies of artifacts can be considered to
be scientific.
The scientific basis of traditional engineering disciplines suchas

mechanical engineering and electrical engineering is physics, but the
scientific basis of computer science is, according to Loui, mathematics.
He thus considers computer science to be a new kind of engineering.
Mathematical engineering, perhaps? [Abrahams, 1987, 472] explicitly
makes this claim, and [Halpern et al., 2001] can be read as making a case
for computer science as being based more on logic than on mathematics,
soif it is a kind of engineeringperhaps it is logical engineering?. (Well
return to computer science as mathematical engineering below.) This
assumes, of course, that you are willing to consider mathematics to be
a natural science, or else that science is not limited to studying natural
entities. But in the latter case, why worry about whether computer science
is concerned with what can, rather that what does, exist?
According to Loui, engineers apply the principles of the scientific base of

their engineering discipline to design a product. [A] computer specialist
applies the principles of computation to design a digital system or a
program [Loui, 1987, 176]. Hence, presumably, computer science is a
branch of engineering. But not all computer scientists (or specialists)
design systems or programs; some do purely theoretical work. And,
in any case, if the scientific basis of computer science is mathematics,
then why does Loui say that computer specialists apply apply the
priniciples of computation? I would have expected him to say the
principles of mathematics. Perhaps he sees computation as being
a branch of mathematics (but thats inconsistent with his objections to
Krantz). Or perhaps he doesnt think that the abstract mathematical
theory of computation is part of computer science (but that seems highly
unlikely, especially in view of his definition of computer science as
including the theory and analysis of algorithms). Its almost as if he sees
computer engineering as standing to computer science in the same way
that mechanical or electrical engineering stand to physics. But then it is not
computer science that is a branch of engineering.
To implement algorithms efficiently, the designer of a computer system

must continually evaluate trade-offs between resources such as time vs.
space, etc. [Loui, 1987, 177]. This is true, but doesnt support his argument
as well as it might. For one thing, not only system designers evaluate such
trade-offs; so do theoreticianswitness the abstract mathematical theory
of complexity. And, as noted above, not all computer scientists design
such systems. So, at most, those who do are doing a kind of engineering.
As for heuristics, Loui seems to have in mind rough-and-ready rules of

thumb rather than formally precise theories. Insofar as engineers rely
on such heuristics (which we will discuss in Ch. 5 when discuss Koens
definition), and insofar as some computer scientists also do, then those
computer scientists are doing something that engineers also do. But so do
many other people: Writers surely rely on such heuristics (write simply
and clearly); does that make them engineers? This is probably his weakest
argument.
So, despite Louis belief that he is arguing for the strong view of computer
science as nothing but engineering, I think hes better interpreted as arguing
for the weak view, namely, that there is an engineering component to
the discipline. Indeed, towards the end of his essay, after approvingly
citing [Denning, 1985]s definition, he says: It is impossible to define
a reasonable boundary between the disciplines of computer science and
computer engineering. They are the same discipline [Loui, 1987, 178].
But that seems to contradict the title of his essay!
30. Computer science is the study of complexity [Ceruzzi, 1989, 268]
Ceruzzi ascribes this to Jerome Wiesner [Wiesner, 1958]. But all

Wiesner says is that Information processing systems are but one facet
of. . . communication sciences. . . that is, the study of. . . the problems of
organized complexity (quoted in [Ceruzzi, 1989, 269]). But even if
computer science is part of a larger discipline that studies complexity, it
doesnt follow that computer science itself is the study of complexity.
According to Ceruzzi, Edsgar Dijkstra also held this view: programming,
when stripped of all its circumstantial irrelevancies, boils down to no
more and no less than very effective thinking so as to avoid unmastered
complexity [Dijkstra, 1975, 4, p. 3]. But, again, at most this makes the
claim that part of computer science is the study of complexity. And really
all that it says is that programming is (the best) way to avoid complexity,
not that it studies it. Whats missing, in any case, is a premise to the effect
that computer science is the study of programming, but Dijkstra doesnt
say that, either in [Dijkstra, 1975] or in [Dijkstra, 1976], the document
that Ceruzzi says contains that premise. (And only [Khalil and Levy, 1978]
makes that claim.)
And [Denning et al., 1989, 11] point out that viewing computer science
[as] the study of abstraction and the mastering of complexity. . . also
applies to physics, mathematics, or philosophy. But, for a contrasting
view of abstraction as the key to computing, see [Kramer, 2007].
[Gal-Ezer and Harel, 1998, 80] also suggest complexity as a unifying
theme, if not the basic subject matter, of computer science: a beneficial
way to view CS is by the kinds of complexity it deals with[:] computational
complexity; system, or behavioral, complexity; and cognitive complexity.
31. Is computer science coextensive with (if not identical to) electrical (more
precisely, electronic) engineering?
[Ceruzzi, 1989] doesnt explicitly say this, but he comes close: First,
Electronics emerged as the technology of choice [over those that
were used in mechanical calculators or even early electric-relay-based
computers] for implementing the concept of a computing machine. . . .
. . . This activity led to the study of computing independently of the
technology out of which computers were built. In other words, it led to
the creation of a new science: Computer Science [Ceruzzi, 1989, 257].
Second, As computer science matured, it repaid its debt to electronics
by offering that engineering discipline a body of theory which served to
unify it above the level of the physics of the devices themselves. In short,
computer science provided electrical engineering a pardigm, which I call
the digital approach, which came to define the daily activities of electrical
engineers in circuits and systems design [Ceruzzi, 1989, 258].
One problem with trying to conclude from this that computer

science is (nothing but) electrical engineering is that there are now
other technologies that are beginning to come into use: quantum
computing [Grover, 1999, Aaronson, 2008, Monroe and Wineland, 2008,
Aaronson, 2011, Hayes, 2014b] and DNA computing [Adleman, 1998,
Shapiro and Benenson, 2006, Qian and Winfree, 2011]. Assuming that
those methods achieve some success, then it becomes clear how
computer science goes beyond any particular implementation technique or
technology, and becomes a more abstract science (or study, or whatever)
in its own right. And Ceruzzi himself declares, The two did not become
synonymous [Ceruzzi, 1989, 273].
32. [Licklider and Taylor, 1968] fairly accurately predicted what computers would
look like, and how they would be used, today (nearly 50 years in the future).
What they were writing about is clearly part of computer science, yet equally
clearly not (directly) part of the abstract, mathematical theory of computation.
This strongly suggests that it would be wrong to treat computer science as
being primarily about algorithms or primarily about computers. It is about both.
Well see this more clearly when we trace the parallel histories of computers
(as calculating machines) and computing (as it evolved from the search for a
foundation for mathematics).
33. HARTMANIS 1993 SHOULD PROBABLY BE
REQUIRED/VERY STRONGLY RECOMMENDED, BUT RE-READ HIS 3
PAPERS (1992 WITH LIN, 1993, AND 1995 TO SEE WHICH SHOULD BE
REQUIRED, WHICH V.S.R, ETC.
I quoted this earlier (3.3.2):
[Computer science differs from the known sciences so deeply

that it has to be viewed as a new species among the sciences.
[Hartmanis, 1993, 1].
Lets consider this claim carefully. First, Hartmanis comes down on the side of
computer science being a science: It is a new species among the sciences. (A
tiger is a different species from a chimpanzee among the animals, but they are
both animals.)
Second, what does it mean to be a new species of science? Both chimps and
tigers are species of animals, and both lions and tigers are species within the
genus Panthera. Is the relation of computer science to other sciences more like
the relation of chimps to tigers (pretty distant) or lions to tigers (pretty close)? A
clue comes in Hartmaniss next sentence:
This view is justified by observing that theory and experiments in
computer science play a different role and do not follow the classic
pattern in physical sciences. [Hartmanis, 1993, 1]
This strongly suggests that computer science is not a physical science (such as
physics or biology), and Hartmanis confirms this suggestion on p. 5: computer
science, though not a physical science, is indeed a science (my italics; cf.
[Hartmanis, 1993, 6], [Hartmanis, 1995a, 11]). The non-physical sciences are
typically taken to include at least the social sciences (such as psychology) and
mathematics. So, it would seem that the relation of computer science to other
sciences is more like that of chimps to tigers: distantly related species of the
same, high-level genus. And, moreover, it would seem to put computer science
either in the same camp as (either) the social sciences or mathematics, or else in
a brand-new camp of its own, i.e., sui generis.
And immediately after his claim that he will not define computer science (see
the motto to 3.4), he gives this characterization:
At the same time [that is, despite his immediately preceding refusal
to define it], it is clear that the objects of study in computer science
are information and the machines and systems which process and
transmit information. From this alone, we can see that computer
science is concerned with the abstract subject of information, which
gains reality only when it has a physical representation, and the
man-made devices which process the representations of information.
The goal of computer science is to endow these information
processing devices with as much intelligent behavior as possible.
[Hartmanis, 1993, 5] (cf. [Hartmanis, 1995a, 10])
Although it may be clear to Hartmanis that information, an abstract subject,
is (one of) the object[s] of study in computer science, he does not share his
reasons for that clarity. Since, as we have seen, others seem to disagree that
computer science is the study of information (others have said that it is the
study of computers or the study of algorithms, for instance), it seems a bit
unfair for Hartmanis not to defend his view. But he cashes out this promissory
note in [Hartmanis, 1995a, 10], where he says that what sets it [computer
science] apart from the other sciences is that it studies processes [such as
information processing] that are not directly governed by physical laws. And
why are they not so governed? Because information and its transmission are
abstract entities [Hartmanis, 1995a, 8]. This makes computer science sound
very much like mathematics (which is not unreasonable, given that it was this
aspect of computer science that led Hartmanis to his ground-breaking work
on computational complexity, an almost purely mathematical area of computer
science)
But its not just information that is the object of study; its also information-
processing machines, that is, computers. Computers, however, dont deal
directly with information, because information is abstract, that is, non-physical.
For one thing, this suggests that, insofar as computer science is a new species
of non-physical science, it is not a species of social science: Despite its name,
the social sciences seem deal with pretty physical things: societies, people,
speech, etc. (This, by the way, is controversial. After all, one of the main
problems of philosophy is the problem of the relation of the mind (which seems
to be non-physical and is studied by the social science of psychology) to the
brain (which is clearly physical and is studied by the natural sciences of biology
and neuroscience). And philosophers such as John Searle [Searle, 1995] have
investigated the metaphysical nature of social institutions such as money, which
seem to be neither purely abstract (many people cash a weekly paycheck and
get real money) nor purely natural or physical (money wouldnt exist if people
didnt exist). So, if computer science is a science, but is neither physical nor
social, then perhaps it is more like mathematics, which is sometimes called a
formal science (http://en.wikipedia.org/wiki/Formal science).
For another thing, to say that computers dont deal directly with information,
but only with representations of information suggests computer science has a
split personality: Part of it deals directly with something abstract (information),
and part of it deals with directly with something real but that is (merely?) a
representation of that abstraction (hence dealing indirectly with information).
Such real (physical?) representations are called implementations; we will look
at that notion in more detail in Ch. 14 and at the relation of computers to the real
world in Ch. 17.
Finally, although Hartmaniss description of the goal of computer science sounds

like he is talking about AI (and he might very well be), another way to think about
that goal is this: Historically, we have endowed calculating machines with the
ability to do both simple and complex mathematics. What other abilities can we
give to such machines? Or, phrased a bit differently, what can be automated
what can be computed? GIVE EITHER A FORWARD OR A BACKWARD
CITATION TO FORSYTHE.
HARTMANIS HAS INTERESTING THINGS TO SAY

ABOUT COMPLEXITY AND LEVELS OF ABSTRACTION ON PP.56. IF I
DISCUSS THIS ELSEWHERE, I SHOULD CITE HARTMANIS.
[Computer science is concentrating more on the how than the what,

which is more the focal point of physical sciences. In general the how
is associated with engineering, but computer science is not a subfield
of engineering. [Hartmanis, 1993, 8]
Here we have another reason why computer science is not a physical science
(and probably also why it is not a social science). But there are two ways to
understand how: Algorithms are the prime formal entity that codify how to
accomplish some goal. But, as Hartmanis quickly notes, engineering is the prime
discipline that is concerned with how to do things, how to build things (which
well look at in more detail in Ch. 5). The first kind of how is mathematical
and abstract (indeed, it is computational!); the second is more physical. One

way to see this as consistent with Hartmaniss description of the objects of study
of computer science is to say that, insofar as computer science studies abstract
information, it is concerned with how to process it (that is, it is concerned with
algorithms), and, insofar as computer science studies computers, it is concerned
with how to process representations (or implementations) of information (that is,
it is concerned with the physical devices).
But that latter task would seem to be part of engineering (perhaps,

historically, electrical engineering; perhaps, in the future, quantum-mechanical
or bioinformatic engineering; certainly computer engineering!). So why does he
say that computer science is not a subfield of engineering? In fact, he seems
to regret this strong statement, for he next says that the engineering in our field
has different characterizations than the more classical practice of engineering
[Hartmanis, 1993, 8]: So, computer science certainly overlaps engineering, but,
just as he claims that computer science is a new species of science, he also
claims that it is a new form of engineering [Hartmanis, 1993, 9]. In fact, he
calls it [s]omewhat facetiously. . . the engineering of mathematics; but he also
says that we should not try to draw a sharp line between computer science and
engineering [Hartmanis, 1993, 9].
To sum up so far, Hartmanis views computer science as a new species both

of science and of engineering. This is due, in part, to his view that it has
two different objects of study: an abstraction (information) as well as its
implementations. But isnt it also reasonable to think that, perhaps, there are
really two different (albeit new) disciplines, namely, a new kind of science and
a new kind of engineering? If so, do they interact in some way more deeply and
integratively than, say, chemistry and chemical engineering, so that it makes
sense to say that they are really a single discipline? Hartmanis suggests
two examples that show a two-way interaction between these two (halves of
the same?) discipline: Alan Turings interest in the mathematical nature of
computation led to his development of real computers and AI; and John von
Neumanns interest in building computers led to his theoretical development of
the structure of computer architecture [Hartmanis, 1993, 10].
Hartmanis explicitly says that computer science is a science and is not

engineering, but his comments imply that it is both. I dont think he can have it
both ways.
HARTMANIS ALSO HAS A NICE CHARACTERIZATION OF THE DUAL

HISTORY OF CS AS BOTH A CULMINATION OF THE HISTORY OF THE
FOUNDATIONS OF MATH AND OF THE HISTORY OF CALCULATING
DEVICESSEE P.9
34. One of the defining characteristics of computer science is the
immense difference in scale of the phenomena computer science

deals with. From the individual bits of programs and data in the
computers to billions of operations per second to the highly complex
operating systems and various languages in which the problems are
described, the scale changes through many orders of magnitude.
[Hartmanis, 1993, 5] (cf. [Hartmanis, 1995a, 10])
[Brooks, 1996, 62, col. 2] echoes this when he says that It is this new world
of complexity that is our peculiar domain. In what way is this a defining
characteristic? Lets say, following Hartmanis, that computer science is the
science of information processing and representation. Is it also the science
of scale differences? Not quite; rather, it seems that scale differences are a
feature of information and its processing and representation. So, insofar as
there might be some other science that also studies information (just as biology,
chemistry, and physics all study physical phenomena), then that science would
also, ipso facto, study such scale differences. Indeed, Hartmanis himself, in
his replies to commentaries on this paper [Hartmanis, 1995b, 60], says that
physics also deals with scale differnences. However, it is interesting to note that
Herbert Simon [Simon, 1962], [Simon, 1996b] also focused on the study of such
complexity and the usefulness of abstraction in studying it (cf. [Hartmanis, 1993,
6], [Hartmanis, 1995a, 11]).
35. What is information? Answering this question is important, but would take
us too far afield. Lots of work has been done on the nature of information
and on the philosophy of information. See, especially, the writings of Floridi,
cited in 1.4, as well as [Piccinini and Scarantino, 2011] and the commentary on
[Hartmanis, 1995a] by Denning (1995), cited below in 3.14.
36. What Do Computer Scientists Do?
One way to answer this new question is to look at some of the advertising
mentioned in 3.3.1.4. Here is a more or less random selection; some of
these definitions of computer science were originally found on departmental
websites or from introductory course materials:
(a) Because the word compute comes from a Latin expression meaning to
reckon together, perhaps computer science is the science of how to
calculate [Jacob, 1999]. As we will see, this is not entirely unreasonable,
but a more plausible characterization is that computer science is the
science of processing information by computers [Jacob, 1999].
But this raises further questions: What is a science? What is

information? What does it mean to process information? And what
is a computer?
(b) According to [Roberts, 2006, 5], computer science is the science of
problem solving in which the solutions happen to involve a
computer.
As Alexander the Great knew when he sliced through the Gordian knot
instead of untying it, there are at least two ways to solve problems.35
Presumably, a science of problem solvingbased on theories of
algorithms, data structures, and complexityis to be contrasted with mere
hacking in the sense of playing around. But one might also wonder
how solving problems might be related to calculating or processing
information, and how the computer is supposed to help.
37. Hartmanis calls computer science a new species among the sciences
[[Hartmanis, 1995a]; but see also [Hartmanis, 1993, 1]]. . . . I have
argued that it would be more accurate to call computer science a new
species of engineering [Loui, 1987]. [Loui, 1995, 31]
But his argument here [Loui, 1995, 31] is interestingly different from his earlier
one:
(a) The goal of engineering is the production of useful things

(b) [C]omputer science is concerned with producing useful things
(c) Computer science is a species of engineering.
This is interesting, because it is invalid! Just because two things share a common
property, it does not follow that one is subsumed under the other. For one
thing, Loui could equally well have concluded that engineering was a species
of computer science. But, more importantly, just because both cats and dogs are
household pets, it doesnt follow that either is a species of the other.
But there are better arguments for considering computer science to be a branch
of engineering, as we have seen (and will look at more closely in Ch. 5. If it
is engineering, what kind of new engineering is it? Heres Louis argument
[Loui, 1995, 31] for that:
(a) [E]ngineering disciplines have a scientific basis.

(b) The scientific fundamentals of computer science. . . are rooted. . . in
mathematics.
(c) Computer science is. . . a new kind of engineering.
This argument is interesting for better reasons than the first one. First, there are
a few missing premises:
A Mathematics is a branch of science.

B No other branch of engineering have mathematics as its basis.
Recall that we can assume that computer science is a kind of engineering (if not
from Louis invalid argument, then, at least, from some of the better ones given
by others). So, from that and 37a, we can infer that computer science (an an
35 http://en.wikipedia.org/wiki/Gordian Knot, accessed 20 September 2011.
engineering discipline) must have a scientific basis. We need premise 37A so that
we can infer that the basis of computer science, which, by 37b is mathematics,
is indeed a scientific one. (Alternatively, one can view 37b as the conjunction of
The fundamentals of computer science are rooted in mathematics and 37A.)
Then, from 37B, we can infer that computer science must differ from all other
branches of engineering. It is, thus, mathematical engineering (as others have
noted).
But this reminds me of the dialogue between [Newell et al., 1967] and
[Knuth, 1974b]. Both Loui and Hartmanis agree that computer science is a new
kind of something or other; each claims that the scientific/mathematical aspects
of it are central; and each claims that the engineering/machinery aspects of it
are also central. But one calls it science, while the other calls it engineering.
Again, it seems to be a matter of point of view.
38. Put this somewhere in 3.4:
Lets remember that there is only one naturethe division into

science and engineering, and subdivision into physics, chemistry,
civil and electrical, is a human imposition, not a natural one. Indeed,
the division is a human failure; it reflects our limited capacity to
comprehend the whole. That failure impedes our progress; it builds
walls just where the most interesting nuggets of knowledge may lie.
[Wulf, 1995, 56]
His last two sentences suggest that, rather that consider computer science to be
either science or else engineering (or even sui generis), it should be thought of
as an interdisciplinary discipline. As such, its goal would be to bring to bear
what is special to itnamely, algorithms and computersto problems in other
disciplines. This is summarized in his final paragraph:
I reject the title question [Are We Scientist or Engineers?].

. . . [Computer science] spans a multidimensional spectrum from deep
and elegant mathematics to crafty programming, from abstraction to
solder joints, form deep truth to elusive human factors, from scholars
motivated purely by the desire for knowledge or practitioners making
my everyday life better. It embraces the ethos of the scholar as
well as that of the professional. To answer the question would be
to exclude some portion of this spectrum, and I would be poorer for
that. [Wulf, 1995, 57]
That would be a great quote to end either this chapter or the chapter on
engineering (before going on to other questions).
39. More comments on [Brooks, 1996]; maybe some or all of this should go in Ch. 5?
Brookss argument:
(a) [A] science is concerned with the discovery of facts and laws.
[Brooks, 1996, 61, col. 2]
(b) [T]he scientist build in order to study; the engineer studies in order to
build. [Brooks, 1996, 62, col. 1]
(c) The purpose of engineering is to build things.
(d) Computer scientists are concerned with making things, be they computers,
algorithms, or software systems. [Brooks, 1996, 62, col. 1]
(e) the discipline we call computer science is in fact not a science but a
synthetic, an engineering, discipline. [Brooks, 1996, 62, col. 1]
As I say earlier, whether or not Brookss notion of what science is is accurate

will be our focus in Chapter 4. So, lets assume the truth of the first premise for
the sake of the argument.
The point of the second premise is this: If a scientists goal is to discover facts
and laws, that is, to study, then anything built by the scientist is only built for
that ultimate purpose; but building is the goal of engineering (and any studying,
or discovery of facts and laws) that an engineer does along the way to building
whatever is being built is merely done for that ultimate purpose. Put another way,
for science, building is a side-effect of studying; for engineering, studying is a
side-effect of building. (Does this remind you of the algorithms-vs.-computers
dispute earlier? Both scientist and engineers, according to Brooks, build and
study, but each focuses more on one than the other.)
To sum up, the point of the second premise is to support the next premise,
which defines engineering as a discipline whose goal is to build things, that is, a
syntheticas opposed to an analyticdiscipline.36 Again, whether or not
Brooks notion of what engineering is is accurate will be our focus in Ch. 5.
So, lets assume the truth of the second and third premises for the sake of the
argument.
Clearly, if the third premise is true, then the conclusion will follow validly (or,
at least, it will follow that computer scientists belong on the engineering side of
the science-engineering, or studying-building, spectrum). So, is it the case that
computer scientists are (only? principally?) concerned with building or making
things? And, if so, what kind of things?
Interestingly, Brooks seems to suggest that computer scientists dont build

computers, even if thats what he says in the conclusion of his argument. Heres
why: He says that Even when we build a computer the computer scientist
designs only the abstract propertiesits architecture and implementation.
36 To analyze is to pull apart; to synthesize is to put together. We speak of engineering as concerned with
synthesis, while science is concerned with analysis. [Simon, 1996b, 4]

Electrical, mechanical, and refrigeration engineers design the realization

[Brooks, 1996, 62, col. 1]. I think this passage is a bit confused or sloppy, but it
makes an interesting point: Brooks seems to be saying that computer scientists
only design abstractions, and other (real?) engineerings implement them in
reality. This is reminiscent of the distinction between the specifications of an
algorithm and its detailed implementation in a computer program (well look
into this in Ch. 10). Where Hartmanis and others have suggested that computer
science is mathematical engineering, Brooks (following [Zemanek, 1971]) calls
it the engineering of abstract objects: If engineering is a discipline that builds,
then what computer science qua engineering builds is implemented abstractions
(see Ch. 14 for further discussion).
Moreover, what computer scientists build, unlike what other engineers build, are
not things for direct human benefit (at least, not in 1977 when he first wrote these
words and when very few people other than scientists, engineers, and business
people had access to (gigantic) computing machinescertainly, there were no
personal computers, laptops, tablets, or smartphones) but, rather, things that in
turn can be used to build such directly beneficial things. Put more simply, his
answer to the question What is a computer? seems to be: A computer is a tool
(and a computer scientist, who makes such tools, is a toolsmith [Brooks, 1996,
62, col. 1].
But much of what he says against computer science being considered a science
smacks of a different battle, one between science and engineering, with scientists
belittling engineers. Brooks takes the opposite position: as we honor the
more mathematical, abstract, and scientific parts of our subject more, and the
practical parts less, we misdirect young and brilliant minds away from a body of
challenging and important problems that are our peculiar domain, depriving the
problems of the powerful attacks they deserve [Brooks, 1996, 62, col. 2].
40. Notes on [Simon, 1996b]:
Argument that computer science is a science of the artificial:

(a) A natural science is a body of knowledge about some class of
things. . . in the world [Simon, 1996b, 1] (Presumably, a natural
science of X is a body of knowledge about Xs in the world. Note
that he does not say that the Xs need to be natural!)
(b) The central task of a natural science is. . . to show that complexity,
correctly viewed, is only a mask for simplicity; to find pattern hidden
in apparent chaos [Simon, 1996b, 1]
(c) The world we live in today is much more a[n]. . . artificial world than
it is a natural world. Almost every element in our environment shows
evidence of human artifice. [Simon, 1996b, 2] (Again, this allows
artifacts to be among the values of X. His justification for this premise
consists of examples: the use of artificial heating and air-conditioning
to maintain temperature, [a] forest may be a phenomenon of nature;

a farm certainly is not. . . . A plowed field is no more part of nature
than an asphalted streetand no less, etc. [Simon, 1996b, 23, my
emphasis]. All artifacts, he also notes, are subject to natural laws
(gravity, etc.) [Simon, 1996b, 2].
Now, Simon doesnt, in this chapter, explicitly draw the conclusion
that there can be sciences of artifacts or, more to the point,
thatbecause computers are symbol systems [Simon, 1996b, 17ff]
(cf. [Newell and Simon, 1976]) and symbols are strings of artifacts
[Simon, 1996b, 2]computer science is an artificial science of
computers. For one thing, thats what his whole book is designed to
argue. But he doesnt have to draw that conclusion explicitly: If it doesnt
follow from the first premise that computer science can even be considered
a natural science, it follows from these premises that any artifacts that
impinge upon, or are produced by, nature or natural objects can be studied
scientifically (in the manner of the second premise).
Its almost as if he really wants to say that artificial sciences are natural
sciences.
41. Just as the scientific basis of electrical engineering is primarily
physics, the scientific basis of software engineering is primarily
computer science. This paper contrasts an education in a science with
an education in an engineering discipline based on the same science.
[Parnas, 1998, 2].
The first sentence, when set alongside the claims that computer science is
mathematical engineering, suggest that software engineering is mathematical
engineering engineering! So, both claims probably cant be true, and Parnass
second sentence shows why: He believes that computer science is a science, not
an engineering discipline. Why?
Part of the reason concerns his definition of engineering: Engineers are

professionals whose education prepares them to use mathematics, science, and
the technology of the day, to build products that are important to the safety and
well-being of the public [Parnas, 1998, 2]. As usual, we will defer analysis of
this to Ch. 5. But note, first, the emphasis on education and, second, the idea that
engineers use science to build things.
To complete his argument, he needs a premise that states that computer science
is not engineering, that is, that computer science education doesnt prepare
computer scientists to use computer science to build things or just that computer
scientists dont build things. (That leaves open the possibility that computer
science might be a branch of math or a technology of the day, but its pretty
clear from the first quote that he thinks that it is a science.) This missing premise
is the gist of his entire article. But at least one part of his argument is this: Proper
training in software engineering (designing, building, testing, and maintaining
software produces [Parnas, 1998, 2]) requires more than a course or two offered
in a computer science curriculum; rather, it requires an accredited professional
programme. . . modelled on programmes in traditional engineering disciplines
[Parnas, 1998, 2].
But we still havent got a clear statement as to why he thinks that computer
science is a science and is not engineering. Well, maybe the latter: Its not
engineering, because there is no rigid accreditation process. . . [hence, no] well
documented core body of knowledge. . . for computer science [Parnas, 1998,
2]. Such accreditation might be necessary, but is surely not sufficient: One might
force such a core body of knowledge and such an accreditation process on, say,
physics, but that wouldnt make physics an engineering discipline.
Some clarity arises here: It is clear that two programmes are needed [for
example, both physics and electrical engineering, or both computer science and
software engineering], not because there are two areas of science involved [for
example, physics and EE], but because there are two very different career paths.
One career path is that of graduates who will be designing products for others
to use. The other career path is that of graduates who will be studying the
phenomena that interest both groups and extending our knowledge in this area
[Parnas, 1998, 3, my emphasis]. So: scientists study phenomena and extend
knowledge; engineers design products. So: computer science studies phenomena
and extends knowledge; software engineers design software products. (What
phenomena does Parnas think that computer scientists study?) The distinction
between science and engineering, for Parnas, is that between learning and
building [Parnas, 1998, 4].
42. Yet another definition (in 3 different versions):
(a) Computer science [is] the science of solving problems with the aid of a
computer [Roberts, 1996]
(b) For the most part, computer science is best thought of as the science of
problem solving in which the solutions happen to involve a computer.
[Roberts, 2006, 5]
(c) Computer Science is the science of using computers to solve problems.
[Computer Science, 2003]
Because these definitions dont limit the kind of problems being solved,
they have the advantage of illuminating the generality and interdisciplinarity
of computer science. And, because they implicitly include the software
(algorithmic) side of computingafter all, you cant use a computer to solve
a problem unless it has been appropriately programmedthey nicely merge the
computer-vs.-algorithm aspects of the possible definitions. (They could just as
easily have said something even more neutral: Computer science is the science of
solving problems computationally, or algorithmicallyafter all, you cant solve
a problem that way without executing its algorithmic solution on a computer.)

But can there really be a science of problem solving?
43. An interesting definition from an invalid argument:
Computer science is philosophy. Logic is the foundation of
philosophy. Its also the foundation of Computer Science.
[Rupp, 2003]
The first sentence is the title of Rupps essay. The next two sentences are the first
sentences of it, and the only support that he offers for his definition. But this is
an invalid argument: Just because two disciplines share a common foundation, it
does not follow that one of them is the other (and which one?) or that they are
identical.
44. [Hammond, 2003], quoting, possibly, Michael R. Fellows (see
http://en.wikiquote.org/wiki/Computer science), says that computer science is
not about computers, because science is not about tools. It is about how we use
them, and what we find out when we do and because Theoretical Computer
Science doesnt even use computers, just pencil and paper. In other words, it is
wrong to define a subject by its tools (see microscopy, above), and it is wrong to
define a subject in such a way that it omits an important part of it.
Rather, she says, computer science is the study of algorithms, including

their. . . Hardware Realizations. In other words, although the surrounding
phenomena of algorithms includes the computers that execute them, its the
algorithms that are the real focus.
Even if this is true, it can be argued that a new tool can open up a new science or,
at least, a new scientific paradigm (in the sense of [Kuhn, 1962]): Paradigm
shifts have often been preceded by a technological or conceptual invention
that gave us a novel ability to see things that oculd not be seen before
[Mertens, 2004, 196], quoting [Robertson, 2003]. Although computer science
may not be about computers any more than astronomy is about telescopes, The
computer is to the naked mind what the telescope is to the naked eye, and it may
well be that future generations will consider all precomputer science to be as
primitive as pretelescopic astronomy [Mertens, 2004, 196].
45. Another science of an artifact: bicycle science
[Wilson and Papadopoulos, 2004]. But its really not clear if this is a science
or a branch of engineering!
46. Computational Intelligence is the manifest destiny of computer

science, the goal, the destination, the final frontier.
[Feigenbaum, 2003, 39]
This isnt exactly a definition of computer science, but it could be turned into
one: Computer science is the study of (choose one): (a) how to make computers
(at least) as intelligent as humans (how to get computers to do what humans

can do); (b) how to understand (human) cognition computationally.
The history of computers supports this: It is a history that began with how to
get machines to do some human thinking (certain mathematical calculations, in
particular), then more and more.
Turings motivation for the TM also supports this: The TM, as a model of
computation, was motivated by how humans compute.
47. On Euclid (for this chapter, or elsewhere):
Euclids geometry was procedural in nature; his proofs can be viewed as
programs that construct certain geometrical figures using only compass and
straightedge, much as computer programs construct (mathematical) objets using
only recursive functions. [Toussaint, 1993] (see also [Goodman, 1987, 4])
48. On microscopy:
Computation is itself a methodology. [Denning, 2013, 37], citing unnamed
others, uses the phrase a new method of thought and discovery in science.
49. On computer science as a natural science:
[Denning, 2013, 37] argues that, insofar as computer science deals with
information processes, it is a natural science, because many information
processes are natural processes, e.g., in biology. (This supports Shapiros
argument.) [Easton, 2006] makes a similar argument.
Denning 2007 offers support for Shapiros argument by citing examples of the
discovery of information processes in the deep structures of many fields:
biology, quantum physics, economics, management science, and even the arts
and humanities. He concludes that computing is now a natural science, not (or
no longer?) a science of the artificial.
An earlier statement by him to this effect is: Computer science. . . is the

science of information processes and their interactions with the world, together
with the later comment that There are many natural information processes
[Denning, 2005, 27, my emphasis].
50. Rather than wonder whether computer science is a science or not, or a branch of
engineering or not, [Lindell, 2001] suggests that it is a liberal art. Today, when
one thinks of the liberal arts, one tends to think of the humanities rather than
the sciences, but, classically, there were seven liberal arts: the linguistic liberal
arts (grammar, rhetoric, and logic) and the mathematical liberal arts (arithmetic,
music, geometry, and astronomy). (For more on this, see the Wikipedia article,
Liberal Arts Education at https://en.wikipedia.org/wiki/Liberal arts education
(accessed 1 July 2013).) Thought of this way, it becomes more reasonable to
consider computer science as a modern version of these.
51. Minsky has also suggested that computer science is concerned with ways to
describe complicated processes (SCS notes from Minskys talk at UB, personal
communication, 25 July 2005).
52. Pedagogically, computer programming has the same relation to
studying computer science as playing an instrument does to studying
music or painting does to studying art. [Tucker et al., 2003, V]
53. Besides trying to say what CS is (a science/study/engineering . . . of
algorithms/computers/information . . . ), another approach is to say that it is a
way of thinking, that algorithmic thinking (about anything) is what makes
CS unique. See [Anthes, 2006], especially the comments by Bernard Chazelle.
54. On the def of procedure (such that computer science might be defined, a la
Minsky and Shapiro, as a science of (computational) procedures:
As [Bringsjord, 2006] argues, talk of procedures can be replaced by talk of
first-order logic, and proofs and interpretations; consider, for example, the
dual interpretations of statements of Prolog as either procedural or declarative
statements, or Lisp programs, which appear to be merely definitions of recursive
functions. So, we need to be clearer about what we mean by procedure (as
well as phrases like computational thinking or algorithmic thinking). This is
a philosophical issue worthy of discussion (and well return to it in Ch. 7).
On computational thinking:
[Wing, 2006]s notion of computational thinking may offer a compromise:
This term avoids a procedural-declarative controversy, by including both
concepts, as well as others. She defines computer science as the study of
computationwhat can be computed and how to compute it. This definition
is nice, too, because the first conjunct clearly includes the theory of computation
and complexity theory (can can include can in principle as well as can
efficiently), and the second conjunct can be interpreted to include both software
programming as well as hardware engineering. Study is nice, too: It avoids
the science-engineering controversy. Computational thinking includes problem
solving, recursive thinking, abstraction, and more.
Denning 2009 finds fault with the notion, primarily on the grounds that it is too
narrow: Computation is present in nature even when scientist are not observing
it or thinking about it. Computation is more fundamental than computational
thinking. For this reason alone, computational thinking seems like an inadquate
characterization of computer science (p. 30). Note that, by computation,
Denning means TM-computation. For his arguments about why it is present
in nature, see Denning 2007.
55. FOR SOME OF THE MORE MINOR POSITIONS, PERHAPS HAVE A
SECTION AT THE END OF (EACH?) CHAPTER PRESENTING THE
POSITIONS AND RAISING QUESTIONS FOR DISCUSSION ABOUT
THEM. The next two items are prime candidates for this.
56. [Eden, 2007] seeks to bring clarity to the science-vs.-math-vs.-engineering

controversy by taking up a distinction due to Peter Wegner [Wegner, 1976]
among three different Kuhnian paradigms (see Ch. 4 for an explanation of
Kuhnian paradigms): a view of computer science as (1) a rationalist or
mathematical discipline, (2) a technocratic or technological discipline,
and (3) a scientific discipline. (Tedre & Sutinen 2008 also discuss these three
paradigms.) Eden then argues in favor of the scientific paradigm.
But must there be a single paradigm? Are there no other disciplines with multiple
paradigms? Does the existence of multiple paradigms mean that there is no
unitary discipline of computer science? Or can all the paradigms co-exist?
57. Wing 2008 offers Five Deep Questions in Computing:
(a) P = NP?37
(b) What is computable?
(c) What is intelligence?
(d) What is information?
(e) (How) can we build complex systems simply?
All but the last, it seems to me, concern abstract, scientific or mathematical
issues: If we consider the second question to be the fundamental one, then the
first can be rephrased as What is efficiently computable?, and the second can
be rephrased as How much of (human) cognition is computable?. The fourth
can then be seen as asking an ontological question about the nature of what
it is that is computed (0s and 1s? Information in some sense (and, if so,
in which sense)?). The last question can be viewed in an abstract, scientific,
mathematical way as asking a question about the structural nature of software
(the issues concerning the goto statement and structural programming would
fall under this category). But it can also be viewed as asking an engineering
question: How should weliterallybuild computers?
If so, then her five questions can be boiled down to two:
What is computation such that only some things can be computed? (And
what can be computed (efficiently), and how?)
(How) can be build devices to perform these computations?
And, in this case, we see once again the two parts of the discipline: the scientific
(or mathematical, or abstract) and the engineering (or concrete).
58. [Lohr, 2008] almost inadvertently points out a confusion (among educators, in
particular) about the nature of computer science. He quotes a high-school math
37 See [Austin, 1983, Mycielski, 1983, Fortnow, 2009, Walsh, 2014] for brief, semi-formal discussions of
what this means.

and computer science teacher as saying, I do feel that computer science really
helps students understand mathematics. . . . And I would use computers more
in math, if I had access to a computer lab. The confusion is that computer
science is seen as the use of a physical tool, rather than as the study of (as
well as the use of) a method of thinking (computational thinking). For that, a
computer is useful, but not necessary, as demonstrated by the Computer Science
Unplugged project (http://csunplugged.org/ (accessed 1 July 2013)).
59. [Vardi, 2011] derives a definition of computing, as an instrument for the human
mind from Leibnizs statement that his lingua characteristica universalis (or
universal formal language) and his calculus ratiocinator (or formal logic) will
offer mankind. . . a new instrument that will enhance the capabilities of the mind
to a far greater extent than optical instruments strengthen the eyes. This is
similar to Daniel C. Dennetts suggestion that the computer is a prosthesis for
the mind (see, e.g., [Dennett, 1982]). Is that a reasonable definition of computer
science?
60. [Cerf, 2012] says, not that computer science is a science, but that there really
is science to be found in computer science (my emphasis). And, presumably,
there is engineering to be found in it, and mathematics, too. This is a slightly
different metaphor from the spectrum.
61. For section on CS is what computer scientists do:

Need to add an organized list of what they do! On the science-engineering
spectrum, we would have something like this:
abstract, mathematical theory of computation

abstract, mathematical theory of computational complexity
abstract, mathematical theory of program development
software engineering
...
operating systems
...
AI
...
computer architecture
...
VLSI, etc.
62. If all true classification. . . [is] genealogical [Darwin, 1872, Ch. 14,
Classification, p. 437], then, to the extent that computer science is a single
discipline, it is both mathematics and engineering. Well see why in Ch. 6, when
we explore the history of computers.
63. Relevant to the above: Computer science is a portmanteau discipline!38
64. On complexity: One way in which computer science might be able to get a
better handle on complexity than merely by scientific experimentation, analysis,
and theorizing is by means of computational modeling, taking into account the
rules of interaction [of the parts of a complex system], the natures of the agents
[that is, of the parts], and the way the agents, rules, and ultimately whole systems
came about [Adami, 2012, 1421]; cf. [Holland, 2012].
65. On interaction between the scientific and engineering aspects of computer

science:
[McMillan, 2013] suggests that faster computers in the future will use a
different kind of computation (such as analog computation). If so, this
will be an engineering accomplishment. But, if so, it will also give rise
to new mathematical analyses of this kind of computation. Something
similar, only in the opposite direction, is happening with quantum computing:
mathematical theories of quantum computing are giving rise to physically
implemented quantum computers. So, here, the science is giving rise
to engineering. (Well explore the nature of computation in Chs. 7
and 8, and the nature of different kinds of computing (hypercomputation)
in Ch. 11. On quantum computing, see [Hayes, 1995, Aaronson, 2008,
Hayes, 2014b], [Grover, 1999]. For more on the digital-analog distinction,
see [Montague, 1960], [Pour-El, 1974], [Haugeland, 1981], [Piccinini, 2004a],
[Piccinini, 2007c], [Piccinini, 2008], [Piccinini, 2009], [Piccinini, 2010b],
[Piccinini, 2011], [Piccinini and Craver, 2011], [Piccinini, 2012].)
66. On microscopy:
There used to be a British Quarterly Journal of Microscopical Science (1853
1965), affiliated with the Microscopical Society of London. From its inaugural
Preface:
Recent improvements in the Microscope having rendered that

instrument increasingly available for scientific research, and having
created a large class of observers who devote themselves to whatever
department of science may be investigated by its aid, it has been
thought that the time is come when a Journal devoted entirely to
objects connected with the use of the Microscope would contribute
to the advancement of science, and secure the co-operation of all
interested in its various applications.
The object of this Journal will be the diffusion of information
relating to all improvements in the construction of the Microscope,
and to record the most recent and important researches made by its
38 Inspired by the philosopher Hilary Putnams comment that his picture of [language] use was a
portmanteau affair. There was the computer program in the brain, and there was the description of the
external causes of the language users words [Putnam, 1994, 458]. A portmanteau word is one with two
meanings packed up into one word, like slithy (meaning lithe and slimy) or smog (meaning smoke
and fog); the term was coined by the great logician and storyteller Lewis Carroll [Carroll, 1871].
aid in different departments of science, whether in this country or on

the continent.. . .
It is, perhaps, hardly necessary to apologise for the title of the
Journal, as the term Microscopical, however objectionable in its
origin, has acquired a conventional meaning by its application to
Societies having the cultivation of the use of the Microscope in view,
and so fully expresses the objects of the Journal, that it immediately
occurred as the best understood word to employ. It will undoubtedly
be a Journal of Microscopy and Histology; but the first is a term but
recently introduced into our language, and the last would give but a
contracted view of the objects to which the Journal will be devoted.
One could replace microscope with computer (along with their cognates) and,
perhaps, histology with something like mathematical calculations, and this
could read like a manifesto for the ACM.
The first issue included, besides many articles on what we now call microbiology,
a paper on Hints on the Subject of Collecting Objects for Microscopical
Examination and a review of a book titled The Microscopist; or a Complete
Manual on the Use of the Microscope.
Here is a passage from that review:
As cutting with a sharp instrument is better than tearing with the nails,
so vision with the microscope is better than with the naked eye. Its
use is, therefore, as extensive as that of the organ which it assists, and
it cannot be regarded as the property of one branch of science more
than another. [Anonymous, 1853, 52]
And here is a paraphrase:
As vision with the microscope is better than with the naked eye, so
thinking with the computer is better than with the mind alone. Its use
is, therefore, as extensive as that of the organ which it assists, and it
cannot be regarded as the property of one branch of science more than
another. [Anonymous, 1853, 52]
This is reminiscent of Daniel Dennetts arguments for the computer as a

prosthesis for the mind [Dennett, 1982].
In the March 1962 issue, the editors announced a change in focus, based on
the nature of many of their articles, to cytology, thus apparently changing their
interest from the tool to what can be studied with it. The change officially
occurred in 1966, when the journal changed its name to the Journal of Cell
Science (and restarted its volume numbers at 1).
67. [Denning, 2003, 15] makes an offhand comment that has an interesting
implication. He says, Computer science was born in the mid-1940s with the
construction of the first electronic computers. This is no doubt true. But it
suggests that that the answer to the question of what computer science is has to
be that it is the study of computers. The study of algorithms is much older, of
course, dating back at least to Turings 1936 formalization of the notion, if not
back to Euclids geometry or ancient Babylonian mathematics [Knuth, 972a].
Yet the study of algorithms is clearly part of modern computer science. So,
modern computer science is the result of a merger (a marriage) between the
engineering problem of building better and better automatic calculating devices
(itself an ancient endeavor) and mathematical problem of understanding the
nature of algorithmic computation. And that implies that modern computer
science is both engineering and science. (And that can be in two ways: that
there are separate branches of computer science, or that they blend into each
other.)
68. Question for end of chapter to think about:
In 3.11, I said that it makes noor very littlesense to have a program without
a computer to run it on. But the some of the earliest AI programs (for playing
chess) were executed by hand (CITE Turing, Shannon, NewellShawSimon,
IJGood). So, did these programs have a computer to run on? (Were the
humans, who hand-executed them, the computers that these programs ran
on?) When you debug a computer program, do you do the debugging by
hand?39
69. On Knuths ideas about teaching computers:
(a) One potential restriction on Knuths theory that we teach computers how
to do something (more specifically, that, insofar as computer science is
the study of what tasks are computable, it is the study of what tasks are
teachable) is that teaching is propositional, hence explicit or conscious.
But one way of getting a computer to do something is by training it
(via a connectionist algorithm, or via a machine-learning algorithmnote
that learning, in this sense of machine learning, is more than being
(propositionally) taught). And such training is implicit or unconscious. We,
as external, third-person observers still dont know how to do it consciously
or explicitly. Knowing how is not necessarily the same as knowing that.
(b) It is a commonplace that a computer can do anything for
which precise and unambiguous instructions can be given.
[Mahoney and Haigh (ed.), 2011, 80].
i. That is, a computer can do anything for which an algorithm can
be given (for, after all, isnt an algorithm merely precise and
unambiguous instructions?). Thought of this way, the comment is
almost trivial.
39 Thanks to Stuart C. Shapiro for this suggestion.
ii. But consider that to give such instructions (to give an algorithm) is
to be able to explicitly teach the computer (or the executor, more
generally) how to do that thing.
iii. And such explicit teaching is to make the task completely conscious
(to make it a System 2 task [Kahneman, 2011]). An algorithm in
the form of an artificial neural network is akin to building in to the
computer the ability, as if innate, to do that thing. Such a computer
could not tell us how it was doing it; it would have no conscious
access to its algorithm. An algorithm in the form of an explicit
machine-learning program that would enable the computer to learn
how to do that thing is somewhere in the middle; it would be conscious
of its ability to learn, but not necessarily of how to do the thing; it
might not necessarily be able to teach someone or something else how
to do it, unless it could observe itself doing it and develop a theory of
how to do it (which theory would be expressed in an algorithm).
iv. On what is computable: Something is computable just in case precise
and unambiguous instructions can be given for it. So, the question
becomes: What tasks are amenable to precise and unambiguous
instructions? Does this mean that chess is computable but vision is
not? After all, one cannot give precise and unambiguous instructions
that would enable someone to see. So there seems to be another notion
of computability: A behavior is computable if it can be described
in terms of such instructions. The entity that exhibits that behavior
naturally might not use, or be able to use, those instructions in order
to behave that way. But we might be able to give those instructions
to another system that could use them to exhibit that behavior. So,
for instance, the human brain might not literally compute in the sense
of executing an algorithm im order to see, but a computer using that
algorithm might be able to exhibit visuall behavior. (Whether it sees,
phenomenologically, is a philosophical question!) Similarly, the solar
system might not be executing Keplers laws, but an artificial solar
system might.
70. [Knuth, 1985, 170171] more emphatically defines computer science as

primarily the study of algorithms simpliciter, primarily because he think[s]
of algorithms as encompassing the whole range of concepts dealing with well-
defined processes, including the structure of data that is being acted upon as
well as the structure of the sequence of operations being performed, preferring
the name algorithmics for the discipline. He also suggests that what computer
scientists have in common (and that differentiates them from people in other
disciplines) is that they are all algorithmic thinkers [Knuth, 1985, 172]. We
will see what it means to think algorithmically in Ch. 7.
71. CS as science of procedures:

Kay argues that computational linguistics is the branch of linguistics that studies
languages computationally, that is, that tries to devise a computational theory
of the nature of language (and is distinct from natural language processing,

or the use of computers to process linguistic data. He considers the former to
be scientific and the latter to be engineering, where scientists try to understand
their subject mainly for its own sake, though they are gratified when some of
what they do proves useful. Engineers seek only the understanding needed
to reach practical ends [Kay, 2010, 1]. [C]omputational linguists. . . look to
computer science for insight into their problems. If communication is. . . about
building structures by remote control in the heads of others, then it is all about
process, and computer science is the science of process, conceived in its most
fundamental and abstract way [Kay, 2010, 2; italics in original; my boldface].
72. There can be algorithmic, that is, computational, theories (or models) of
biological phenomena such as cells. See [Covert, 2014]. Or computational
theories of plants; see [Pollan, 2013, 104105].
73. Turing and von Neumann not only played leading roles in the
design and construction of the first working computers, but were also
largely responsible for laying out the general logical foundations for
understanding the computation process, for developing computing
formalisms, and for initiating the methodology of programming:
in short, for founding computer science as we now know it.
[Robinson, 1994, 5]
74. The ascendancy of logical abstraction over concrete realizaion has

ever since been a guiding principle in computer science, which has
kept itself organizationally almost entirely separate from electrical
engineering. The reason it has been able to do this is that computation
is primarily a logical concept, and only secondarily an engineering
one. To compute is to engage in formal reasoning, according to
certain formal symbolic rules, and it makes no logical difference
how the formulas are physically represented, or how the logical
transformations of them are physically realized.
Of course no one should underestimate the enormous importance
of the role of engineering in the history of the computer. Turing
and von Neumann did not. They themselves had a deep and quite
expert interest in the very engineering details from which they were
abstracting, but they knew that the logical role of computer science is
best played in a separate theater. [Robinson, 1994, 12]
75. Just as philosophy can be about anything (for any x, there is a philosophy of
x), so computer science can be about pretty much anything. This can happen
in two ways: For any x, there can be a computational theory of x. (Come to
think of it, there are other disciplines like this, such as history.) But one can also
use computational methods to study pretty much any x (see comments by the
computer scientist Mehran Sahami in [Reese, 2014a]).
76. On computational thinking:

The ability to program or to understand different aspects of computing

systems really involves clarity of thought. If youre going to program
a computer, you need to specify a set of instructions for the computer
to execute, and those have to be sufficiently clear to get your problem
solved. Otherwise, you have errors. Just having the clarity of thinking
to lay out instructions is an analytical skill thats useful for everyone.
(Mehran Sahami, cited in [Reese, 2014a].)
Algorithmic thinking is an interpretation of the world in which a

person understands and formulates actions in terms of step-by-step
procedures that give unambiguous results when carried out by anyone
(or by a suitable machine). It resembles scientific thinking, which
seeks to invent standard ways of observing that allow anyone to see
and reproduce physical effects. Algorithmic thinking emphasizes the
standard procedure[,] and scientific thinking the standard observer.
[Denning, 1999, 3]
77. Another good source on complexity is [Pylyshyn, 1992]. It is also a good source
on computer science as the science of information processing.
78. On computational thinking:

An idea due to Alan Perlis:
. . . procedural literacythe ability to read and write processes

. . . a fundamental skill that everyone should learn at the university,
regardless of field of study. . . . It is not fluency with computer
technology per se but rather is about how one can rationally navigate
the tower of abstraction that exists within any computation system.
The purpose of a course in programming is to teach
people how to construct and analyze processes.
[Wheeler, 2013, 296], quoting [Perlis, 1962, 209]
Also:
A course in programming. . . , if it is taught properly, is concerned

with abstraction: the abstraction of constructing, analyzing, and
describing processes.
I personally feel that the ability to analyze and construct processes
is av ery important ability, one which the student has to acquire sooner
or later. I believe that he [sic] does acquire it in a rather diluted way
during four years of an engineering or science program. I consider it
also important to a liberal arts program.
This, to me, is the whole importance of a course in programming.
It is a simulation. The point is not to teach the students how to use
ALGOL, or how to program the 704. These are of little direct value.
The point is to make ths students construct complex processes out of
simpler ones (and this is always present in programming) in the hope
that the basic concepts and abilities will rub off. A properly designed
programming course will develop these abilites better than any other
course. [Perlis, 1962, 210]
And:
Programming is the art of stating procdures. (John McCarthy, in

[Perlis, 1962, 211])
79. Taught right, programming. . . is the hands-on, inquiry-based way
that students learn what we call computational thinking: problem
analysis and decomposition, abstraction, algorithmic thinking,
algorithmic expression, stepwise fault isolation (we call it
debugging) and modeling. [Lazowska, 2014, A26]
80. Computer science is actually a study of things that happen. Its not
only a study of a basic theory, and it is not just the business of making
things happen. Its actually a study of how things happen. So the
advice [to students] is: Dont lose the link [between practice and
theory]! [Frenkel, 1993, 97, quoting Robin Milner]
81. The primary subject matter of computer science is computational
processes, or processes that can be specified and studied
independently of any physical realization. Processes, whether
computational or not, are composed of interactions among objects,
so we can also characterize the subject matter of computer science as
interaction patterns. [Colburn and Shute, 2011, 249]
The main claim is stated without argument. The or clause is a bit hard to
interpret. On one possible reading, it is an elaboration of the main claim.
But surely being independent of any physical realization is not the only
feature of being computational. For instance, the halting problem is not
computational, yet is independent of any physical realization. And surely some
processes (a better term, of course, would be procedures), perhaps recipes (as
we will discuss in Ch. 10), would also be counterexamples. Another possible
reading is that computer science has two kinds of processes as subject matter:
computational ones and ones that are realization independent. But that leaves
open the question of which ones are computational. It is also a bit odd to
say that computer science should not be interested in the realization of such
processes (or is that more properly the concern of computer engineering?); well
return to this topic in Ch. 14.
82. On complexity and (software) engineering:
[P]rograms are built from programs. . . . Programs are compilations

in another sense as well. Even the smallest sub-program is also
a compilation of sub-components. Programmers construct sub-
programs by assembling into a coherent whole such discrete program
elelments as data, data structures, and algorithms. The engineering
in software engineering involves knowing how to assemble these

components to produce the desired behavior. [Samuelson et al., 1994,
23262327]
The idea that a complex program is just a construction from simpler things,
each of whichrecursivelycan be analyzed down to the primitive operations
and data structures of ones programming system (for a Turing machine, these
would be the operations of printing and moving, and the data structures of 0
and 1) is, first, the underlying way in which complexity can be dealt with and,
second, where engineering (considered as a form of construction) comes into the
picture.
83. Computer science offers a standard way to handle complexity:
hierarchical structure. [Lamport, 2012, 16]
84. The following material might go here, or else in Ch. 4 (or both):
As we mentioned in Ch. 2, [McGinn, 2015b] argues that philosophy is a science

just like physics (which is an empirical science) or mathematics (which is a
formal science), likening it more to the latter than the former. To make
his argument, he has to come up with a characterization of science, which
we will look at in more details in Ch. 4. But an interesting exercise for the
reader is to take that characterization, and see whether computer science fits it.
[McGinn, 2015b, 85] thinks that it is, like math, a formal science.
85. If not used in this chapter, then put in Ch. 4:
As we enter life, we all struggle to understand the world. Some
of us continue this struggle with dogged determination. These are
the scientists. Some of them realize that computation provides a
privileged perspective to understand the world outside and the one
within. These are the computer scientists. [Micali, 2015, 52].
86. On carving nature at its joints 3.4:
(There may be a better place to put this observation, perhaps in a section on what
a computer is or what thinking is.) When we define or formalize something,
we often find that our definition or formalization includes something that was
not previously thought to fall under the concept or excludes something that was
thought to fall under it. We can then either reformulate the definition or bite the
bullet about the inclusions and exclusions.
Here is an observation about this point:

In mathematics, tentative proofs are sometimes refuted not
by logical contradiction. . . , but instead by a logically-sound
consequence. . . that is unexpected and unwanted. . . . Lakatos [1976]
argues that in such situations mathematicians should not just
implicitly reject the unwelcome result (monster barring), but
should instead either change their acceptance of the result (concept

stretching. . . ), or change the statement of the conjecture either
by explicitly incorporating a condition that disallows the unwanted
consequence (lemma incorporation. . . ), or by inventing a wholly
new conjecture. [Staples, 2015, 3.1]
87. On the first occurrence (here or earlier) of procedure, note that this is a term of
art that is being used relatively informally here but that will be studied in more
detail in later chapters, especially Ch. 7.
88. compare and contrast computer science with library science, political
science? [Mahoney and Haigh (ed.), 2011, 195] has a relevant point to make
here: Is wondering whether computer science is a science, given its name,
. . . laboring under a misapprehension rooted in the English use of
computer science to denote a subject other languages refer to as
informatics.
But he then goes on to point out that textbooks and courses from informatics
departments cover exactly the same material as textbooks and courses from
computer science departments.
So I must wonder, as Shakespeares Juliet once did, Whats in a
name? Here too the rose smells the same.
89. [Mahoney and Haigh (ed.), 2011, 26; my italics] suggests the spectrum view of
the nature of computer science, given its historical origins in both mathematics
and engineering:
Between the mathematics that makes the device theoretically possible
and the electronics that makes it practically feasible lies the
programming that makes it intellectually, economically, and socially
useful. Unlike the extremes, the middle remains a craft, technical
rather than technological, mathematical only in appearance.
90. Computing has no nature. It is what it is because people have made it
so. [Mahoney and Haigh (ed.), 2011, 109]
This is not exactly of the form computer science is what computer scientists do,
though it bears a superficial resemblance. But I think that it is more sophisticated.
Mahoneys point is that, unlike the other natural sciences (for example, physics,
chemistry, biology, and so on), computer science only came into existence when
its two histories (logical-mathematical and engineering) began to intersect in the
1940s, so its nature only came to be what those logicians, mathematicians, and
engineers were doing.
91. [Mahoney and Haigh (ed.), 2011, 159ff] cites the argument that computer
science is not a natural science because the computer is an artifact, not a
natural phenomenon, and science is about natural phenomena (p. 159). This
goes straight against [Newell et al., 1967], though it is consistent with Simons
claim that computer science is a science of the artificial. It is also orthogonal

to [Shapiro, 2001]s claim that computer science is a natural science, because
Shapiros argument is based on a different premise, namely, that computer
science is the study of procedures, and procedures are natural phenomena.
But does not subscribe to this viewpoint, not because he thinks that computers
are natural phenomena, but because he thinks that there is no sharp dividing line
between nature and artifact (p. 159). And he thinks that that is the case because
we use artifacts to study nature (p. 159)[w]e know about nature through the
models we buid of it (p. 161) and because [a]rtifacts work by the laws of
nature, and by working reveal those laws (p. 160). In other words, artifacts are
part of nature.
He goes on to suggest that nature itself might be ultimately computational in

nature (so to speak). We will explore some of these ideas in Ch. 9 when we
consider whether the universe might be a computer.
92. Academically and professionally, computer engineering took charge
of the hardware, while computer science concerned itself with the
software. . . . [Mahoney and Haigh (ed.), 2011, 108]
To the extent, then, that software and hardware cannot easily be distinguished
(which we will begin to investigate in Ch. 12), then neither can computer
engineering and computer science.
93. Computation is sure to become one of the most important of the
sciences. This is because it is the science of how machines can be
made to carry out intellectual processes. [McCarthy, 1963, 1, my
italics]
First, note that he thinks that it is a science, presumably just like other sciences
(else it wouldnt be destined to become one of the most important of them).
Second, the nature of this science is akin to the view that computer science is the
study of what is computable: Machines can be made to carry out intellectual
processes if those processes are computable. Why intellectual processes?
Well, surely mathematical processes are intellectual, and to the extent that other
intellectual processes are expressible mathematicallyfor example, by being
codable into symbolic notation (and, ultimately, into binary notation)those
other intellectual processes are mathematical, hence potentially computable.
Why is it a science? McCarthy doesnt say, but I would guess that it is a science
in the same sense that mathematics is.
94. On whether one of the fundamental questions of computer science is What can
be automated?:
Automatable does not imply computable (though computable does seem to
imply automatable): Witness the invention of the direct dialing system in
telephony, which automated the task of the human operator. Yes, direct dialing
is computable, but it wasnt a computer that did this automation.
95. On Knuths distinction between art and science,

see also [Simon and Newell, 1958, 4ff]s distinction between well-structured
and ill-structured problems. The latter would probably still fall under Knuths
science category, given that Simon & Newell think that they are computable
(if only heuristically).
96. What Is Computer Science?

There is no simple, one-sentence characterization of computer science. Any
attempt at one is no better than the celebrated descriptions of an elephant by the
blind men (https://en.wikipedia.org/wiki/Blind men and an elephant). Many, if
not most or all, such attempts wind up describing the entire subject, but focusing
on only one aspect of it: Newell, Perlis, & Simons definition:
Computer science is the study of the phenomena surrounding

computers. [Newell et al., 1967, 1374, my emphasis]
(where the surrounding phenomena include the hardware, their programs or

algorithms, and all that goes along with them) differs only in emphasis from
Knuths definition:
[C]omputer science is. . . the study of algorithms and related

phenomena such as computers [Knuth, 1974b, 323].
What I will call computer science (but that could just as easily be called
computing, computing science, informatics, etc.) is the scientific study of a
family of topics surrounding both theoretical and practical computing.
It evolved or descended from (or is the offspring of the marriage or merger
of) two historical traditions: (1) the foundations of mathematics (leading from
inquiries into the nature of logic to the Turing machine) and (2) the attempts to
design and construct a calculating machine (from the Antikythera Mechanism
of ancient Greece, through Pascals and Leibnizs calculators and Babbages
machines, to the ENIAC and beyond).
The topics studied in contemporary computer science roughly align along a
spectrum ranging from the mathematical theory of computing, at one end, to
the engineering of physical computers, at the other. (Newell et al., above,
were looking at this spectrum from one end; Knuth was looking at it from
the other end.) The topics share (probably at most) a family resemblance,
not only to each other, but also to other disciplines (including mathematics,
electrical engineering, information theory, communication, etc.), and they
overlap with issues discussed in the cognitive sciences, philosophy (including
ethics), sociology, education, the arts, and business.
The single most central question of computer science is:
(a) What can be computed?
Several other questions follow logically from this:

2 What can be computed efficiently?

2 What can be computed practically?
2 What can be computed physically?
How can computations be performed practically?
2 What should be computed?
Question 96a is the question of computability vs. non-computability. It is the
question that Church, Turing, Godel, and others were originally concerned with
and whose answer has been given as the Church-Turing Computability Thesis:
A function is computable if and only if it is computable by a Turing machine
(or any formalism logically equivalent to a Turing machine, such as Churchs
lambda calculus or Godels general recursive functions. A standard example of
a non-computable function is the Halting Problem.
Although similar in intent, George Forsythes questionWhat can be
automated? [Forsythe, 1968]can be understood in a slightly different way:
Presumably, a process can be automatedthat is, done automatically, by a
machine, without human interventionif it can be expressed as an algorithm.
But there might be automatable processes that are not algorithmic, not so much
because they are not Turing-machine computable, but because they are not the
kinds of processes expressible as mathematical functions.
This is the central question, because all other questions presuppose it.
The fundamental task of any computer scientist, whether at the purely
mathematical/theoretical end of the spectrum or the purely practical/engineering
end is to determine whether there is a computational solution to a given
problem, and, if so, how to implement it. But those implementation questions
are covered by the rest of the questions on the above list, and only make
sense after the first question has been answered. Alternatively, they facilitate
answering that first question; in any case, they serve the goal of answering
it. Finding a computational solution to a problem requires what has been
called computational thinking [Wing, 2006], which I understand to be,
algorithmic (or procedural) thinking (http://www.cs.cmu.edu/CompThink/).
This has been characterized as thinking in such a way that a problems
solution can effectively be carried out by an information-processing agent
(https://www.cs.cmu.edu/CompThink/resources/TheLinkWing.pdf,
https://computinged.wordpress.com/2011/03/22/a-
definition-of-computational-thinking-from-jeanette-wing/). Here, it is important
not to limit such agents to computers, but to include humans!
So, for example, the fundamental question of AI is: Is cognition computable?
This is one reason why I prefer to call AI computational cognition
[Rapaport, 1995], [Rapaport, 2003]. Given the advances that have been made in
AI to date, it seems clear that at least some aspects of cognition are computable,
so a slightly more precise question is: How much of cognition is computable?
[Rapaport, 2012b, 2, pp. 3435]
The how question is also important: Computer science is not only interested
in a mere existence statemernt to the effect that a problem is computable; it also
requires a constructive answer in the form of an algorithm that explicitly shows

how it is computable. (See the Calvin-Hobbes cartoon about a toaster making
bread disappear and toast appear.; this may also be relevant to the notion of
computer science as magic.)
Question 96? is the question studied by the branch of computer science known
as complexity theory. Given the set of computable functions, the question
here is, so to speak, which of them can be computed in less time than the
age of the universe or less space than the size of the universe. The principal
distinction here is whether a function is in the class called P (in which case, it
is efficiently computable) or in class NP (in which case it is not efficiently
computable but it is efficiently verifiable). (Technically, P is the class of
functions computable in Polynomial time, and NP is the class of functions
computable in Non-deterministic Polynomial time.) Almost all standard
algorithms are in P. One important function that is in NP is the Satisfaction
problem (given a molecular proposition of propositional logic with n atomic
propositions, under what assignment of truth-values to those atomic propositions
is the molecular proposition true (or satisfied)? Whether P = NP is one of
the major open questions in mathematics and computer science; most computer
scientists both hope and believe that P 6= NP. (See [Fortnow, 2009] for a semi-
technical discussion, and [Fortnow, 2013] for a non-technical discussion.)
Question 96? is considered both by complexity theorists as well as by more
practically-oriented software engineers. Given a computable function in P, what
are some practically efficient methods of actually computing it? For example,
under certain circumstances, certain sorting algorithms are more efficient in a
practical sense (e.g., faster) than others. Even a computable function that is in
NP might be practically computable in certain special cases. And some functions
might only be practically computable via a heuristic: A heuristic for problem
p can be defined as an algorithm for some problem p0 , where the solution to p0 is
good enough as a solution to p [Rapaport, 1998, 406]. Being good enough
is, of course, a subjective notion, but the idea behind it is related to Herbert
Simons notion of bounded rationality: We might not be able to solve p because
of limitations in space, time, or knowledge. But we might be able to solve p0
algorithmically within the required spatio-temporal-epistemic limits. And if the
algorithmic solution to p0 gets us closer to a solution to p, then it is a heuristic
solution to p. But it is still an algorithm. [Oommen and Rueda, 2005, 1] call
the good enough solution a sub-optimal solution that, hopefully, is arbitrarily
close to the optimal..
Question 96? brings in engineering considerations. Even a practically
efficient algorithm for computing some function might run up against physical
limitations. Just as one example, even if, eventually, computational linguists
devise practically efficient algorithms for natural-language competence
(understanding and generation; [Shapiro, 1989], [Shapiro and Rapaport, 1991]),
it remains the case that humans have a finite life span, so the infinite capabilities
of natural-language competence are not really required (a Turing machine isnt
needed; a finite-state machine might suffice). This is also where I would
place all of those aspects of the discipline that are usually covered under the
heading computer engineeing. It is the question that issues in the design and
construction of real computers are concerned with.
Finally, question 96? brings in ethical considerations [Tedre, 2015, 167168].
Actually, the question is slightly ambiguous. It could simply refer to questions
of practical efficiency: Given a sorting problem, which sorting algorithm should
be used; that is, which one is the best or most practical or most efficient
in the actual circumstances? But this sense of should is not really does not
really differentiate this question from question 96?. It is the ethical interpretation
that makes this question interesting: Let us suppose that the goal of AI is
practically and efficiently computable by physically plausible computers. One
can and should still raise the question whether such artificial intelligences
should be created, and whether we (their creators) have any ethical or moral
obligations towards them. (See [Lem, 1971], [LaChat, 1986], and ARTICLE IN
NEW SCIENTIST??)

Items in boldface are especially interesting.
NOTE FOR NEXT DRAFT:
FACTOR OUT REFS TO: DNA, MAGIC, & QUANTUM COMPUTING (ETC.) IN
SEPARATE SECTION.
Abrahams, Paul (1987), What Is Computer Science?, Communications of the
ACM 30(6) (June): 472473.
Abrahams, Paul W.; & Lee, Graham (2013), Computer Science Is Not a Science
(two separate letters to the editor), Communications of the ACM 56(1) (January): 8.
Anthes, Gary (2006), Computer Science Looks for a Remake, Computerworld
(1 May).
Bornat, Richard (2006), Is Computer Science Science?, 4th European
Conference on Computing and Philosophy (ECAP06, 2224 June, Trondheim,
Norway).
Carey, Kevin (2010), Decoding the Value of Computer Science, Chronicle of
Higher Education 58(12) (12 November): A88.
Cerf, Vinton G. (2012), Where Is the Science in Computer Science?,
Communications of the ACM 55(10) (October): 5.
Argues that, even if computer science might once have focused on computing
machines, it should now be more focused on predict[ing] likely outcomes
based on models[, which] is fundamental to the most central notions of the
scientific method.
The College Board (2012), AP Computer Science: Principles; Course
Annotations, online at:
http://www.collegeboard.com/prod downloads/computerscience/1 Annotations-
08-21-12 ld.pdf (accessed 8 July 2013)
Cook, Stephen A. (1983), An Overview of Computational Complexity,
Communications of the ACM 26(6) (June): 400408.
Crowcroft, Jon (2005), On the Nature of Computing, Communications of the
ACM 48(2) (February): 1920.
Argues that computer science is not the study of either the natural or the
artificial but of the virtual.
Another essay on computer science and the virtual is [Pylyshyn, 1992].
CSTA [Computer Science Teachers Association] Curriculum Improvement Task
Force (2005), The New Educational Imperative: Improving High School Com-
puter Science Education (New York: Association for Computing Machinery); on-
line at:
http://csta.acm.org/Communications/sub/DocsPresentationFiles/White Paper07 06.pdf
(accessed 1 July 2013).
Relevant to this chapter, see 1.1 (pp. 1719), What Is Computer Science
Anyway?.
Denning, Peter J. (1980), What Is Experimental Computer Science?,
Denning, Peter J. (1995), Can There Be a Science of Information?, ACM

Computing Surveys 27(1) (March): 2325.
Denning, Peter J. (1999), Computer Science: The Discipline, in Anthony Ralston
& David Hemmindinger (eds.) (2000), Encyclopedia of Computer Science: PAGES.
Denning, Peter J. (2005), Is Computer Science Science?, Communications of the
ACM 48(4) (April): 2731.
Denning, Peter J. (2007), Computing Is a Natural Science, Communications of
the ACM 50(7) (July): 1318.
Denning, Peter J. (2009), Beyond Computational Thinking, Communications of
the ACM 52(6) (June): 2830.
Denning, Peter J.; Comer, Douglas E.; Gries, David; Mulder, Michael C.;
Tucker, Allen; Turner, A. Joe; & Young, Paul R. (1989), Computing as a
Discipline, Communications of the ACM 32(1) (January): 923.
Denning, Peter J.; & Rosenbloom, Paul S. (2009), Computing: The Fourth Great
Domain of Science, Communications of the ACM 52(9) (September): 2729.
Denning, Peter J.; & Freeman, Peter A. (2009), Computings Paradigm,
Communications of the ACM 52(12) (December): 2830.
Dijkstra, E.W. (1987), Mathematicians and Computing Scientists: The Cultural
Gap, Abacus 4(4): 2631.
Easton, Thomas A. (2006), Beyond the Algorithmization of the Sciences,
Communications of the ACM 49(5) (May): 3133.
Ensmenger, Nathan (2011), Computing as Science and Practice, Science 333
(9 September): 1383.
Fant, Karl M. (1993), A Critical Review of the Notion of Algorithm in Computer
Science, Proceedings of the 1993 ACM Annual Computer Science Conference
(Indianapolis) (New York: ACM): 16;
http://tinyurl.com/qdaojox40 (accessed 6 December 2013).
Feitelson, Dror G. (guest ed.) (2007), Special Issue on Experimental Computer
Science, Communications of the ACM 50(11) (November): 2459.
Forsythe, George F. (1967), A Universitys Educational Program in Computer
Science, Communications of the ACM 10(1) (January): 38.
Green, Penelope (2014), Magic in the Mundane, New York Times (17 July):
D1, D4.
Accessed 26 July 2014 from:
http://www.nytimes.com/2014/07/17/garden/putting-magic-in-the-
mundane.html
An inventor of enchanted objects [embedded with computers] sees a future
where we can all live like wizards.
Guzdial, Mark (2008), Paving the Way for Computational Thinking,
Communications of the ACM 51(8) (August): 2527.
Hartmanis, Juris (1995), On Computational Complexity and the Nature of
Computer Science, ACM Computing Surveys 27(1) (March): 715.
40 http://staffweb.worc.ac.uk/DrC/Courses%202006-7/Comp%204070/Reading%20Materials/FAnt.pdf
This paper covers much of the same ground, and in many of the same words,
as [Hartmanis, 1993], but it is more easily accessible.
Among several follow-up commentaries, see esp.:
Denning, Peter J. (1995), Can There Be a Science of Information?,
ACM Computing Surveys 27(1) (March): 2325.
Loui, Michael C. (1995), Computer Science Is a New Engineering
Discipline, ACM Computing Surveys 27(1) (March): 3132.
Plaice, John (1995), Computer Science Is an Experimental Science,
Stewart, N.F. (1995), Science and Computer Science, ACM Computing
Surveys 27(1) (March): 3941; a longer version is online at:
http://www.iro.umontreal.ca/stewart/science computerscience.pdf
(accessed 5 June 2013).
Wulf, W.A. (1995), Are We Scientists or Engineers?, ACM
Computing Surveys 27(1) (March): 5557.
And Hartmaniss replies:
Hartmanis, Juris (1995), Response to the Essays On Computational
Complexity and the Nature of Computer Science , ACM Computing
Surveys 27(1) (March): 5961.
Hayes, Brian (2015), Cultures of Code, American Scientist 103(1) (January-
February): 1013, http://www.americanscientist.org/issues/pub/cultures-of-code
Distinguishes three communities in the world of computation: computer
science, computational science, and software development.
Hendler, James; Shadbolt, Nigel; Hall, Wendy; Berners-Lee, Tim; & Weitzner,
Daniel (2008), Web Science: An Interdisciplinary Approach to Understanding the
Web, Communications of the ACM 51(7) (July): 6069.
Hsu, Francis (2013), Is Computing Science?, Communications of the ACM 56(8)
(August): 9.
Argues that there are no clear boundaries between branches of knowledge.
Kroes, Peter (2010), Engineering and the Dual Nature of Technical Artefacts,
Cambridge Journal of Economics 34: 5162.
Argues that technical artefacts are physical structures with functional
properties
Kukla, A. (1989), Is AI an Empirical Science?, Analysis 49: 5660.
The answer is: No; it is an a priori science or discipline like mathematics.
For the opposite point of view, see:
Burkholder, Leslie (1999), Are AI and Mechanics Empirical Disciplines?,
Journal of Experimental and Theoretical Artifical Intelligence 11: 497500.
Lloyd, Seth (1990), The Calculus of Intricacy: Can the Complexity of a Forest
Be Compared with That of Finnegans Wake?, The Sciences (September/October):
3844.
An informal discussion of the nature of complexity and how it can be
measured using information theory or ideas from computing.
Lohr, Steve (2006), Computing 2016: What Wont Be Possible?, New York Times
(31 October): F3.
Naur, Peter (1995), Computing as Science, Appendix 2 of An Anatomy of Human
Mental Life (Gentofte, Denmark: naur.com Publishing): 208217; online at
http://www.naur.com/naurAnat-net.pdf (accessed 1 July 2013) via
http://www.naur.com/Nauranat-ref.html (accessed 1 July 2013).
Parnas, David Lorge (1998), Software Engineering Programmes Are Not
Computer Science Programmes, Annals of Software Engineering.
See also:
Dewar, Robert; & Astrachan, Owen (2009), CS Education in the U.S.:
Heading in the Wrong Direction?, Point/Counterpoint, Communications of
the ACM 52(7) (July): 4145.
Pylyshyn, Zenon W. (1987), Computers and the Symbolization of Knowledge, in
Ralph Morelli, W. Miller Brown, Dina Anselmi, Karl Haberlandt, & Dan Lloyd
(eds.), Minds, Brains and Computers: Perspectives in Cognitive Science and
Artificial Intelligence (Norwood, NJ: Ablex, 1992): 8294; portions online at:
http://tinyurl.com/Morelli-etal1987 (accessed 19 July 2013); preprint online at:
http://ruccs.rutgers.edu/ftp/pub/papers/suffolk.pdf (accessed 19 July 2013).
Shneiderman, Ben (2007), Web Science: A Provocative Invitation to Computer
Science, Communications of the ACM 50(6) (June): 2527.
Simon, Herbert A. (1996), Understanding the Natural and Artificial Worlds,
Ch. 1, pp. 124 of Simons The Sciences of the Artificial, Third Edition (Cambridge,
MA: MIT Press).
Originally written in 1969; updated in 1981.
In this chapter, Simon also has interesting things to say about whether
computer programs are scientific theories, which we will investigate in Ch. 15
Tedre, Matti (2015), The Science of Computing: Shaping a Discipline (Boca Raton,
FL: CRC Press).
An investigation of the philosophical issues around the nature and history of
computer science, examining whether it is a science, and, if so, what kind of
science it might be.
Tedre, Matti; & Moisseinen, Nella (2014), Experiments in Computing:
A Survey, Scientific World Journal, Vol. 2014, Article ID 549398,
http://dx.doi.org/10.1155/2014/549398 (accessed 3 April 2015).
A survey of the nature of experiments in science, and whether computer
science is experimental in nature.
Turner, Ray (2010), Programming Languages as Mathematical Theories, in Jordi
Vallverdu (ed.), Thinking Machines and the Philosophy of Computer Science:
Concepts and Principles (IGI Global): 6682.
Vardi, Moshe Y. (2014), Boolean Satisfiability: Theory and Engineering,
Communications of the ACM 57(3) (March): 5
http://cacm.acm.org/magazines/2014/3/172516-boolean-satisfiability/fulltext
An editorial on the difference between computer science approaches to an
important theoretical problem and computer engineering approaches to a
practical version of the same problem.
Wing, Jeannette M. (2006), Computational Thinking, Communications of the

ACM 49(3) (March): 3335,
https://www.cs.cmu.edu/15110-s13/Wing06-ct.pdf
(accessed 30 September 2015)
For a contrasting view, see:
Denning, Peter J. (2009), Beyond Computational Thinking,
Communications of the ACM 52(6): 2830.
Wing, Jeannette M. (2008), Five Deep Questions in Computing, Communications
of the ACM 51(1) (January): 5860.
Wings five questions are:
(a) P=NP?
(b) What is computable?
(c) What is intelligence?
(d) What is information?
(e) (How) can we build complex systems simply?
2. Websites:
Chazelle, Bernard (2006), The Algorithm: Idiom of Modern Science
http://www.cs.princeton.edu/chazelle/pubs/algorithm.html
accessed 5 November 2012.
Denning, Peter J., Great Principles of Computing,
http://cs.gmu.edu/cne/pjd/GP/GP-site/welcome.html (accessed 8 July 2013)
Denning, Peter J. (ed.), Symposium on The Science in Computer Science,
Ubiquity, http://ubiquity.acm.org/symposia.cfm
Department of Computing Sciences, Elon University, Elon, NC (2007), What Is
Computer Science?
http://www.elon.edu/e-
web/academics/elon college/computing sciences/curriculum/cs.xhtml
This is from the departmental website of a small colleges computing
sciences (note the name!) department. It discusses pretty much all the issues
weve been looking at: Is it a science? What does it study? Is it an engineering
discipline?
Shapiro, Stuart C. (2001), Computer Science
http://www.cse.buffalo.edu/shapiro/Courses/CSE115/notes2.html
Stairs, Allen (2014), Response to question about the definition of magic,
Tedre, Matti (2007), Lecture Notes in the Philosophy of Computer Science, esp.
Introduction to the Course
(http://cs.joensuu.fi/mmeri/teaching/2006/philcs/files/lecture notes1.pdf),
Part I: Struggling for Status
(http://cs.joensuu.fi/mmeri/teaching/2006/philcs/files/lecture notes2.pdf),
and Part II: Emerging Interdisciplinarity

(http://cs.joensuu.fi/mmeri/teaching/2006/philcs/files/lecture notes3.pdf)
(all accessed 3 July 2013).
Tedre, Matti; & Sutinen, Erkki (2008), Three Traditions of Computing: What
Educators Should Know, Computer Science Education 18(3) (September): 153
170.
Figure 3.4: http://abstrusegoose.com/215

Figure 3.5: http://xkcd.com/894/

Figure 3.6: Outline of Positions on What CS Is; if not legible, please contact author
Chapter 4
What Is Science?
Version of 13 May 2015
The most remarkable discovery made by scientists is science itself. The

discovery must be compared in importance with the invention of cave-
painting and of writing. Like these earlier human creations, science
is an attempt to control our surroundings by entering into them and
understanding them from inside. And like them, science has surely made a
critical step in human development which cannot be reversed. We cannot
conceive a future society without science.
Jacob Bronowski, Scientific American (1958)
Science is the field in which rationality reigns.

John Worrall [Worrall, 1989, 109]
[A] scientific education is, to a considerable extent, an exercise in taming

the authority of ones intuition.
H. David Orr [Orr, 2013, 27]
149
150 CHAPTER 4. WHAT IS SCIENCE?

In doing these readings, remember that our ultimate question is whether computer science is a
science.
1. Required:
Kemeny, John G. (1959), A Philosopher Looks at Science (Princeton: D. van

Nostrand).
Introduction, pp. ixxii
Ch. 5, The [Scientific] Method, pp. 85105.
You can skim Ch. 10, What Is Science?, pp. 174183, because his answer is
just this: A science is any study that follows the scientific method.
Kolak, Daniel; Hirstein, William; Mandik, Peter; & Waskan, Jonathan (2006),
Cognitive Science: An Introduction to Mind and Brain (New York: Routledge),
4.4.2. The Philosophy of Science
Papineau, David (1996), Philosophy of Science, in Nicholas Bunnin & E.P. Tsui-
James (eds.), The Blackwell Companion to Philosophy (Oxford: Blackwell): 290
324.
1.11.3 (The Problem of Induction, Poppers Falsificationism, and The
Failings of Falsificationism), pp. 290294
1.52.1 (Instrumentalism versus Realism, Theory, Observation and
Incommensurability, The Underdetermination of Theory by Evidence,
The Pessimistic Meta-Induction, Naturalized Epistemology of Science,
Causation), pp. 298310
Skim the rest.
Quine, Willard van Orman (1951), Two Dogmas of Empiricism, Philosophical

Review 60: 2043, http://www.ditext.com/quine/quine.html
Popper, Karl R. (1953), Science: Conjectures and Refutations, in Karl R. Popper,
Conjectures and Refutations: The Growth of Scientific Knowledge (New York:
Harper & Row, 1962).
Kuhn, Thomas S. (1962), The Structure of Scientific Revolutions (Chicago:
University of Chicago Press), Ch. IX, The Nature and Necessity of Scientific
Revolutions; online at:
http://www.marxists.org/reference/subject/philosophy/works/us/kuhn.htm
Hempel, Carl G. (1966), Philosophy of Natural Science (Englewood Cliffs, NJ:

Prentice-Hall), Scope and Aim of this Book.
On empirical vs. non-empirical sciences.
Kyburg, Henry E., Jr. (1968), Philosophy of Science: A Formal Approach (New
York: Macmillan), Ch. 1, The Concept of a Formal System.
4.1. RECOMMENDED READINGS: 151
Ziman, John M. (1968), What Is Science?, in John M. Ziman, Science Is Public

Knowledge (Cambridge, UK: Cambridge University Press), reprinted in Alex C.
Michalos (ed.) Philosophical Problems of Science and Technology (Boston: Allyn
& Bacon, 1974): 127.
Salmon, Wesley C. (1984), Scientific Explanation and the Causal Structure of
the World (Princeton: Princeton University Press), Ch. 1, Scientific Explanation:
Three General Conceptions.
4. Recommended:
Okasha, Samir (2002), Philosophy of Science: A Very Short Introduction (Oxford:
Oxford University Press).
This is the very best introduction to philosophy of science, although its an
entire (but short) book. You may read it instead of any of the above.
4.2 Introduction
We began by asking, What is computer science?, and we have seen that one answer
is that it is a science (or that parts of it are science). Some say that it is a science
of computers, some that it is a science of algorithms or procedures, some that it is a
science of information processing. And, of course, some say that it is not a science
at all, but that it is a branch of engineering; we will explore that option in Chapter 5.
In the present chapter, we will explore what it means to be a science, so that we can
decide whether computer science is one (or whether parts of it are).
In keeping with my definition of philosophy as the personal search for truth
by rational means, I wont necessarily answer the question, Is computer science a
science?. But I will provide considerations to help you find and defend an answer that
you like. It is more important for you to determine an answer for yourself than it is for
me to present you with my view; this is part of what it means to do philosophy in the
first person for the first person. And it is very important for you to be able to defend
your answer; this is part of what it means to be rational (it is the view that philosophy
is intimately related to critical thinking).
The word science originally simply meant knowledge or knowing; it derives
from the Latin verb scire, which meant to know1 But, of course, it has come to mean
much more than that.
Lets begin by contrasting the term science with some other terms. First, science
is sometimes opposed to art, not in the sense of the fine arts (such as painting, music,
and so on) but in the sense of an informal body of experiential knowledge, or tricks of
the trade: information that is the result of personal experience, perhaps unanalyzable
(or, at least, unanalyzed), and creative. This is art in the sense of the art of cooking.
By contrast, science is formal, objective, and systematic; scientific knowledge is the
result of analysis.
This contrast can be seen in the titles of two classic texts in computer science:
Donald Knuths The Art of Computer Programming [Knuth, 1973] and David Griess
The Science of Programming [Gries, 1981]. The former is a (multi-volume) handbook
of different techniques, catalogued by type, but analyzed (albeit incompletely by
todays standards). The latter is a compendium of formal methods for program
development and verification, an application of logic to programming. (For a detailed
defense of the title of Knuths work, see [Knuth, 1974a].)
Second, as we have seen, science is also sometimes opposed to engineering.
Because this will be our focus in Chapter 5, I wont say more about it here.
Finally, science is opposed (in more ways than one) to pseudo-science (such
as astrology). We will return to this in 4.7.1, because to explain the contrast
between science and pseudo-science (disciplines that masquerade as science, but are
not science) is part of the philosophical exploration of what science is.
One might think that the philosophy of science would be the place to go to find out
what science is, but philosophers of science these days seem to be more interested in
questions such as the following (the first two of which are the closest to our question):
1 According to the Oxford English Dictionary, http://www.oed.com/view/Entry/172672 (accessed 16 July
2013).
4.3. EARLY MODERN SCIENCE 153
What is a scientific theory? (Here, the emphasis is on the meaning of the term
theory.)
What is scientific explanation? (Here, the emphasis is on the meaning of the
term explanation.)
What is the role of probability in science?
What is the nature of induction? (Why) will the future resemble the past?
What is a theoretical term? (That is, what do the terms of (scientific) theories
mean? Do they necessarily refer to something in the real world? For example,
there used to be a scientific concept in the theory of heat called phlogiston, but
we no longer think that this term refers to anything.)
How do scientific theories change? When they do, are their terms
commensurable: that is, do they mean the same thing in different theories?
(For example, what is the relationship between phlogiston and heat? Does
atom, as used in ancient Greek physics, or even 19th-century physics, mean
the same as atom as used in 21st-century physics?)
Are scientific theories realistic (do they attempt to describe the world?) or
merely instrumental (are they just very good predicting-devices that dont
necessarily bear any obvious resemblance to reality, as sometimes seems to be
the case with our best current theory of physics, namely, quantum mechanics)?
And so on.
These are all interesting and important questions, and it is likely that a good answer
to our question, What is science?, will depend on answers to all of these. If so,
then a full answer will be well beyond our present scope, and the interested reader is
urged to explore a good book on the philosophy of science (such as those listed in the
Recommended and Further Readings sections of this chapter). Here, we will only be
able to consider a few of these questions.
4.3 Early Modern Science

Francis Bacon (15611626), a contemporary of Shakespeare, who lived about 400
years ago, devised one of the first scientific methods. He introduced science as a
systematic study. (So, when you read about computer scientists who call computer
science a study rather than a science, maybe they are not trying to deny that
computer science is a science but are merely using a euphemism.)
To study something, X, systematically is:
to find positive and negative instances of Xto find things are are Xs and things
that are not Xs;
to make changes in Xs or their environment (that is, to do experiments);
to observe Xs and to observe the effects of experiments performed with them;
to find correlations between Xs, their behavior, and various aspects of their
environment.
One important question in the history of science has concerned the nature of these
correlations. Are they (merely) descriptions, or are they explanations? In other words,
is the goal of science to describe the world, or is it to explain the world?
4.4 The Goals of Science

4.4.1 Description as the Goal of Science
Ernst Mach (18381916) was a physicist and philosopher of science who lived about
130 years ago, at the time when the atomic theory was being developed. He was
influenced by Einsteins theory of relativity and is probably most famous for having
investigated the speed of sound (which is now measured in Mach numbers, Mach 1
being the speed of sound).
For Mach, the goal of science was to discover regular patterns among our sensations
in order to enable the prediction of future sensations. Thus, Mach was a realist,
because, in order to discover something, it must really be there. Perhaps it seems odd
to you that a physicist would be interested in our sensations rather than in the world
outside of our sensations. This makes it sound as if science should be done in the
first person, for the first person, just like philosophy! Thats almost correct; many
philosophically oriented scientists at the turn of the last century believed that science
should begin with observations, and what are observations but our sensations? The
great 18th-century philosopher Immanual Kant distinguished between what he called
noumena (or things in themselves) and what he called phenomena (or things
as we perceive and conceive them); he claimed that we could only have knowledge
about phenomena, not noumena, because we could not get outside of our first-person,
subjective ways of conceiving and perceiving the world.
For Mach, scientific theories are (merely) shorthand descriptions of how the world
appears to us. According to the philosophy of science known as physicalism,
sensory perception yields reliable (but corrigible)2 knowledge of ordinary, medium-
sized physical objects and events. For Mach, because atoms were not observable, there
was no reason to think that they exist.
4.4.2 Explanation as the Goal of Science

By contrast, the atomic theory was an attempt to explain why the physical world
appears the way it does. Such a goal for science is to devise theories that explain
observed behavior. Such theories are not merely summaries of our observations, but
go beyond our observations to include terms that refer to things (like atoms) that might
not be able to be observed (yet).
Here is a question for you to think about: Are some computer programs theories?
In particular, consider an AI program that allows a robot to see or to use natural
2 I.e., correctable.
4.5. INSTRUMENTALISM VS. REALISM 155
language. Does such a program constitute a psychological (hence scientific) theory

of vision or language? If so, would it be a descriptive theory or an explanatory one?
(Well return to this topic in Ch. 15.)
4.5 Instrumentalism vs. Realism
Suppose we have an explanatory scientific theory of something, say, atomic theory.

Such theories, as we have seen, often include unobservablesterms referring to
things that we have not (yet) observed but whose existence would help explain things
that we have observed. One way of looking at this is to think of an experiment as taking
some input (perhaps some change deliberately made to some entity being studied) and
observing what happens after the experiment is overthe output of the experiment.
Between the input and the output, something happens, but we dont necessarily know
what it is. It is as if what we are studying is a black box, and all we can observe are
its inputs and outputs. A scientific theory (or, for that matter, a computer algorithm!)
can be viewed as an explanation of what is going on inside the black box. (Can it
be viewed merely as a description of whats going on inside? Probably not, because
you can only describe what you can observe and, by hypothesis, we cant observe
whats going on inside the black box.) Such a theory will usually involve various
unobservables structured in various ways.
Do the unobservables that form part of such an explanatory theory exist? If you
answer yes, then you are a realist; otherwise, you are an instrumentalist. A realist
believes in the real existence of explanatory unobservables. An instrumentalist believes
that they are merely useful tools (or instruments).
In Machs time, it was not clear how to treat the atomic theory. Atoms were clearly
of instrumental value, but there was no observable evidence of their existence. But
they were so useful scientifically that it eventually became unreasonable to deny their
existence, and, eventually, they were observed.
In our time, quantum mechanics poses a similar problem. If the world really
is as quantum mechanics says that it is (and quantum mechanics is our best current
theory about how the world is), then the world is really weird. So, possibly quantum
mechanics is merely a useful calculating tool for scientific prediction and shouldnt be
taken literally as a description of the real world. (On the other hand, recent research in
cognitive science suggests that quantum-mechanical methods applied at the macrosopic
level might provide better explanations of certain psychological findings about human
cognition than more standard methods [Wang et al., 2013].)
The great 20th-century philosopher Willard van Orman Quine, in his classic paper
On What There Is [Quine, 1948] argued that to be is to be the value of a bound
variable. In other words, if your best theory talks about Xsthat is, postulates the
existence of Xsthen, according to that theory, Xs exist.
4.6 The Scientific Method

People often talk about the scientific method. There probably isnt any such thing;
rather, there are many scientific methods of studying something. But lets look at
one version of a scientific method, a version that is interesting in part because it was
described by the mathematician John Kemeny, who was also a computer scientist. (He
was the inventor of the BASIC computer programming language and helped develop
time sharing. (He also worked with Einstein and was president of Dartmouth College.)
His book, A Philosopher Looks at Science [Kemeny, 1959, Chs. 5, 10] presents the
scientific method as a cyclic procedure. Because cyclic procedures are called loops
in computer programming, I will present Kemenys version of the scientific method as
an algorithm that does not halt (an infinite loop):3
Algorithm Scientific-Method
begin
while there is a new fact to observe, do:
{i.e., repeat until there are no new facts to observe,
which will never happen, so we have permanent inquiry}
begin
1. observe things & events;
{Express these observations as statements about

particular objects a, b, . . . :
Pa Qa,
Pb Qb,
...
Note: observations may be theory-laden,

that is, based on assumptions}
2. induce general statements;
{x[Px Qx]}
3. deduce future observations; {prediction}
{Pc/ Qc}
4. verify predictions against observations;
{if Qc
then general statement confirmed
or consistent with theory
else revise theory (or. . . }
end
end.
3 Thisalgorithm is written in an informal pseudocode. Terms in boldface are control structures.
Expressions in {braces} are comments.
4.7. ALTERNATIVES TO THE SCIENTIFIC METHOD 157
Kemenys version of the scientific method is a cycle (or loop) consisting of

observations, followed by inductive inferences, followed by deductive predictions,
followed by verifications. The scientist begins by making individual observations of
specific objects and events, and expresses these in language: Object a is observed to
have property P, object a is observed to have property Q, object as having property P
is observed to precede object as having property Q, and so on. Next, the scientist
uses inductive inference to infer from a series of events of the form Pa precedes
Qa, Pb precedes Qb, etc., that whenever any object x has property P, it will also
have property Q. So, the scientist who observes that object c has property P will
deductively infer (that is, will predict) that object c will also have property Q before
observing it. The scientist will then perform an experiment to see whether Qc. If Qc
is observed, then the scientists theory that x[Px Qx] will be verified; otherwise,
the theory will need to be revised in some way. (Similarly, [Denning, 2005, 28]
says that The scientific paradigm. . . is the process of forming hypotheses and testing
them through experiments; successful hypotheses become models that explain and
predict phenomena in the world. He goes on to say, Computing science follows
this paradigm in studying information processes.)
For Kemeny, explanation in science works this way: An observation is explained by
means of a deduction from a theory. This is the philosopher Carl Hempels Deductive-
Nomological Theory of Explanation. It is deductive, because the statement that some
object c has property Q (Qc) is explained by showing that it can be validly deduced
from two premises: that c has property P (Pc) and that all Ps are Qs (x[Px Qx]).
And it is nomological, because the fact that all Ps are Qs is lawlike or necessary, not
accidental: Anything that is a P must be a Q. (This blending of induction and deduction
is a modern development; historically, Bacon (and other empiricists, chiefly in Great
Britain) emphasized experimental induction and probabilism, while Descartes (and
other rationalists, chiefly on the European continent) emphasized deduction and
logical certainty [Uglow, 2010, 31].)
Finally, according to Kemeny, any discipline that follows the scientific method is
a science. This rules out astrology, on the grounds that astrologers never verify their
predictions.4
4.7 Alternatives to The Scientific Method

This hypothetical-deductive method of formulating theoretical hypotheses and
testing their predictions abainst systematic observation and controlled experiment is
the classical, or popular, view of what science and the scientific method are. And, at
a sufficiently abstract level, all sciences count as using it [Williamson, 2011]. But
there are at least two other views of the nature of science that, while generally agreeing
on the distinctions between science as opposed to art, engineering, and pseudo-science,
differ on the nature of science itself.
4 Or that their predictions are so vague that they are always trivially verified. See 4.7.1, below.
4.7.1 Falsifiability
According to the philosopher Karl Popper (19021994), the scientific method (as
propounded by people like Kemeny) is a fiction. The real scientific method sees science
as a sequence of conjectures and refutations:
1. Conjecture a theory (to explain some phenomenon).
2. Compare its predictions with observations
(i.e., perform experiments to test the theory).
3. If an observation differs from a prediction,
then the theory is refuted (or falsified)
else the theory is confirmed.
There are two things to note about the last part of this: First, by falsifiable, Popper
meant something like capable of being falsified in principle, not capable of being
falsified with the techniques and tools that we now have available to us. Seond,
confirmed does not mean true! Rather, it means that the theory is not yet falsified!
This is because there might be some other explanation for the predicted observation.
Just because a theory T predicts that some observation O will be made, and that
observation is indeed made, it does not follow that the theory is true! In other words,
the argument:
T O
O
T
is an invalid argument (called the Fallacy of Affirming the Consequent). If T is
false, then the first premise is still true, so we could have true premises and a false
conclusion. This might also be called the fallacy of circumstantial evidence, where T
is the (possibly false) circumstantial evidence that could explain O, but there might be
other evidence that also explains O and which is true.
So, what is science according to Popper? A theory or statement is scientific if
and only if it can be falsified. This is why astrology (and any other superstition) is
only pseudo-science: It cannot be falsified, because it is too vague. (As the physicist
Freeman Dyson once wrote, Progress in science is often built on wrong theories that
are later corrected. It is better to be wrong than to be vague [Dyson, 2004, 16].)
For Popper, falsifiability also ruled out two other candidates for scientific theories:
Freudian psychotherapy and Marxism as an economic theory. The vaguer a statement
or theory is, the harder it is to falsify. When I was in college, one of my friends came
into my dorm room, all excited about an astrology book he had found that, he claimed,
was really accurate. He asked me what day I was born; I said September 30th. He
flipped the pages of his book, read a horoscope to me, and asked if it was accurate. I
said that it was. He then smirked and told me that he had read me a random horoscope,
for April 16th. The point was that the horoscope for April 16th was so vague that it
also applied to someone born on September 30th!
But is astrology really a Popperian pseudo-science? Although I agree that the
popular newspaper style of astrology no doubt is (on the grounds of vagueness),
4.7. ALTERNATIVES TO THE SCIENTIFIC METHOD 159
real astrology, which might be considerably less vague, might actually turn out to be
testable and, presumably, falsified, hence falsifiable. But that would make it scientific
(albeit false)!
It is worthwhile to explore the logic of falsifiability a bit more. Although the invalid
argument form above seems to describe what goes on, it can be made more detailed. It
is not the case that scientists deduce predictions from theories alone. There are usually
background beliefs that are independent of the theory being tested (for example, beliefs
about the precision of ones laboratory equipment). And one does not usually test a
complete theory T but merely one new hypothesis H that is being considered as an
addition to T . So it is not simply that the argument above should begin with a claim
that the theory T implies prediction O. Rather, theory T , conjoined with background
beliefs B, conjoined with the actual hypothesis H being tested is supposed to logically
predict that O will be observed.
(T & B & H) O
Suppose that O is not observed:
What follows from these two premises? By the rule of inference called modus tollens,
we can infer:
(T & B & H)
But from this, it follows (by DeMorgans Law) that:
T B H
That is, either T is false, or else B is false, or else H is false, or any combination of them
is false. What this means is that if you strongly believe in your hypothesis H that seems
to be inconsistent with your observation O, you do not need to give up H. Instead, you
could give up the rest of your theory T (or simply revise some part of it), or you could
give up (some part of) your background beliefs B (for example, you could blame your
measuring devices as being too inaccurate). As [Quine, 1951] pointed out, you could
even give up the laws of logic if the rest of your theory has been well confirmed; this
is close to the situation that obtains in contemporary quantum mechanics.
4.7.2 Scientific Revolutions

Thomas Kuhn (19221996), a historian of science, rejected both the classic scientific
method and Poppers falsifiability criterion (a plague on both your houses, he might
have said). Based on his studies of the history of science, Kuhn claimed that the real
scientific method works as follows [Kuhn, 1962, Ch. 9]:
1. There is a period of normal science, based on a paradigm, that is, on a

generally accepted theory. During that period of normal science, a Kemeny-like
(or Popper-like) scientific method is in operation.
2. If that theory is challenged often enough, there will be a revolution, and a new
theorya new paradigmwill be established, completely replacing the old one.
3. A new period of normal science follows, now based on the new paradigm, and
the cycle repeats.
[Dyson, 2004, 16] refers to the revolutionaries as those who build grand castles in
the air and to the normal scientists as conservatives. . . who prefer to lay one brick
at a time on solid ground.
The most celebrated example was the Copernican revolution in astronomy
[Kuhn, 1957]. Normal science at the time was based on Ptolemys paradigm of an
earth-centered theory of the solar system. But this was so inaccurate that its advocates
had to keep patching it up to make it consistent with observations. Copernicuss
new paradigmthe heliocentric theory that we now believeoverturned Ptolemys
paradigm. Other revolutions include those of Newton (who overthrew Aristotles
physics), Einstein (who overthrew Newtons), Darwin (whose theory of evolution
further demoted humans from the center of the universe), Watson and Crick (whose
discovery of the. . . structure of DNA. . . changed everything in biology [Brenner, 2012,
1427]), and Chomsky in linguistics (even though some linguists today think that
Chomsky was wrong [Boden, 2006]).
4.8 Is Computer Science a Science?

These are only three among many views of what science is. Is computer science a
science according to any of them? This is a question that I will leave to the reader to
ponder. But here are some things to consider:
Does computer science follow Kemenys scientific method? For that matter,
does any science (like physics, chemistry, or biology) really follow it? Does
every science follow it (what about astronomy or cosmology)? Or is it just an
idealized vision of what scientists are supposed to do?
Even if it is just an idealization, does computer science even come close? What
kinds of theories are there in computer science? How are they tested?
Are any computer-science theories ever refuted?
Have there been any revolutions in computer science? (Is there even a current
Kuhnian paradigm?)
And there are other considerations: What about disciplines like mathematics? Math
certainly seems to be scientific in some sense, but is it a science like physics or biology?
Is computer science, perhaps, more like math than like these (other) sciences? This
raises another question: Even if computer science is a science, what kind of science is
it?
4.9. WHAT KIND OF SCIENCE MIGHT COMPUTER SCIENCE BE? 161
4.9 What Kind of Science Might Computer Science Be?

Hempel once distinguished between two kinds of science: empirical sciences and non-
empirical sciences.
The former explore, describe, explain, and predict various occurrences in the world.
Such descriptions or explanations are empirical statements that need empirical (that is,
experimental) support. The empirical sciences include the natural sciences (physics,
chemistry, biology, some parts of psychology, etc.) and the social sciences (other parts
of psychology, sociology, anthropology, economics, political science, history, etc.).
Does computer science fit in here? (If so, where?)
The non-empirical sciences are logic and mathematics. Their non-empirical
statements dont need empirical support. Yet they are true of, and confirmed by,
empirical evidence (though exactly how and why this is the case is still a great mystery;
see [Wigner, 1960], [Hamming, 1980]).5 So, does computer science fit here, instead?
Computer science arose from logic and math (we will see how in Chapter 6,
when we look into the nature of computers and algorithms). But it also arose
from the development of the computer as a tool to solve logic and math problems.
This brings it into contact with the empirical world and empirical concerns such as
space and time limitations on computational efficiency (or complexity [Loui, 1996,
Aaronson, 2013]).
One possible way of adding computer science to Hempels taxonomy is to realize
that psychology doesnt neatly belong to just the natural or just the social sciences.
So, perhaps computer science doesnt neatly belong to just the empirical or just the
non-empirical sciences, but that parts of it belong to both. And it might even be the
case that the non-empirical aspects of computer science are not simply a third kind of
non-empirical science, on a par with logic and math, but are themselves parts of both
logic and of math.
Or it might be the case that we are barking up the wrong tree altogether. What if
computer science isnt a science at all? This possibility is what we now turn to.

1. Chams PhD cartoon about scientific method.
2. Calvin and Hobbes cartoon on horoscopes.
3. Natural science is knowledge about natural objects and phenomena.

We ask whether there cannot also be artificial scienceknowledge
about artificial objects and phenomena. Simon 1996:3
4. xkcd cartoon about difference between scientists and normal people.
5. discuss para on Galileo and Descartes in Dyson 2006, bottom of p2 to top of p4

5 A related question about whether scientists need to study mathematics at all has recently been discussed
in [Wilson and Frenkel, 2013].

6. Steedman 2008 has some interesting things to say on the differences between a
discipline such as physics and a discipline such as computer science (in general)
and computational linguistics (in particular), especially in 1, The Public Image
of a Science.
7. Petroski 2008: The first two sections argue that all scientists are sometimes
engineers and all engineers are sometimes scientists.
8. Vardi 2010 argues that computing (or computational science) is not a new kind
of science or a new way of doing science, but just a more efficient way of doing
the same kind of science that humans have always done.
9. Non Sequitur cartoon on pre-conceptual science.

10. Maybe put this in an exercises for the reader section at the end of the chapter:
Read [Johnson, 2001]. Then try to reconstruct and evaluate his argument for the
proposition that All Science Is Computer Science.
11. Perhaps science is merely any systematic activity, as opposed to a chaotic

one. There is a computer program called AlphaBaby, designed to protect
your computer from young children who want to play on your computer but
who might accidentally delete all of your files while randomly hitting keys.
AlphaBabys screen is blank; when a letter or numeral key is hit, a colorful
rendition of that letter or numeral apears on the screen, and when any other key
is hit, a geometric figure or a photograph appears. Most children hit the keys
randomly (chaotically) rather than systematically investigating which keys do
what (scientifically).
[Williamson, 2011] suggests something similar when he characterizes the

scientific spirit as emphasizing values like curiosity, honesty, accuracy,
precision and rigor.
12. Heres a simplfied way of thinking about what a scientific theory is: We can begin
by considering two things: the world and our beliefs about the world (or our
descriptions of the world). Those beliefs or descriptions are theoriestheories
about the world, about what the world is like. Such theories are scientific if
they can be tested by emmpirical or rational evidence, in order to see if they are
good beliefs or descriptions, that is, beliefs or descriptions that are true (that
correspond to what the world is really like). The testing can take one of two
forms: verification or refutation. A theory is verified if it can be shown that it is
consistent with the way the world really is. And a theory is refuted if it can be
shown that it is not the way the world really is. A picture can help:
Line Aa continuous lineis intended to represent the world, a continuum.

Line Ba line with gapsis intended to represent observations that we can
make about the world; some parts of the world we can observe, but others we
cant (those are the gaps). The solid lines in B represent things that we believe
about the world; the gaps represent things that we dont know (yet) about the
world. Line C is intended to represent a scientific theory about the world (about
line A); here, the gaps are filled in. Those fillings-in are predictions about what
the world is like at those locations where we cannot observe it; they are guesses
(hypotheses) about the world.
13. On art vs. science:
According to [Knuth, 1974a, 668], art in the non-painting sense once meant
something devised by mans intellect, as opposed to activities derived from
nature or instinct, as in the liberal arts (grammar, rhetoric, logic, arithmetic,
geometry, music, and astronomy), and it later came to be used. . . for the
application of knowledge (where science was used to stand for knowledge):
The situation was almost exactly like the way in which we now
distinguish between science and engineering.
But he later has something more interesting to suggest:
Science is knowledge which we understand so well that we can teach
it to a computer; and if we dont fully understand something, it is an
art to deal with it. Since the notion of an algorithm or a computer
program provides us with an extremely useful test for the depth of
our knowledge about any given subject, the process of going from an
art to a science means that we learn how to automate something.
(In [Knuth, 2001, 168], he adds this: Every time science advances, part of an art
becomes a science, so art loses a little bit. Yet, mysteriously, art always seems
to register a net gain, because as we understand more we invent more new things
that we cant explain to computers.)
14. Distinguish between the everyday sense of theory and the technical sense.
In the everyday sense, a theory is merely an idea which may or may not
have any evidence to support it. In the technical sense, a theory is a set
of statements that describe and explain some phenomenon, often formalized
mathematically or logically; to be scientific, such a theory must also be
accompanied by confirming evidence (and, if Popper is right, its statements must
be falsifiable). Note that both theory and theorem are etymologically related.
Anti-evolutionists (creationists and advocates of intelligent design) sometimes

criticize the scientific theory of evolution as merely a theory. Anyone who
does so is confusing the everyday sense (in which theory is opposed to fact)
with the technical senseevolution is a theory is the technical sense.
For a wonderful discussion of the nature of science with respect to evolution,

see the judges opinion in the celebrated legal case of Kitzmiller v. Dover Area
School District (especialy 4, Whether ID [Intelligent Design] Is Science),
online at:
http://ncse.com/files/pub/legal/kitzmiller/highlights/2005-12-
20 Kitzmiller decision.pdf
Also note: Evolution can be considered as automated design, instead of

intelligent design.
15. [A] science is an evolving, but never finished, interpretive system.

And fundamental to science. . . is its questioning of what it thinks
it knows. . . . Scientific knowledge. . . is a system for coming to an
understanding. [Barr, 1985]
Science is all about the fact that we dont know everything. Science
is the learning process. [Dunning, 2007]
[Science is not a collection of truths. It is a continuing exploration of

mysteries. [Dyson, 2011b, 10]
16. Francis Bacon. . .
told us to ask questions instead of proclaiming answers, to collect

evidence instead of rushing to judgment, to listen to the voice of
nature rather than to the voice of ancient wisdom. [Dyson, 2011a,
26]
17. In a commentary on a paper [Zhai and Kristensson, 2012] describing a

new technology for entering text on a computer, [Buxton, 2012] lauds the
combination of engineering and pure science involved in the develpment of
this new technology, and comments that it is an exemplary demonstration that
reinforces that computer science is a science, and that today the nature of that
science has a breadth that extends beyond its traditional roots. This, perhaps,
reinforces the portmanteau nature of computer science.
18. More on Bacon:
The concept of replicability in scientific research was laid out by

Francis Bacon in the early 17th century, and it remains the principal
epistemological tenet of modern science. Replicability begins with
the idea that science is not private; researchers who make claims must
allow others to test those claims. [Wainer, 2012, 358]
19. Besides the triumverate of Bacons scientific method, Poppers falsificationism,

and Kuhns paradigms and revolutions, there are other approaches to the nature
of science. For instance, the philosopher of science Michael Polanyi argued in
favor of science as being socially constructed, not purely rational or formal
(see [Kaiser, 2012] for an overview). And another philosopher of science, Paul
Feyerabend, also critiqued the rational view of science, from an anarchic point
of view (see [Feyerabend, 1975], [Preston, 2012], and compare similar remarks
in [Quillian, 1994, 2.2]). But exploration of alternatives such as these are
beyond our scope.
20. The history of science is partly the history of an idea that is by now
so familiar that it no longer astounds: the universe, including our
own existence, can be explained by the interations of little bits of
matter. We scientists are in the business of discovering the laws
that characterize this matter. We do so, to some extent at least, by
a kind of reduction. The stuff of biology, for instance, can be reduced
to chemistry and the stuff of chemistry cna be reduced to physics
[Orr, 2013, 26]
The reductionist picture can be extended at both ends of the spectrum that Orr
mentions: At one end, sociology can be reduced to psychology, and psychology
to biology; at the other, perhaps physics can be reduced to mathematics, and
mathematics to logic and set theory. CITE PHD CARTOON ON THIS. On
this view, its hard to see where computer science fits into this sequence. On
the other hand, not everyone thinks that this chain of reductions is legitimate
[Fodor, 1974]. For another thing, its not obvious that computer science is in
the business of discovering the laws that characterize. . . matter. We might try to
argue that the universe isnt made up of matter, but of information. Then, if you
are also willing to say that computer science is the science of information (or of
information processing), you could conclude that it is a science.
21. On instrumentalism:
Galileo. . .
. . . and the Church came to an implicit understanding: if he would

claim his work only as istoria, and not as dimonstrazione, the
Inquisitors would leave him alone. The Italian words convey the
same ideas as the English equivalents: a new story about the cosmos
to contemplate for pleasure is fine, a demonstration of the way things
work is not. You could calculate, consider, and even hypothesize with
Copernicus. You just couldnt believe in him. [Gopnik, 2013, 107]
Gopnik goes on to observe, with respect to the tension between empirical

observation and rational deduction:
Though Galileo. . . wants to convince. . . [readers] of the importance

of looking for yourself, he also want to convince them of the
importance of not looking for yourself. The Copernican system
is counterintuitive, he admitsthe earth certainly doesnt seem to
move. It takes intellectual courage to grasp the argument that it does.
[Gopnik, 2013, 107]
22. [Colburn, 2000, 168] draws an analogy between the scientific method of
formulating, testing, and (dis)confirming hypotheses and the problem-solving
method of computer science consisting of formulating, testing, and accepting-
or-rejecting an algorithm. Besides suggesting that computer science is (at least
in part) scientific, this analogizes algorithms to scientific hypotheses or theories.
(See Ch. 15 for further discussion.)
23. Question for end-of-chapter discussion:
Computer science is not the only alleged science whose scientific credentials
have been questioned. Even excepting such pseudosciences as astrology, the
disciplines of psychiatry and political science have both come under attack. At
the time of writing, economics is another such discipline whose status as a
science (even as a social science) has been doubted by some critics. Find some
essays that have been critical of the scientific status of one of these disciplines
(or of some other discipline) and consider whether the arguments that have been
used to justify or to challenge their status as a science can be applied to computer
science. (For economics, you might consider [Rosenberg, 1994], [Leiter, 2004],
[Leiter, 2005], [Leiter, 2009], [Chetty, 2013]. For psychiatry, take a look at the
writings of Adolf Grunbaum at: http://tinyurl.com/lsstp7a6
For astrology, [Thagard, 1978].)
24. On the pessimistic meta-induction that all statements
of science are false, see http://plato.stanford.edu/entries/structural-realism/ and
https://en.wikipedia.org/wiki/Pessimistic induction
25. Probably need to add a short section on structuralism, to complement sections
on realism and instrumentalism. Can then cite under Further Sources:
Floridi, Luciano (2004), Informational Realism, in J. Weckert & Y. Al-
Saggaf (eds.), Computing and Philosophy Conference (Canberra, Australia);
Conferences in Research and Pratice in Information Technology, Vol. 37: 7
12 (Australian Computer Society);
accessed 18 June 2014 from:
http://philsci-archive.pitt.edu/2538/1/ir.pdf
which discusses the relevance of object-oriented programming to structuralism.
@misc{pittphilsci2538,
month = {January},
6 http://grunbaum.pitt.edu/wp-content/plugins/downloads-
manager/upload/Medicine%20Bibliography%201-13.pdf
title = {Informational Realism},

author = {Luciano Floridi},
year = {2005},
note = {published in ACS Conferences in Research and
Practice in Information Technology (CRPIT), 2004, (3), pp. 7-12.},
keywords = {Structural realism, epistemic structural realism,
ontic structural realism, levels of abstraction, informational ontology,
object oriented programming},
url = {http://philsci-archive.pitt.edu/2538/},
abstract = {What is the ultimate nature of reality? This paper
defends an answer in terms of informational realism (IR). It does so in
three stages. First, it is shown that, within the debate about
structural realism (SR), epistemic (ESR) and ontic (OSR) structural
realism are reconcilable by using the methodology of the levels of
abstractions. It follows that OSR is defensible from a
structuralist-friendly position. Second, it is argued that OSR is also
plausible, because not all related objects are logically prior to all
relational structures. The relation of difference is at least as
fundamental as (because constitutive of) any relata. Third, it is
suggested that an ontology of structural objects for OSR can reasonably
be developed in terms of informational objects, and that Object Oriented
Programming provides a flexible and powerful methodology with which to
clarify and make precise the concept of ?informational object?. The
outcome is informational realism, the view that the world is the
totality of informational objects dynamically interacting with each
other.?}
}
26. [Cockshott and Michaelson, 2007, 2.5,

p. 235] suggest that the hypercomputation challenges to the CTCT are examples
of Kuhnian revolutionary paradigmatic challenges to the normal science of
computer science. (They dont think that the challenges are successful, however.)
27. At least one major philosopher has said something that suggests that engineering
might be a part of science:
I have never viewed prediction as the main purpose of science,

although it was probably the survival value of the primitive precursor
of science in prehistoric times. The main purposes of science are
understanding (of past as well as future), technology, and control of
the environment. [Quine, 1988]
If technology can be identified with engineering (and why shouldnt it be?),

then this puts engineering squarely into the science camp, rendering the science-
vs.-engineering debates moot (though still not eliminating the need to ask what
engineeringor technologyis).
28. In a word, the general public has trouble understanding the

provisionality of science. Provisionality refers to the state of
knowledge at a given time. Newtons laws of gravity, which we all
learn in school, were once thought to be complete and comprehensive.
Now we know that while those laws offer an accurate understanding
of how fast an apple falls from a tree or how friction helps us take
a curve in the road, they are inadequate to describe the motion of
subatomic particles or the flight of satellites in space. For these we
needed Einsteins new conceptions. [Natarajan, 2014, 64]
Einsteins theories did not refute Newtons; they simply absorbed

them into a more comprehensive theory of gravity and motion.
Newtons theory has its place and it offers an adequate and accurate
description, albeit in a limited sphere. As Einstein himself once put
it, The most beautiful fate of a physical theory is to point the way
to the establishment of a more inclusive theory, in which it lives as a
limiting case. It is this continuously evolving nature of knowledge
that makes science always provisional. [Natarajan, 2014, 65]
Somewhere, need to mention this issue of provisionality, maybe in connection
with Poppers falsificationism. Question to think about: In what sense, if
any, is computer science provisional? Nelson Pole has suggested (private
communication, 3/9/15) that if there is a bug lurking in every moderately
complex program, then all programs are provisional.
29. A great closing/summary quote:
Science, after all, is simply a logical, rational and careful examination
of the facts that nature presents to us. (The Amazing Randi, as quoted
in [Higginbothan, 2014, 53].) Although Shapiro would be happy with
the word nature here, others might not be, but I think that it can
be eliminated without loss of meaning and still apply to computer
science.
30. Comments on [Simon, 1996b]:
(a) A natural science is a body of knowledge about some class
of thingsobjects or phenomenain the world: about the
characteristics and properties that they have; about how they
behave and interact with each other.
The central task of a natural science is to make the wonderful
commonplace: to show that complexity, correctly viewed, is only
a mask for simplicity; to find pattern hidden in apparent chaos.
[Simon, 1996b, Ch. 1, p. 1]
The emphasis here is on natural, because he is ultimately going to discuss
sciences of the artificial. So, presumably, a science of the artificial would
be such a body of knowledge about some class of artificial things (as
opposed to things in the world). But, because artificial things are, surely,
things in the world (think of beehives and bird nests), I suspect that what
he means is things that are not naturally in the world: Note that the word
natural does not appear in his characterization of natural science. But
what difference is there between what might be called natural artifacts
such as beehives and birdnests, on the one hand, and artifacts such as
machines and computers, on the other? The only difference that I can see
is in who made them: non-humans vs. humans (and Simon says things that
are consistent with this (p. 2). In that case, a science of the artificial would
merely be a science of things made by humans. But it would still be a
science.
(b) For instance, he says that A forest may be a phenomenon of nature; a farm
certainly is not (p. 3). Granted, there might not have been any farms had
there not been humans (but what about ant farms?). But would he be
equally willing to say that a beehive may be a pheomenon of nature; an
log house certainly is not? Beehives are just as artificial or non-natural as
log houses, differing only in who made them. (Another, better?, example:
bird nests vs. shelters made of branches and leaves by some members of
Amazonian or sub-Saharan African peoples.)
(c) But Simon is aware of this; he goes on to say:
A plowed field is no more part of nature than an asphalted street
and no less [Simon, 1996b, 3, my italics].
And, indeed, he explicitly says:
But you will have to understand me as using artificial. . . as
meaning man-made as opposed to natural. [Simon, 1996b, 4].
I would only amend this slightly to read: human-made as opposed to non-
human-made
(d) I suppose that there are echoes here of the more general issue of the
generalization of terms, such as whether or not to apply thinking to
computers (in addition to humans) or whether to apply flying to airplanes
(in addition to birds) or even noodle to non-wheat products (see
http://www.askphilosophers.org/question/5738). (See also Simons
comments (pp. 1516) about the meaning of moonare artificial
satellites moons?)
(e) His description of the purpose of science is consistent both with Machs
view that science is (merely) a way to describe the world efficiently and
with the more modern view that its purpose is to explain or help us
understand the world. Note, too, the emphasis on complexity, which
elsewhere (WHERE?) we saw/will see that Simon believes is one of the
central roles of computer science.
(f) Nevertheless, he comes down later on the side of explanation:
This is the task of natural science: to show that the wonderful
is not incomprehensible, to show how it can be comprehended. . .
[Simon, 1996b, 1, my italics]
So, the task is not, pace Mach, merely to describe the complexity of the
world in simple terms, but to explain the world, to make it understandable.
(g) One of the general paradoxes of analysis or of explanation in general is that,
by showing how something mysterious or wonderful or complex is really
just a complex of things that are non-mysterious or mundane or simple, we
lose sight of the original thing that we were trying to understand or analyze.
(We saw/will see this again in Dennetts notion of Turings inversion.)
Simon demurs:
. . . the task of natural science. . . [is] to show how it [the
wonderful] can be comprehendedbut not to destroy wonder.
For when we have explained the wonderful, unmasked the hidden
pattern, a new wonder arises at how complexity was woven out of
simplicity. [Simon, 1996b, 12, my italics]
So, for instance, the fact (if it is a fact; we will explore this issue
later WHERE?) that non-cognitive computers can exhibit (or merely
simulate) cognitive behaviors is not necessarily a reductio of cognition
as computation but is itself something worthy of wonder and further
(scientific) explanation.
(h) He also goes on to suggest an analogy between the emergence of
complexity from simplicity and the emergence of a representational
picture or of a piece of music from partially concealed patterns,
presumably implemented by brush strokes or individual musical notes
although he seems to chalk this up to aesthetics (p. 2).
31. Science is what we understand well enough to explain to a computer;

art is everything else. [Knuth, 2001, 168]
Despite the sense of facetiousness, there is a serious point here: We must
understand something well enoughin full detailin order to express it in, or
model it with, a computer program.
Nelson Pole (private communication, 3/9/15) points out that programs that learn
need not begin by understanding things that well. I would respond by observing
that the programmer of the learning program might not have to understand the
topic to be learned that well, but would have to understand the nature of learning
well enough to write a learning program.
32. The goal of any scientific theory or model is to construct mechanisms

that account for observable phenomena which can be detected by our
sensory systems (possibly mediated by gauges of scientific apparati).
[Peschl and Scheutz, 2001a, 1]
One ambiguity in this characterization is whether the constructed mechanism
(which, of course, could be expressed as a computer program) is the one that
actually caused the phenomena that we observed, and not merely one that could
have caused it. In other words, we want our theores to be true. Many different
mechanisms might produce the same phenomena (think of certain kinds of

optical illusions or distorted images).
33. As we saw in Chs. 2 and 3, [McGinn, 2015b, 85] thinks that both philosophy
and computer science are sciences, just as physics and chemistry are. Unlike
the latter two, the formal sciences are not empirical in the usual sensethey
are not based on observation and experimentbut they are sufficiently rigorous,
organized, and systematic to qualify as science. To make his argument, he has
to come up with a characterization of science, but he is more concerned with a
broad, almost ordinary-language (or dictionary-definition) characterization than
a formal, philosophical analysis.
I take it that what distinguishes a discourse as scientific are such

traits as these: rigor, clarity, literalness, organization, generality
(laws or general principles), technicality, explicitness, public
criteria of evaluation, refutability, hypothesis testing, expansion
of common sense (withe the possibility of undermining common
sense), inaccessibility to the layman, theory construction, symbolic
articulation, axiomatic formulation, learned journals, rigorous and
lengthy education, professional societies, and a sense of apartness
from nave opinion [McGinn, 2015b, 86]
(Possibly add this article as required, or whatever, reading for this chapter.)
34. Put in this chapter, if not used in Ch. 3:

As we enter life, we all struggle to understand the world. Some
of us continue this struggle with dogged determination. These are
the scientists. Some of them realize that computation provides a
privileged perspective to understand the world outside and the one
within. These are the computer scientists. [Micali, 2015, 52].
35. On the nature of theories:
Theories are syntactic in nature:
When we say that theories are abstract, we mean that theoretical
constructs are typically phrased in a vocabulary that is different
from the vocabulary of the evidence that supports the theory.
[Gopnik, 1996, 3.1, p. 496]
First, theories are phrased in a vocabulary, hence syntactic. Second, this
vocabulary is different from the semantic interpretation of the theory.

Items in boldface are especially important, interesting, or useful.
Recommendations on texts. . . whose subject is the philosophy of science,
AskPhilosophers.org http://www.askphilosophers.org/question/1515
Arden, Bruce W. (1980), COSERS Overview, in Bruce W. Arden (ed.), What Can
(Cambridge, MA: MIT Press), Ch. 1, pp. 131; see esp. the section titled About
Names and Labels (pp. 57).
Burkholder, Leslie (1999), Are AI and Mechanics Empirical Disciplines?,
Journal of Experimental and Theoretical Artificial Intelligence 11: 497500;
DOI:10.1080/095281399146373
Discusses the difference between empirical experimental disciplines (like
mechanics, a branch of physics) and a priori disciplines (like mathematics).
Chaitin, Gregory J. (1968), On the Difficulty of Computations, in Gregory J.
Chaitin, Thinking about Godel and Turing: Essays on Complexity, 19702007
(World Scientific Publishing): 315; online at:
http://www.worldscientific.com/doi/suppl/10.1142/6536/suppl file/6536 chap01.pdf
See esp. 7, on classical problems of the methodology of science.
Corcoran, John (DATE), Scientific Revolutions, Encyclopedia of American
Philosophy.
Denning, Peter J. (2005), Is Computer Science Science?, Communications of the
ACM 48(4) (April): 2731.
Denning, Peter J.; Comer, Douglas E.; Gries, David; Mulder, Michael C.; Tucker,
Allen; Turner, A. Joe; & Young, Paul R. (1989), Computing as a Discipline,
Communications of the ACM 32(1) (January): 923.
Dyson, Freeman (2006, 19 October), Writing Natures Greatest Book, New York
Review of Books 53(16).
An interesting book review that discusses some of the origins of science. In
particular, take a look at fifth and sixth paragraphs, which discuss Galileo and
Descartes.
Fine, Arthur (1986), Unnatural Attitudes: Realist and Instrumentalist Attachments
to Science, Mind 95(378) (April): 149179
A state of the art survey article on realism vs. instrumentalism in science.
Gordin, Michael D. (2012), Separating the Pseudo from Science, The Chronicle
[of Higher Education] Review 59(4) (21 September): B10B12; online at:
http://chronicle.com/article/Separating-the-Pseudo-From/134412/
Hartmanis, Juris, & Lin, Herbert (eds.?) (1992), What Is Computer Science and

Hitchcock, Christopher (2007), Conceptual Analysis Naturalized:
A Metaphilosophical Case Study, Journal of Philosophy 103(9) (September):
427451.
I, Theories of Causation, pp. 429433, contains an excellent summary of
various contemporary and historical theories of the nature of cause and effect.
Lehoux, Daryn; & Foster, Jay (2012), A Revolution of Its Own, Science 338
(16 November): 885886.
Review of the 4th edition (2012) of [Kuhn, 1957].
Lloyd, Seth (2007), You Know too Much, Discover Magazine (28 April).
Most of our exploding collection of science information is junk.
Lohr, Steve (2006, 2 November), Group of University Researchers to Make Web
Science a Field of Study, New York Times: C6.
Is Web science a science?
How is it related to computer science?
Pigliucci, Massimo; & Boudry, Maarten (eds.) (2013), Philosophy of
Pseudoscience: Reconsidering the Demarcation Problem (Chicago: University of
Chicago Press).
The demarcation problem is the problem of determining the dividing line
between real science and pseudoscience; for example, almost everyone
will agree that astronomy is a real science and that astrology is not. But
what is the difference between real and pseudo sciences?
Also see:
(a) Pigliucci, Massimo; & Boudry, Maarten (2013), The Dangers of
Pseudoscience, Opinionator/The Stone, New York Times (10 October);
accessed 18 October 2013 from:
http://opinionator.blogs.nytimes.com/2013/10/10/the-dangers-of-
pseudoscience/ (accessed 18 October 2013)
(b) Curd, Martin (2014), Review of Pigliucci & Boudrys book, Notre Dame
Philosophical Reviews (22 July); accessed 21 July 2014 from:
https://ndpr.nd.edu/news/49425-philosophy-of-pseudoscience-
reconsidering-the-demarcation-problem/
Petroski, Henry (2008), Scientists as Inventors, American Scientist 96(5)
(September-October): 368371.
Popper, Karl R. (1959), The Logic of Scientific Discovery (New York: Harper &
Row).
Quillian, M. Ross (1994), A Content-Independent Explanation of Sciences
Effectiveness, Philosophy of Science 61(3) (September): 429448.
7 This book was the subject of a petition (online at:
http://www-formal.stanford.edu/jmc/petition/whysign/whysign.html (accessed 16 July 2013)) to withdraw it

from publication because we consider it misleading and even harmful as an agenda for future research.
An explanation, in terms of the communication of information, of why the

natural sciences are more effective than the social sciences, written by one
of the early researchers in AI who later became a sociologist of science.
Although written in the early days of the World Wide Web, this paper has some
interesting implications for the role of social media in political discourse.
Reed, Dan (2007), Computing: All Our Constants Are Variables, Computing
Research News 19(4) (September): 3.
Makes an interesting distinction between science as natural philosophy and
as scientia (= knowledge).
Rosenberg, Alex (2000), Philosophy of Science: A Contemporary Introduction
(London: Routledge).
Simon, Herbert A. (1996), The Sciences of the Artificial, 3rd edition (Cambridge,
MA: MIT Press).
Ch. 1, The Natural description Artificial Worlds, has sections on the nature
of understanding by simulating and on computers as artifacts, as abstract
objects, and as empirical objects.
Steedman, Mark (2008), On Becoming a Discipline, Computational Linguistics
34(1) (March): 137144.
Tedre, Matti (2011), Computing as a Science: A Survey of Competing
Viewpoints, Minds and Machines 21(3) (August): 361387.
Tedre, Matti; & Moisseinen, Nella (2014), Experiments in Computing:
A Survey, Scientific World Journal, Vol. 2014, Article ID 549398,
http://dx.doi.org/10.1155/2014/549398 (accessed 3 April 2015).
A survey of the nature of experiments in science, and whether computer
science is experimental in nature.
Thagard, Paul R. (1978), Why Astrology Is a Pseudoscience, PSA: Proceedings
of the Biennial Meeting of the Philosophy of Science Association: 223234.
Vardi, Moshe Y. (2010), Science Has Only Two Legs, Communications of the
ACM 53(9) (September): 5.
Weinberger, David (2012), Shift Happens, The Chronicle [of Higher Education]
Review (27 April): B6B9.
Wigner, Eugene (1960), The Unreasonable Effectiveness of Mathematics in
the Natural Sciences, Communications in Pure and Applied Mathematics 13(1)
(February); online at:
http://nedwww.ipac.caltech.edu/level5/March02/Wigner/Wigner.html
An interesting reply from a computer scientist:
Hamming, R.W. (1980), The Unreasonable Effectiveness of Mathematics,
American Mathematical Monthly 87(2) (February), online at:
http://www.dartmouth.edu/matc/MathDrama/reading/Hamming.html
Willingham, Daniel T. (2011), Trust Me, Im a Scientist, Scientific American
(May): 12; online at:
http://www.scientificamerican.com/article.cfm?id=trust-me-im-a-scientist
(accessed 23 July 2013)
Why so many people choose not to believe what scientists say

2. Websites:
Abelson, Harold, & Sussman, Gerald Jay, with Sussman, Julie (1996), Preface to
the First Edition, Structure and Interpretation of Computer Programs,
http://mitpress.mit.edu/sicp/full-text/book/book-Z-H-7.html#% chap Temp 4
Gopnik, Adam (2015), The Evolution Catechism, The New Yorker online,
http://www.newyorker.com/news/daily-comment/evolution-catechism (accessed
24 March 2015)
A discussion of the nature of theory as it is used in science and in ordinary
language.
Tedre, Matti (2007), Introduction to the Course, Lecture Notes in the Philosophy
of Computer Science,
http://cs.joensuu.fi/mmeri/teaching/2006/philcs/files/lecture notes1.pdf
Chapter 5
What Is Engineering?
Version of 11 February 2015
Figure 5.1: 2009

c by Universal UClick
177
178 CHAPTER 5. WHAT IS ENGINEERING?

In doing these readings, remember that our ultimate question is whether computer science is an
engineering discipline.
1. Required:
Davis, Michael (2009), Defining Engineering from Chicago to Shantou, The
Monist 92(3) (July): 325338.
OR [Davis, 1995a] or [Davis, 1996]?? (and move Davis 2009 to further
reading)
Petroski, Henry (2003), Early [Engineering] Education, American Scientist 91
(May-June): 206209.
Loui, Michael C. (1987), Computer Science Is an Engineering Discipline,
Engineering Education 78(3): 175178.
Davis, Michael (1998), Introduction to Engineering, Part I of Thinking Like an
Engineer: Studies in the Ethics of a Profession (New York: Oxford University
Press), esp.:
Introduction (p. 3)
Ch. 1, Science, Technology, and Values (pp. 317)
Who Is an Engineer? (from Ch. 2, pp. 2528)
Introduction to Ch. 3, Are Software Engineers Engineers? (pp. 3132)
The Standard Definition of Engineer (from Ch. 3, pp. 3234)
Three Mistakes about Engineering (from Ch. 3, pp. 3436)
Membership in the Profession of Engineering (from Ch. 3, pp. 3637)
Petroski, Henry (2008), Scientists as Inventors, American Scientist 96(5)
(September-October): 368371.
The first two sections argue that all scientists are sometimes engineers and all
engineers are sometimes scientists.
Brooks, Frederick P., Jr. (1996), The Computer Scientist as Toolsmith II,
5.2 Introduction
We began by asking what computer science is (Ch. 3), and we considered that it might
be what it says it is: a science; so we then looked at what science is (Ch. 4).
We also considered that computer science might be a branch of engineering; so
now it is time to ask what engineering is. What is its relationship to science? And what
is the relationship of computer science to engineering?
Answering these questions depends even more on what people say about it than
does answering the question, What is science?. This is because the philosophy of
engineering is much less well developed than the philosophy of science and, for some
reason, there seem to be fewer attempts to try to define engineering. For instance, if
you go to various university websites for schools or departments of engineering, you
will rarely find a definition. ([Koen, 1988, 307] makes a similar point.)
For engineering, more than for science or even computer science, it seems that
engineering is what engineers do, and so we may need to look at what it is that they
do. (As the computer scientist R.W. Hamming once pointed out [Hamming, 1968,
4], the only generally agreed upon definition of mathematics is Mathematics is
what mathematicians do, which is followed by Mathematicians are people who do
mathematics.
Dictionary definitions of engineering are even less helpful than usual. Actually,
dictionary definitions are rarely useful: First, dictionaries dont always agree, and
some are better than others. Second, dictionaries at best tell you how other people
use a term, but, if people use a term incorrectly, dictionaries are duty bound to record
that. Third, dictionaries can be misleading: Websters Ninth New Collegiate Dictionary
[Mish, 1983, 259] defines college as a body of clergy living together and supported
by a foundation! This may once have been true, and may even still be true in a very
limited sense of the term, but why is it listed as the first definition? Because Merriam-
Webster dictionaries list definitions in historical order!
[Mish, 1983] defines engineering as:
1. The activities or function of an engineer.
2. The application of science and mathematics to make matter and energy useful to
people.
The first of these cannot be understood without understanding engineer, which is

defined thus:
1. A member of a military group devoted to engineering work.
2. A person trained in engineering.
So this dictionary agrees explicitly with Hamming: engineering is what engineers

do; engineers are people who do engineering! Independently of the military group
condition, both of these definitions require us to already understand engineering.
Only the second definition of engineering holds out some hope for independent
understanding, but it is not at all clear that computer scientists apply science and math
to make matter and energy useful. Some might, by a stretch of meaning, but surely not
all.
It might help to look at the etymology of engineer. (It might, but it wont.) The
word comes from in- + gen- + -or, where in- means in, -or means, roughly, agent,
and gen- can mean mind, intellect, talent; to produce; etc., none of which seems to
helpthe word has simply changed too much for us to be able to figure out what it
means from its origins.
One philosopher who has tried to explain engineering [Bunge, 1974] does it this
way: There are two kinds of science (where science is any discipline that applies the
scientific method): pure and applied. Pure sciences apply the scientific method to
increasing our knowledge of reality (e.g., cell biology). Applied sciences apply the
scientific method to enhancing our welfare and power (e.g., cancer research). Among
the applied sciences are operations research (mathematics applied to management),
pharmacology (chemistry applied to biology), and engineering. Given this taxonomy,
computer science would not be a branch of engineering, though it might be an applied
science alongside engineering. Yet there is a pure component of computer science,
namely, the mathematical theory of algorithms, computability, and complexity (on the
first two, see Ch. 7; on the latter, see [Loui, 1996, Aaronson, 2013]).
5.3 A History of Engineering

Rather than treat software engineering as a subfield of computer science,
I treat it as an element of the set, {Civil Engineering, Mechanical
Engineering, Chemical Engineering, Electrical Engineering, . . . }. this is
not simply a game of academic taxonomy, in which we argue about the
parentage of ownership of the field; the important issue is the content and
style of the education. [Parnas, 1998, 1, my emphasis]
So let us turn to another philosopher for an insight as to what engineering might

be [Davis, 1998] (cf. [Davis, 1995a], [Davis, 1996], which contain earlier versions of
the same considerations). Davis begins with a history of engineering, which began
some 400 years ago in France, where there were enginesthat is, machinesand
engineers who worked with them, such as soldiers who used engines of war such
as catapults and artillery, or carpenters and stonemasons in civilian life who plied these
trades as soldiers. From this background comes the expression railroad engineer and
such institutions as the Army Corps of Engineers.
In 1676, the French army created a separate corps (separate from the infantry) of
engineers who were charged with military construction. So, at least in 17th-century
France, an engineer was someone who did whatever it was that those soldiers did.
Forty years later, in 1716, there were civil engineers: soldiers who built bridges, roads,
etc., for civilians.
A distinction was drawn between engineers and architects. The former were
trained in math and physics, and were concerned with reliability and other practical
matters. They were trained as army officers, hence more discliplined for larger projects.
Architects, on the other hand, were more like artists, chiefly concerned with aesthetics.
5.4. ENGINEERING AS DESIGN 181
Engineers in France at that time were trained at the Ecole Polytechnique

(Polytechnic School), whose curriculum began with a year of science and math,
followed gradually by more and more applications to construction (e.g., of roads),
culminating in a specialization.
So, at this time, engineering was the application of science for the use and
convenience of people and for improving the means of production [Davis, 1998].
Engineering was not science: Engineers use science but dont create new knowledge.
Nor is engineering applied science: Engineers are concerned with human welfare (and
not even with generality and precision), wheras applied scientists are concerned with
applying their scientific knowledge.
Davis concludes that engineering must be defined by its curriculum (by its specific
knowledge) and by a professional commitment to use that knowledge consistent with
an engineering code of ethics. So, rather than saying that engineering is what engineers
do, Davis says that engineering is what engineers learn and how they ought (ethically)
to use that knowledge. This, of course, raises the question: What is it that engineers
learn?
One question we might raise concerns the one branch of computer science that
has always explicitly called itself engineering: software engineering. Is software
engineering engineering? For Davis, it could be, if and only if there is a professional
curriculum for it, along with a code of professional ethics.
Another philosopher [Mitcham, 1994] echoes much of what Davis says. He notes
that engineer comes from the Latin ingenero, meaning to implant, generate, or
produce [Mitcham, 1994, 144]. And he rehearses the military history (albeit with
different datesCHECK THIS OUT). He also distinguishes between the engineer as
designer and the technician or technologist as builder. ELABORATE ON THIS.
5.4 Engineering as Design

An alternative view of engineering is due to Henry Petroski [Petroski, 2003, 206],
who says that its fundamental activity is design. So, engineering is not science,
whose fundamental activity is analysis [Petroski, 2003, 207]. But a science and an
engineering discipline can have the same object, can be about the same thing.
If so, then what is the common object of computer science and computer
engineering? Is it computers? Algorithms? Information? Perhaps computer science
studies algorithms and procedures, whereas computer engineering studies computers
and computer systems. If so, then who studies the relations between these?
One aspect of design has been picked up by [Hamming, 1968]. When one designs
something, one has to make choices. Hamming suggests that science is concerned
with what is possible while engineering is concerned with choosing, from among the
many possible ways, one that meets a number of often poorly stated economic and
practical objectives. This is interesting for a number of reasons.
First, it fits well with much of the workeven theoretical workthat is done by
computer scientists. As we saw earlier (3.6,3.8, one definition of computer science
is that it is concerned with what can be automated. One way of expressing this is
as follows: For what tasks can there be an algorithm that accomplishes it? But there
can be many algorithms all of which accomplish the exact same task. How can we
choose among them? We can ask which ones are more efficient: Which use less
memory (space)? Which take less time? So, in addition to asking what can be
automated, computer science also asks: What can be automated efficiently? If that is
computer engineering, so be it, but that would put one of the most abstract, theoretical,
and mathematical branches of computer sciencenamely, the theory of computational
complexitysmack dab in the middle of computer engineering, and that doesnt seem
correct.
5.5 A New Kind of Engineering?

One possible answer comes from the computer scientist Michael Loui [Loui, 1987].
According to Loui, computer science is not the study of computers: Contrary to
Newell, Perlis, and Simon, not all phenomena give rise to a science of that phenomenon
(there are toasters, but no science of toasters). In any case, any given computer will
soon be outdated.
Nor is computer science a branch of math, because math is declarative
(mathematical theorems are true propositions), whereas computer science is
procedural: Algorithms tell you how to do things. I dont think that this distinction is
quite as clearcut as Loui would make it out. For one thing, one of the earliest and most
successful branches of mathEuclids geometrywas procedural! [Toussaint, 1993]
A typical statement would be To construct an angle bisector with compass and
straightedge, do the following:. (And, just as there are theorems in computer science
about what cannot be computed, there are theorems in Euclidean geometry about what
cannot be constructed using only compass and straightedge, such as an angle trisector.)
Nor is computer science a branch of electrical engineering, because computers need
not be electronic. There can be mechanical computers, and researchers are studying
optical, quantum, and biological computers.1 (However, as my colleague Peter Scott,
formerly of our universitys electrical and computer engineering department, pointed
out to me, todays electrical engineering departments dont study just electronics, but
also optics.)
Rather, according to Loui, computer science is the theory, design, analysis, and
software- and hardware-implementation of algorithms for processing information,
where, by processing, he includes storing, transforming, retrieving, and transmitting.
Note that this is a very complicated definition, seemingly designed to make everyone
happy: Theres a little bit of engineering (design, hardware implementation), a little bit
of science (theory, analysis), and some information thrown in for good measure.
But, for Loui, this means that computer science is engineering, but a new kind of
engineering. Whereas classical engineering has a basis in science, computer science
has a basis in math (as [Hartmanis, 1993]? suggested, perhaps computer science is
mathematical engineering). Whereas classical engineering applies scientific principles
to design physical objects, computer science designs, not only physical computers, but
also non-physical programs. Finally, whereas classical engineering analyzes tradeoffs
1 The computing scientist could not care less about the specific technology that might be used to realize
machines, be it electronics, optics, pneumatics, or magic [Dijkstra, 1986], quoted in [Loui, 1987, 177].
5.6. THE ENGINEERING METHOD 183
such as gain vs. bandwidth (in electrical engineering) or stability vs. aesthetics (in civil
engineering), the tradeoffs that computer science deals with are of a different kind:
software vs. hardware, and space vs. time efficiency.
5.6 The Engineering Method

Just as there is a scientific method, some scholars have proposed an engineering
method. We saw in Chapter 4 one view of the scientific method, according to which
it is a loop that cycles through observation of facts, induction of general statements,
deduction of future observations, and verification of the deduced predictions against
observations, before cycling back to more observations.
Similarly, [Malpas, 2000] describes the engineering method both linearly and as a
cycle. It begins by inputting a set of requirements, followed by analysis, then synthesis,
then evaluation and execution, and outputting a solution. The cycle comes in between
the input and the output: The evaluation and execution cycles back both to the analysis
and to the synthesis, as well as adding to a knowledge base that, along with a set of
resources, interact with the analysis, synthesis and evaluation-execution.
But neither this nor the scientific method are carved in stone; they are more like
guidelines or even after-the-fact descriptions of behavior rather than rules that must be
slavishly followed.
Whether computer science is a science or an engineering discipline, it has to be the
science or engineering of something. We have seen at least two possibilities: It studies
computers, or it studies algorithms. To further explore which of these might be central
to computer science, let us begin by asking, What is a computer? Later, we will
inquire into what an algorithm is.

1. It is generally recognized...that engineering is the art or science of
making practical application of the knowledge of pure sciences....
The engineer uses the logic of science to achieve practical results.
(pp. x-xi.) of Florman 1994
2. Here, in a mere 66 words, [former President Herbert] Hoover captured

the essence of engineering and its contributions to society and
culture:
Engineering is a great profession. There is the fascination
of watching a figment of the imagination emerge through
the aid of science to a plan on paper. Then it moves to
realization in stone or metal or energy. Then it brings jobs
and homes to men. Then it elevates the standards of living
and adds to the comforts of life. That is the engineers high
privilege. (p.304 of Petroski 2003)
3. Perhaps end this chapter, before going on to the other topics in PHICS, with
Wulfs quote from 38
4. On contrasts between science and engineering:
In a scientific discipline you discover how things work;

in an engineering discipline you learn how to build things.
[Abrahams, 1987, 472]
From this, Abrahams concludes that Computer science is both a scientific

discipline and an engineering discipline. The conclusion appears to be valid.
Is the argument sound? The premises seem to be true. Yet one way to discover
how things work is to try to build them; so, is all engineering a kind of science?
But is the first premise really true? Is life, or the universe, a thing? Do
scientists really try to learn how the kinds of physical objects that engineers
build work (and nothing else)? This seems overly simplistic. Nevertheless, this
analytic vs. synthetic distinction (that is, a distinction between analyzing
taking something apartin order to learn how it works, on the one hand, and
synthesizingputting things togetherin order to build something, on the other
hand) seems to be a feature of many characterizations of science vs. engineering.
5. Characteristics of engineering:
(a) Engineering emphasizes what can exist, as opposed to science

(especially physics), which studies what does exist [Loui, 1987, 176]. At
first glance, this appears odd: One might think that the sciences are more
concerned with abstract possibilities, whereas engineering would be more
concerned with nitty gritty reality. But another way to understand this
distinction is that the sciences are only concerned with things that currently
exist, whereas engineering is (also) concerned with creating new things that
couldbut do not yetexist.
(b) Computer science has all the significant aspects of engineering: a
scientific basis, application of principles in design, analysis of trade-
offs, and heuristics and techniques [Loui, 1987, 176]. This suggests the
following argument:
i. Any engineering discipline has a scientific basis, etc.
ii. Computer science has a scientific basis, etc.
iii. computer science is an engineering discipline.
But this is invalid, having the form All Ps are Qs; c is Q; c is P, which
is a variety of the fallacy of affirming the consequent. (Compare: All dogs
have 4 legs; my cat has 4 legs; , my cat is a dog.) Louis second aspect
ties in with the first characteristic in our present list; he elaborates it as
follows: The ultimate goal of an engineering project is a product. . . that
benefits society [Loui, 1987, 176], giving bridges and computer programs
as sample products. But not all computer programs benefit society (think
of computer viruses); presumably, Loui meant something like product that
is intended to benefit society. But does that mean, then, that a computer
programmer who writes a spreadsheet program is an engineer (no matter
how sloppily the programmer writes it), whereas a computer programmer
who writes a computer virus is not an engineer (even if the program was
designed according to the best software engineering principles)?
6. Notes on [Koen, 1988]:
Rather than defining engineering, Koen defines the engineering method;

presumably, just as science can be defined as any discipline that follows
the scientific method, so engineering can be defined as any discipline
that follows the engineering method. Here is his definition:
The engineering method is the use of engineering heuristics to
cause the best change in a poorly understood situation within the
available resources [Koen, 1988, 308].
He then spends a lot of time defining heuristics, which well turn to in
a moment. What he doesnt seem to define is the adjective modifying
heuristics, namely, engineering! Presumably, the engineering method
is more specific than merely the general use of heuristics. So, what makes
a heuristic an engineering heuristic? Without this, his definition isnt very
useful for us.
On heuristic:
A heuristic is anything that provides a plausible aid or direction in
the solution of a problem but is in the final analysis . . . incapable
of justification and fallible. It is anything that is used to guide,
discover and reveal a possible, but not necessarily, correct way
to solve a problem. Though difficult to define, a heuristic has
four characteristics that make it easy to recognize: it does not
guarantee a solution; it may contradict other heuristics; it reduces
the search time for solving a problem; and its acceptance depends
on the immediate context instead of on an absolute standard.
[Koen, 1988, 308].
On this definition of heuristic, the first two characteristics differentiate
the use of them from science and mathematics, according to Koen. So,
they demarcate engineering from science and math. The third and fourht
characteristics make their use more practical than at least some scientific or
mathematical theories.
In a somewhat puzzling passage on p. 309, Koen seems to suggest that
engineering simply is identical with the use of heuristics, or, perhaps, with
the use of heuristics for causing desirable change in an unknown situation
within the available resources. If this interpretation is correct, then part of
what Koen is saying that what engineers do is to cause changes. This does
contrast with science (and math), whose goal is, presumably, to understood
things, not to change them.2

Much of what Koen says suggests that the kind of heuristic reasoning used
by engineers is akin to what Herbert Simon called bounded rationality
and satisficing (which we discussed briefly in 2.6.1.4): being satisfied
with having a reasonable answer to a question rather than the correct
one. Bounded rationality is necessary in practical situations, given limits
(bounds) on our time and knowledge.
7. According to the National Research Councils Committee on the Education and

Utilization of the Engineer, engineering is, by their definition,
Business, government, academic, or individual efforts in which

knowledge of mathematical and/or natural sciences is employed in
research, development, design, maufacturing, systems engineering,
or technical opeations with the objective of creating and/or delivering
systems, products, processes, and/or services of a technical nature and
content intended for use. [Florman, 1994, 174175]
Even Florman admits that this is a mouthful! Perhaps it can be simplified to

something like this: Efforts in which math and natural science are used in order
to produce something useful. If so, then engineering is (simply?) applied science.
And [Davis, 1996] points out that this definition, because of its vagueness (the
overuse of and/or), includes too much (such as accountants, because they use
mathematics). He does say that it emphasizes three important elements of
engineering: the centrality of math and science, the concern with the physical
world (which might, therefore, rule out software; but see 12.3 on that topic), and
the fact that unlike science, engineering does not seek to understand the world
but to remake it. But he goes on to say that those three elements. . . do not
define engineering. So, it would seem that they are necessary but not sufficient
conditions.
8. [Florman, 1994, 175] and many others [Davis, 1995a] also cite an 1828
definition due to one Thomas Tredgold: Engineering is the art of directing the
great sources of power in nature for the use and convenience of man [sic].
9. Florman prefers this as his characterization, not of engineering, but of the

engineering view:
a commitment to science and to the values that science demands

independence and originality, dissent and freedom and tolerance; a
comfortable familiarity with the forces that prevail in the physical
universe; a belief in hard work, not for its own sake, but in
the quest for knowledge and understanding and in the pursuit of
excellence; a willingness to forgo perfection, recognizing that we
have to get real and useful products out the door; a willingness
2 Compare Ch. 2, 2.6.4, where we saw Marx saying that philosophers should change the world, not
merely understand it. Does that mean that Marx was proposing a disclipline of philosophical engineering?
to accept responsiblity and risk failure; a resolve to be dependable;

a commitment to social order, along with a strong affinity for
democracy; a seriousness that we hope will not become glumness;
a passion for creativity, a compulsion to tinker, and a zest for change.
[Florman, 1994, 183]
There are clear echoes of Koens emphasis on the role of heuristics, a recognition
of the foundational role of science, andperhaps of most significance for our
purposesan emphasis on products.
So, if the central feature of engineering is, lets say, the application of scientific
(and mathematical) techniques for producing or building something, then surely
part of computer science is engineering: especially those parts that build
computers and that write programs. Heres something to think about: Just as
(some) computer scientists write programs, so journalists and novelists write
essays. Are journalists and novelists therefore engineers? Their products are not
typically application of science and math, so perhaps they arent. But might they
not be considered to be, say, language engineers?
10. Computer science is engineering, because: writing a computer program is giving

instructions how to do something, that is, how to build something. On the other
hand computer science could also be considered as the science of engineering,
because it has theoretical aspects that describe what it means to give instructions
for how to build something.
11. The goal of engineering is the production of useful things,

economically, for the benefit of society. [Loui, 1995, 31]
Among the things that (some) engineers produce are (chemical) processes,
which are similar to a computational algorithm, which is a method for
processing information [Loui, 1995, 31]. So, because computer science
produces useful things (including algorithms, but also microprocessors,
software, and databases), economically, for the benefit of society, it is a kind
of engineering.
12. According to Michael Davis, scientists are primarily interested in discovering

new knowledge and only secondarily, if at all, interested in inventing
something useful, whereas engineers are primarily interested in the latter and
hardly at all in the former [Davis, 1995a, 4243] (possibly repeated verbatim in
[Davis, 1998]; CHECK THIS)
13. Question for the reader to think about: In Ch. 3, we saw that Brooks argued that
computer science was not a science, but a branch of engineering, in part because
the purpose of engineering is to build things, and that thats what computer
scientists do. How would you evaluate this argument now that you have thought
more deeply about what engineering is?
14. On software engineering:

In addition to the question of whether computer science is a kind of engineering,
there is the question of the nature of software engineering. Computer scientists

(whether or not they consider themselves to be scientists or engineers) often
consider software engineering as a branch of computer science: Courses in
software engineering are often, perhaps even usually, taught in computer-science
departments. [Parnas, 1998] argues that software engineering is a branch of
engineering, not of computer science (nor of science more generally). In fact, he
says, Just as the scientific basis of electrical engineering is primarily physics,
the scientific basis of software engineering is primarily computer science
[Parnas, 1998, 2].
There are two interesting implications of this. First, it suggests that Parnas
views computer science as a science, because he takes it to be the scientific basis
of a branch of engineering. Second, this view of things is inconsistent with the
view advocated by, for instance, Loui and Hartmanis, who take computer science
(or parts of it) as being a kind of engineering whose scientific basis is primarily
mathematics! (On the other hand, one might argue that if software engineering
is based on computer science, which, in turn, is based on mathematics, then
software engineering must ultimately be based on mathematics, too. And thats
not far from the truth, considering that much of formal software engineering
(such as the formal analysis of computer programs and their development, or the
use of program-verification methods in the development of programs; see, for
example, [Mili et al., 1986] and our discussion of program verification in Ch. 16)
is based on (discrete) mathematics and logic.
Parnas says that engineers are professionals whose education prepares
them to use mathematics, science, and the technology of the day, to build
products that are important to the safety and well-being of the public
[Parnas, 1998, 2]. This echoes Daviss claim about the central role of education
in the nature of being an engineer, as well as Brookss (and others) claim about
the purpose of engineering to build humanly useful things.
15. Even if computer science qua engineering is based on mathematics and logic
as its underlying science, there is still a difference between more classic
engineering disciplines and computer science, because, historically, computers
(which, if anything, would seem to fall clearly on the engineering side of
the spectrum) were developedengineered, if you willwithout very much
benefit of mathematical or logical analysis. It took von Neumann to see that the
engineering accomplishments of Eckert and Mauchly (the developers of ENIAC)
could be combined with the mathematical insights of Turing [Davis, 2012,
Introduction]. We will look into this more closely when we investigate the
historical answer to the question of what a computer is (in Ch. 6).
16. Engineering is the knowledge required, and the process applied,

to conceive, design, make, build, operate, sustain, recycle or retire,
something of significant technical content for a specified purpose;
a concept, a model, a product, a device, a process a system, a
technology. [Malpas, 2000, 31]
This is another definition-by-committee (note the lists of verbs and nouns). But
it comes down to much the same thing as others have said: designing or building
useful things. It emphasizes two aspects to this: One is that the design or
activity of building must be knowledge-based; this presumably rules out design
or building that is based, not on scientific knowledge, but on experience alone.
The other aspect is that engineering is a process, in the sense of knowing
how to do something [Malpas, 2000, 5]; this has an algorithmic flairafter
all, algorithms are methods of describing how to do something.
17. [Petroski, 2005, 304] notes that we speak of the sciences in the plural (as we
do of the humanities), but of engineering in the singular, even though there
are many engineerings. Moreover, engineering. . . is an activity that creates
things. Note two points: First, it is creative; this is related to claims about
engineering as designing and building things. But, second, it is an activity, even
grammatically: The word is a gerund, which expresses. . . action. Is science
also an activity (or is engineering different from science in this respect)? Insofar
as science is an activity, is an activity that produces knowledge. Engineering
is an activity that uses that scientific knowledge to design and build artifacts.
18. [Petroski, 2006, 304] quotes former US president Herbert Hoover, who was
an engineer by profession, as having captured the essence of engineering as
beginning with. . .
a figment of the imagination [that] emerge[s] through the aid of

science to a plan on paper. Then it moves to realization in stone or
metal or energy. Then it bring jobs and homes to men [sic]. Then it
elevates the standards of living and adds to the comforts of life.
This rehearses the design and build useful things mantra of engineering, but
it also highlights the idea of engineering as implementing an abstract idea. For
more on implementation and its relationship to computer science, see Ch. 14.
19. In a follow-up article to [Davis, 1996], Michael Davis cites four different
conceptions of engineering:
(a) engineering as tending engines, e.g., railroad engineers, building-

superindent engineers; clearly, neither computer scientists nor software
engineers are engineers in this sense, but neither are electrical, civil,
mechanical, or chemical engineers engineers in this sense [Davis, 2011,
32];
(b) engineers in the functional sense, engineering-as-invention-of-useful-
objects [Davis, 2011, 32]. He criticizes this sense as both too broad
(including architects and accountants) and anachronistic (applying to
inventors of useful objects before 1700, which is about when the modern
sense of engineer came into use). Note that this seems to be the sense of
engineering used by many who argue that computer science is engineering
(who view engineering as designing and building useful artifacts).
(c) engineering-as-discipline [Davis, 2011, 3233]. Here, the issue concerns

the body of knowledge engineers are supposed to learn, which includes
courses concerned with the material world, such as chemistry and
statistics. Again, this would seem to rule out both computer science and
software engineering, on the grounds that neither needs to know any of the
material natural sciences like chemistry or physics (though both software
engineers and computer scientists probably need some statistics) and both
need to know things other engineers do not.
(d) engineering-as-profession [Davis, 2011, 33]. This is Daviss favored
sense, which he has argued for in his other writings. And here his explicit
argument against software engineering (currently) being engineering
(and his implicit argument against computer science (currently?) being
engineering) is that, although both are professions, neither is (currently)
part of the profession of engineering as it is taught and licensed in
engineering schools. (Even computer science departments that are
academically housed in engineering schools typically do not require their
students to take engineering courses, nor are their academic programs
accredited in the same way, nor are their graduates required to become
professional engineers in any legal senses.)
Interestingly, he does suggest that software engineers, at least, might eventually

become real engineers when real engineering starts becoming more
computational [Davis, 2011, 34].
20. Consider the validity and soundness of the following arguments:3
(a) Argument 1:
i. Engineers are cognitive agents that build artifacts for some identifiable
purpose.
ii. Birds build nests for housing their young.
iii. Beavers build dams because the sound of rushing water annoys them.4
iv. Computer engineers build computers for computation.5
v. Birds, beavers, and computer engineers are all engineers.
(b) Argument 2:
i. Engineers are cognitive agents that build artifacts for some identifiable
purpose and who know what that purpose is.
ii. Birds and beavers do not know why they build nests and dams,
resepectively; they are only responding to biological or evolutionary
instincts.
iii. Computer do know what the purpose of computation is.
iv. computer engineers are engineers, but birds and beavers are not.
3 With thanks to Albert Goldfain.
4 http://naturealmanac.com/archive/beaver dams/beaver dams.html (accessed 21 October 2013).
5 We will discuss what this means later; see Chs. 6, 7, and 9.
21. On declarative propositions vs. procedural statements:

[Goodman, 1987] argues that constructive algorithms, which describe how
something can be done, are intensional, hence go beyond the standard
extensional statements of ordinary mathematics. (On the difference between
intensional and extensional, see [Rapaport, 2009].) One reason is that we
need to be able to say what a program is for, because, in turn, The point
of discussin a particular algorithm is usually that it solves some problem that
interests us [Goodman, 1987, 476].
22. Mark Staples (personal communication, 2015) observes that Daviss definition
of engineering in terms of its curriculum is circular. . . . How does engineering
knowledge become accepted into engineering curricula?
Also: Contra Petroski, engineering is more than just design, because architects
design, but are not engineers.
Also: the key objective of engineering is to provide justified assurance (using

theories and evidence from empirical test) that artefact are suitable for use
(because they meet requirements specifications).
For a reply to Koen, see [Staples, 2014, 6.2].
On their being no widely accepted definition of engineering, see [Pawley, 2009].
. . . engineering research is only similar to science, and is not part of

science (4.1).
The comment that follows may be more relevant to Ch. 4: Perhaps it would be
useful to distinguish between a narrow and a wide notion of science: Science
in the narrow sense is what questions about what is science are focused on.
Science in the wide sense includes any logical, detailed, systematic study or
investigation of some phenomenon. It is in the latter sense that engineering is
similar to science. And one way that it is not part of science is in its use of
theories that would not be acceptable within science (4.1), namely, theories
that are phenomenological, . . . highly approximate, and have a highly limited
scope but are not fundamental or explanatory, . . . as precise as possible, . . . [or]
as universal as possible (4.1).
Unlike science, it is sufficient for engineering theories to be

phenomenological. [Staples, 2015, 5]
Presumably, it is necessary that scientific theories go beyond the merely

phenomenological (cf. Mach?).
Note:
A phenomenological theory is one which predicts phenomena, but

whose elements are not intended to correspond ontologically to real
entities in the world. [Staples, 2015, 2.2]
science tries to understand the world, whereas engineering tries to

change it. (4.1)
Compare Marx/Feuerbach on philosophy! (see 2.6.4). Computer science tries

to do both: to understand the world computationally, and to change the world by
building computational artifacts.
. . . engineering theories express claims that an artefact (represented

in the theory by a design) will perform in a way that satisfies its
requirements for use (represented in the theory by a requirements
specification). (8)
Empirical theories in engineering concern not just the behaviour

of artefacts, but also specifications of requirements for use, deigns
fro artefacts, and claims that artefacts meet requirements for use.
[Staples, 2015, 1].
I define engineering theories as explicit falsifiable claims used to

predict and analyse the performance of artefacts with respect to
requirements. [Staples, 2015, 2.1]

Abelson, Harold, & Sussman, Gerald Jay, with Sussman, Julie (1996), Preface to
the First Edition, Structure and Interpretation of Computer Programs
http://mitpress.mit.edu/sicp/full-text/book/book-Z-H-7.html#% chap Temp 4
Abrahams, Paul (1987), What Is Computer Science?, Communications of the
ACM 30(6) (June): 472473.
Arden, Bruce W. (1980), COSERS Overview, in Bruce W. Arden (ed.), What Can
(Cambridge, MA: MIT Press), Ch. 1, pp. 131; see esp. the section titled About
Names and Labels (pp. 57).
Bunge, Mario (1974), Towards a Philosophy of Technology, in Alex C. Michalos
(ed.), Philosophical Problems of Science and Technology (Boston: Allyn & Bacon):
2847.
Davis, Martin (2000), Introduction, in Martin Davis, The Universal Computer
(New York: W.W. Norton).
Davis, Michael (1995), Questions for STS [Science & Technology Studies] from
Engineering Ethics, talk given at the Society for the Social Study of Science,
Charlottesville, VA, October 22, 1995,
http://ethics.iit.edu/publication/Questions for STS.pdf
Denning, Peter J. (2013), Design Thinking, Communications of the ACM 56(12)
(December): 2931.
Denning, Peter J.; Comer, Douglas E.; Gries, David; Mulder, Michael C.; Tucker,
Allen; Turner, A. Joe; & Young, Paul R. (1989), Computing as a Discipline,
Denning, Peter J.; & Riehle, Richard D. (2009), Is Software Engineering
Engineering?, Communications of the ACM 52(3) (March): 2426.
Fiske, Edward B. (1989), Between the Two Cultures: Finding a Place in the
Curriculum for the Study of Technology, New York Times (29 March): B8; online
at:
http://www.nytimes.com/1989/03/29/us/education-lessons.html
(accessed 9 September 2013)
Florman, Samuel C. (1994), The Existential Pleasures of Engineering, 2nd edition
(New York: St. Martins Press).
esp. Ch. 14, The Engineering View, pp. 174184.
Hartmanis, Juris, & Lin, Herbert (eds.?) (1992), What Is Computer Science and
6 This book was the subject of a petition to withdraw it from
publication because we consider it misleading and even harmful as an agenda for future research
(http://www-formal.stanford.edu/jmc/petition/whysign/whysign.html).
Hoare, C.A.R. (2009), Retrospective: An Axiomatic Basis for Computer

Programming, Communications of the ACM 52(10) (October): 3032.
Has some interesting comments on the complementary nature of pure
academic research (science) and applied industrial research (engineering).
Kaag, John; & Bhatia, Sujata K. (2014), Fools for Tools: Why Engineers Need
to Become Philosophers, The Chronicle [of Higher Education] Review 61(13)
(28 November): B13B15,
http://chronicle.texterity.com/chronicle/20141128b?folio=B13#pg13
An essay on engineering ethics.
Koen, Billy Vaughn (2009), The Engineering Method and Its Implications for
Scientific, Philosophical, and Universal Methods, The Monist 92(3) (July): 357
386.
Koen seeks a universal method (not merely a scientific method or a method
used in the humanities); he finds it in the engineering method, which
he identifies with heuristics. cf. Simons notion of satisficing, which we
discussed in 2.6.1.4.
properties
Loui, Michael C. (1995), Computer Science Is a New Engineering Discipline,
Malpas, Robert (2000), The Universe of Engineering: A UK Perspective
(London: Royal Academy of Engineering)
http://tinyurl.com/Malpas00 (accessed 5 November 2012).
Petroski, Henry (2006), A Great Profession, American Scientist 94 (July-August):
304307.
Petroski, Henry (2007), Lab Notes, American Scientist 95(2) (March-April):
114117.
How a theoretician develop[ed] his applied side.
Popper, Karl R. (1972), Objective Knowledge: An Evolutionary Approach (Oxford:
Oxford University Press).
Parts of this book discuss the relation of engineering to science.
Simons, Peter M.; & Michael, Sir Duncan (eds.) (2009), Special Issue on
Philosophy and Engineering, The Monist 92(3) (July).
Staples, Mark (2014), Critical Rationalism and Engineering: Ontology, Synthese
191(10): 22552279,
http://www.nicta.com.au/pub?doc=7397
(preprint accessed 10 February 2015).
Presents a deductive view of theories in engineering, arguing that they
express claims that an artefact. . . will perform in a way that satisfies its
requirements for use (8). Definitions of engineering are discussed in 2.
The relation of engineering to science is discussed in 4.
Staples, Mark (2015), Critical Rationalism and Engineering: Methodology,

Synthese 192(1): 337362,
http://www.nicta.com.au/pub?doc=7747
A sequel to Staples 2014. Contains a useful taxonomy (3) of ways in which
artifacts can fail to conform to specifications (a topic that is also relevant to
our later discussion of implementation, Ch. 14).
Tedre, Matti (2009), Computing as Engineering, Journal of Universal Computer
Science 15(8): 16421658.
Vincenti, Walter G. (1990), What Engineers Know and How They Know It:
Analytical Studies from Aeronautical History (Baltimore: Johns Hopkins University
Press).
Argues that engineering is a kind of knowledge that is different from scientific
knowledge.
2. Websites:
Exercise for the reader: Link to various engineering websites, and try to find a definition
of engineer or engineering!
What Is Engineering?, Whiting School of Engineering, Johns Hopkins University
http://www.jhu.edu/ virtlab/index.php ().accessed 5 November 2012
What is engineering and what do engineers do?, National Academy of
Engineering of the National Academies
http://www.nae.edu/About/FAQ/20650.aspx (accessed 5 November 2012).
Chapter 6
What Is a Computer?
A Historical Perspective

Let us now return to the analogy of the theoretical computing machines. . . It can
be shown that a single special machine of that type can be made to do the work of
all. It could in fact be made to work as a model of any other machine. The special
machine may be called the universal machine. . . Alan Turing (1947), cited in
[Davis, 2012]
If it should turn out that the basic logics of a machine designed for the
numerical solution of differential equations coincide with the logics of a machine
intended to make bills for a department store, I would regard this as the most
amazing coincidence I have ever encountered. Howard Aiken (1956), cited in
[Davis, 2012]
Many people think that computation is for figuring costs and charges in a grocery
store or gas station. [Hill, 2008]
One of my daughters looked at a pile of MacBooks in the living room, and asked,
Can you hand me a computer? (28 March 2015)
Figure 6.1: 2012

c by ??
197
198 CHAPTER 6. WHAT IS A COMPUTER? A HISTORICAL PERSPECTIVE

1. Highly Desired (if you have enough time to read two full-length books!):
Davis, Martin (2012), The Universal Computer: The Road from Leibniz to Turing;
Turing Centenary Edition) (Boca Raton, FL: CRC Press/Taylor & Francis Group).
This will give you a good background in the logical history of computers,
written by one of the leading mathematicians in the field of theory of
computation.
At least try to read the Introduction (http://tinyurl.com/Davis00 accessed
5 November 2012), which is only one page long!
See Davis 1987, below. See also Papadimitriou 2001 and Dawson 2001 in
Further Sources of Information for reviews of Davis 2012.
Aspray, William (ed.) (1990), Computing before Computers (Ames, IA: Iowa State
University Press), http://ed-thelen.org/comp-hist/CBC.html
This will give you a good background in the engineering history of
computers.
2. Required:
Browse the linked websites at A Very Brief History of Computers
http://www.cse.buffalo.edu/rapaport/584/history.html
OConnor, J.J., & Robertson, E.F. (1998), Charles Babbage,
http://www-gap.dcs.st-and.ac.uk/ history/Mathematicians/Babbage.html
Simon, Herbert A., & Newell, Allen (1958), Heuristic Problem Solving: The Next
Advance in Operations Research, Operations Research 6(1) (January-February):
110.
Read the brief history of Babbages work (pp. 13); skim the rest.
In this paper, Simon and Newell predicted that (among other things) a
computer would be the worlds chess champion (p. 7) within 10 years, i.e.,
by 1968.1
For more on Adam Smiths pin-factory story (told in Simon & Newell 1958
and discussed below in 6.4.3), see:
Lawson, Robert (2004), Division of Labour,
http://www.divisionoflabour.com/archives/000006.php
Reprints the passage from Adam Smith on the pin factory.
Kennedy, Gavin (2008, 11 May), The Talmud on the Division of
Labour, Adam Smiths Lost Legacy (blog),
http://tinyurl.com/p52fh6e2 (accessed 5 November 2012).
Cowen, Tyler (2008), Division of labor in the Babylonian Talmud,
Marginal Revolution (blog),
http://tinyurl.com/m7e6s343 (accessed 5 November 2012).
1I once asked Simon about this; our email conversation can be found at:
http://www.cse.buffalo.edu/rapaport/584/S07/simon.txt (accessed 5 November 2012).
2 http://adamsmithslostlegacy.blogspot.com/2008/05/talmud-on-dvision-of-labour.html
3 http://marginalrevolution.com/marginalrevolution/2008/05/division-of-lab.html
Davis, Martin (1987), Mathematical Logic and the Origin of Modern Computers,
Studies in the History of Mathematics, reprinted in Rolf Herken (ed.), Universal
Turing Machine: A Half-Century Survey; Second Edition (Vienna: Springer-Verlag,
1995): 135158; online at:
http://staff.science.uva.nl/edoh/COMPHIS/davispaper.pdf
Article-length version of Davis 2012.
Ensmenger, Nathan (2003), Bits of History: Review of A.R. Burkss Who Invented
the Computer? The Legal Battle that Changed Computing History, in American
Scientist 91 (September-October): 467468;
http://www.americanscientist.org/bookshelf/pub/bits-of-history
(accessed 30 October 2013).
Robinson, J. Alan (1994), Logic, Computers, Turing, and von Neumann, in
K. Furukawa; D. Michie; & S. Muggleton (eds.), Machine Intelligence 13: Machine
Intelligence and Inductive Learning (Oxford: Clarendon Press): 135.
A personal history of the development of computers and the related history
of logic, by the developer of the resolution method of automated theorem
proving.
Browse through:
Computer History Museum, http://www.computerhistory.org/
Copeland, Jack (2000), A Brief History of Computing,
http://www.alanturing.net/turing archive/pages/Reference%20Articles/BriefHistofComp.html
Lee, J.A.N. (19952002), The History of Computing,
http://ei.cs.vt.edu/history/ (accessed 5 November 2012).
Hitmill.com (2004), History of Computers,
http://www.hitmill.com/computers/computerhx1.html
Hoyle, Michelle A. (19942006), The History of Computing Science
http://lecture.eingang.org/ (accessed 5 November 2012).
6.2 Introduction
Let us take stock of where we are. We began by asking what computer science is;
we saw that it might be a science, a branch of engineering, a combination of both, or
something sui generis; so we then investigated the nature of science and of engineering.
Now we want to turn to the question, What is computer science the science (or
engineering, or study) of ? There are at least three options to consider: The subject
matter of computer science might be computers (the physical objects that compute), or
it might be computing (the algorithmic processing that computers do), or it might be
something else, such as the information that gets processed. In this and a subsequent
chapter, we will examine the first two options. Readers interested in exploring the third
optioninformationare encouraged to look at some of the readings cited at the end
of this chapter [OR AT END OF CH. 1? IN ANY CASE, SEPARATE THEM OUT,
SO THAT THEYRE EASIER TO FIND]
To help answer the question What is a computer?, we will look first at the history
of computers and then at some philosophical issues concerning the nature of computers.
Lets begin our historical investigation with some terminology. Some 120 years
ago, in the May 2, 1892, issue of The New York Times, the following ad appeared:4
A COMPUTER WANTED.
WASHINGTON, May 1.A civil service ex-
amination will be held May 18 in Washington,
and, if necessary, in other cities, to secure
eligibles for the position of computer in the
Nautical Almanac Office, where two vacancies
existone at $1,000, the other at $1,400.
The examination will include the subjects of
algebra, geometry, trigonometry, and as-
tronomy. Application blanks may be obtained
of the United States Civil Service Commission.
So, a century or so ago, the answer to the question, What is a computer?, was: a
human who computes! In fact, until at least the 1940s (and probably the 1950s), that
was the meaning of computer:
Towards the close of the year 1850, the Author first formed the design
of rectifying the Circle to upwards of 300 places of decimals. He was
fully aware, at that time, that the accomplishment of his purpose would
add little or nothing to his fame as a Mathematician, though it might as a
Computer;. . . William Shanks (1853), as cited in [Hayes, 2014a, 342]
When people wanted to talk about a machine that computed, they would use the phrase
computing machine or (later) electronic (digital) computer. (Later on, when we look
at Alan Turings foundational papers in computer science and artificial intelligence, this
distinction will be important.) Interestingly, nowadays when one wants to talk about a
human who computes, we need to use the phrase human computer [Pandya, 2013].
4 The original may be seen online at [http://tinyurl.com/NYT-computer].
For the sake of familiarity, in this book, I will use the word computer for the machine
and the phrase human computer for a human who computes.
6.3 Two Histories of Computers

There seem to be two histories of computers; they begin in parallel, but eventually
converge and intersect.
The goal of one of these histories was to build a machine that could compute (or
calculate), that is, a machine that could duplicate, and therefore assist, or even replace,
or eventually supersede human computers. This is an engineering goal!
The goal of the other history was to provide a foundation for mathematics. This is
a scientific (or, at least, a mathematical) goal!
These histories were probably never really parallel but more like a tangled web,
with two bridges connecting them: the first was a person who lived about 330 years
ago and the other was someone who was active only about 80 years agoGottfried
Wilhelm Leibniz (16461716) and Alan Turing (19121954). (There were, of course,
other significant people involved in both histories, as we will see.) Both histories begin
in ancient Greece, the engineering history beginning with the need for computational
help in navigation and the scientific history beginning with Aristotles study of logic.
6.4 The Engineering History

The engineering history concerns the attempt to create machines that would do certain
mathematical computations. The two main reasons for wanting to do this seem to be
(1) to make life easier for humans (let a machine do the work) andperhaps of more
importance(2) to produce computations that are more accurate (both more precise
and with fewer errors) than those that humans produce. It is worth noting that the goal
of having a machine perform an intellectual task that would otherwise be done by a
human is one of the motivations underlying AI.
Here, I will only sketch some of the highlights of the engineering history.
Articles such as [Chase, 1980], books such as [Aspray, 1990], and websites such as
[Carlson et al., 1996] go into much more detail. And [Sloman, 2002, 2] argues that
even the engineering history of computers has two strands: the development of
machines for controlling physical mechanisms and [the] development of machines for
performing abstract operations, e.g. on numbers.
6.4.1 Ancient Greece

The very early history of the attempt to build machines that could calculate can
be traced back at least to the second century B.C.E., when a device now known
as the Antikythera Mechanism was constructed [Freeth, 2006], [Wilford, 2006],
[Seabrook, 2007], [Wilford, 2008], [Freeth, 2009]. This was a device used to calculate
astronomical information, possibly for use in agriculture or religion. Although the
Antikythera Mechanism was discovered in 1900, a full understanding of what it was
and how it worked was not figured out until the 2000s.
6.4. THE ENGINEERING HISTORY 203
6.4.2 17th Century Adding Machines

Skipping ahead almost 2000 years to about 350 years ago, two philosopher-
mathematicians are credited with more familiar-looking calculators: Blaise Pascal
(16231662), who helped develop the theory of probability, also invented an adding
(and subtracting) machine that worked by means of a series of connected dials, and
Leibniz (who invented calculus, almost simultaneously but independently of Isaac
Newton) invented a machine that could add, subtract, multiply, and divide. As well
see later on, Leibniz also contributed to the scientific history of computing with an idea
for something he called a calculus ratiocinator (loosely translatable as a reasoning
system).
6.4.3 Babbage Machines

Two of the most famous antecedents of the modern electronic computer were due to
the English mathematician Charles Babbage, who lived about 175 years ago (1791
1871). The first of the machines he developed was the Difference Engine (18211832).
A French mathematician named Gaspard de Prony, who later headed Frances civil-
engineering college, needed to construct highly accurate, logarithmic and trigonometric
tables for large numbers. He realized that it would take him too long using difference
equations by hand.5 In the 1790s, he had read Adam Smiths 1776 text on economics,
The Wealth of Nations, which discussed the notion of the division of labor: The
manufacture of pins could be made more efficient by breaking the job down into
smaller units, with each laborer who worked on one unit becoming an expert at his one
job. This is essentially what modern computer programmers call top-down design
and stepwise refinement [Wirth, 1971]: To accomplish some task T , analyze it into
subtasks T1 , . . . , Tn , each of which should be easier to do than T . This technique can be
repeated: Analyze each Ti into sub-subtasks Ti1 , . . . , Tim , and so on, until the smallest
sub. . . subtask is so simple that it can be done without further instruction (this is the
essence of recursion; see 7.7). De Prony applied this division of labor to computing
the log and trig tables, using two groups of human computers, each as a check on the
other. (For a discussion of this, written by two pioneers in computer science and AI,
see [Simon and Newell, 1958]; see also [Pylyshyn, 1992].)
Babbage wanted a machine to replace de Pronys people; this was his Difference
Engine. He later conceived of an Analytical Engine (18341856), which was
intended to be a general-purpose, problem solver (perhaps more closely related to
Leibnizs goal for his calculus ratiocinator). Babbage was unable to completely build
either machine: The techniques available to him in those days were simply not up to
the precision required. However, he developed techniques for what we would today
call programming these machines, using a 19th-century version of punched cards
(based on a technique invented by Joseph Marie Jacquard for use in looms). Lady
Ada Lovelace (18151852), daughter of the poet Lord Byron, working with Babbage,
wrote a description of how to program the (yet-unbuilt) Analytical Engine; she is, thus,
considered to be the first computer programmer. (But, for a more historically accurate
5 Difference equations are a discrete-mathematical counterpart to differential equations. They involve
taking successive differences of sequences of numbers.

discussion of the history of programming, see [Randell, 1994].) A partial model of

the Difference Engine was finally built around 1991 [Swade, 1993], and efforts are
underway to build a version of the Analytical Engine (http://www.plan28.org/).
6.4.4 Electronic Computers

The modern history of electronic, digital computers is rather confusing and the source
of many historical and legal disputes. Here is a brief survey:
1. John Atanasoff (19031995) and his student Clifford Berry (19181963),

working at Iowa State University, built the ABC (Atanasoff-Berry Computer) in
19371942. This may have been the first electronic, digital computer, but it was
not a general-purpose (programmable) computer, and it was never completed. It
was, however, the subject of a patent-infringement suit, about which more in a
moment.
2. Konrad Zuse (19101995), in Germany, developed the Z3 computer in 1941,

which was programmable.
3. Howard Aiken (19001973), influenced by Babbage, built the Harvard Mark I

computer in 1944; it was designed to compute differential equations.
4. John Presper Eckert (1919-1995) and his student John Mauchly (19071980),
working at the University of Pennsylvania, built the ENIAC (Electronic
Numerical Integrator And Computer) in 1946. This was the first general-
purposethat is, programmableelectronic computer. In 1945, with the
collaboration of the mathematician John von Neumann (19031957)who
outlined an architecture for computers that is still used todaythey developed
the EDVAC (Electronic Discrete Variable Automatic Computer), which used
binary arithmetic (rather than decimal). This evolved into the first commercial
computer: the UNIVAC (UNIVersal Automatic Computer). UNIVAC became
famous for predicting, on national TV, the winner of the 1952 US presidential
election. The company that made UNIVAC evolved into the Sperry Rand
Corporation, who owned the patent rights. The Honeywell Corporation, a
rival computer manufacturer, successfully sued Sperry Rand in 1973, on the
grounds that Mauchly had visited Atanasoff in 1941, and that it was Atanasoff
(and Berry), not Mauchly (and Eckert) who had invented the computer,
thus vacating Sperry Rands patent claims. (A useful summary of some of
the issues involved can be found in [Ensmenger, 2003], who observes that
identifying Atanasoff as the inventor of the computer (my phrasing and italics)
is an answer to what is fundamentally the wrong question [Ensmenger, 2003,
italics in original], because any particular claim to priority of invention must
necessarily be heavily qualified: If you add enough adjectives, you can always
claim your own favorite. The story of von Neumanns involvement in the
development of computers can be found in [Dyson, 2012b].)
5. There were two significant developments in the UK:

6.5. THE LOGICAL AND MATHEMATICAL HISTORY 205
(a) In 1943, the Colossus computer was developed and installed at Bletchley
Park for use in cryptography during World War II. Bletchley Park was
where Alan Turing worked on cracking the Nazis code-making machine,
the Enigma.
(b) After the war, in 1945, Turing decided to try to implement his a-machine
(the Turing machine), and developed the ACE (Automatic Computing
Engine) [Copeland, 1999]. (It was also around this time that Turing started
thinking about AI and neural networks.)
6.5 The Logical and Mathematical History

Logics dominant role in the invention of the modern computer is not widely
appreciated. The computer as we know it today was invented by Turing in 1936,
an event triggered by an important logical discovery announced by Kurt Godel in
1930. Godels discovery. . . decisively affected the outcome of the so-called Hilbert
Program. Hilberts goal was to formalize all of mathematics and then give positive
answers to three questions about the resulting formal system: is it consistent?
is it complete? is it decidable? Godel found that no sufficiently rich formal
system of mathematics can be both consistent and comlete. In proving this, Godel
invented, and used, a high-level symbolic programming language: the formalism
of primitive recursive functions. As part of his proof, he composed an elegant
modular functional program. . . . This computational aspect of his work. . . is
enough to have established Godel as the first serious programmer in the modern
sense. Godels computational example inspired Turing. . . [who] disposed of the
third of Hilberts questions by showing. . . that the formal system of mathematics
is not decidable. Although his original computer was only an abstract logical
concept,. . . Turing became a leader in the design, construction, and oepration of
the first real computers. [Robinson, 1994, 67]
The parallel historical story concerns, not the construction of a physical device that
could compute, but the logical and mathematical analysis of what computation itself
is.
This story begins, perhaps, with Leibniz, who not only constructed a computing
machine, as we have seen, but who also wanted to develop a calculus ratiocinator:
a formalism in a universally understood language (a characteristica universalis) that
would enable its speakers to precisely express any possible question and then to
rationally calculate its answer. Leibnizs motto (in Latin) was: Calculemus! (Let us
calculate!). In other words, he wanted to develop an algebra of thought.
This task was taken up around 160 years later (around 160 years ago) by the
English mathematician George Boole (18151864), who developed an algebra of logic,
which he called The Laws of Thought [Boole, 2009]. This was what is now called
propositional logic. But it lacked a procedure for determining the truth value of a given
(atomic) statement.
Booles work was extended by the German mathematician Gottlob Frege (1848
1925, around 120 years ago), who developed what is now called first-order logic (or the
first-order predicate calculus).6 Frege was a follower of a philosophy of mathematics

called logicism, which viewed mathematics as a branch of logic. Thus, to give a firm
foundation for mathematics, it would be necessary to provide a system of logic that
itself would need no foundation.
Unfortunately, the English philosopher Bertrand Russell (18721970, around 90
years ago), upon reading Freges book The Foundations of Arithmetic in manuscript,
discovered a problem. This problem, now known as Russells Paradox, concerned the
logic of sets: A set that had as members all only those sets that do not have themselves
as members would both have itself as a member and not have itself as a member. This
inconsistency in Freges foundation for mathematics began a crisis that resulted in the
creation of the theory of computation.7
That story continues with work done by the German mathematician David
Hilbert (18621943, around 100 years ago), another logicist who wanted to set
mathematics on a rigorous, logical foundation, one that would be satisfactory to all
philosophers of mathematics, including intuitionists and finitists. (Intuitionists believe
that mathematics is a construction of the human mind, and that any mathematical claim
that can only be proved by showing that its assumption leads to a contradiction should
not be accepted. Finitists believe that only mathematical objects constructible in a
finite number of steps should be allowed into mathematics.) Hilbert proposed the
following Decision Problem (in German: Entscheidungsproblem) for mathematics:
to devise a procedure according to which it can be decided by a finite number of
operations whether a given statement of first-order logic is a theorem.8 A mathematical
statement that was decidable in this way was also said to be effectively computable
or effectively calculable, because one could compute, or calculate, whether or not it
6 None of these things called calculus (plural: calculi) are related to the differential or integral calculus.
Calculus just means system for calculation.

7 An enjoyable graphic-novel treatment of the Russell-Frege story, written by a well-known computer
scientist, is [Doxiadis et al., 2009].

8 There are varying statements and translations of Hilberts statements about the decision problem. Here
are some:
. . . determining whether or not a given formula of the predicate calculus is universally valid.
[Hilbert and Ackermann, 1928, 112]
In the broadest sense, the decision problem can be considered solved if we have a method
which permits us to decide for any given formula in which domains of individuals it is
universally valid (or satisfiable) and in which it is not. [Hilbert and Ackermann, 1928, 115]
The Entscheidunsproblem is solved if one knows a procedure that allows one to decide
the validity (respectively, satisfiability) of a given logical expression by a finite number of
operations. (Translation in [Sieg, 1994, 77]; my italics.)
An earlier version (dating from 1900) appeared in Hilberts list of 23 math problems that he thought should
be investigated during the 20th century. The 10th problem was this:
Given a diophantine equation with any number of unknown quantities and with rational
integral numerical coefficients: to devise a process according to which it can be determined
by a finite number of operations whether the equation is solvable in rational integers.
(English translation from http://aleph0.clarku.edu/djoyce/hilbert/problems.html#prob10
(accessed 11 December 2013.)
(Like the Halting Problem, this turns out to be non-computable; that is, there is no such process, no such
algorithm.) For more on the decision problem, see 8.3.
6.5. THE LOGICAL AND MATHEMATICAL HISTORY 207
was a theorem in a finite number of steps.9

(A brief digression on what a theorem is: When you studied geometry, you
may have studied it in a modern interpretation of Euclids original presentation of
it as an axiomatic system. Most branches of mathematics (and, according to some
philosophers, most branches of science) can be formulated axiomatically. One begins
with a set of axioms; these are statements that are assumed to be true (or are
considered to be so obviously true that their truth can be taken for granted). Then
there are rules of inference that tell you how to logically infer other statements from
the axioms in such a way that the inference procedure is truth preserving; that is, if
the axioms are true (which they are, by assumption), then whatever logically follows
from them according to the rules of inference are also true. (See Appendix. A for more
details.) Such statements are called theorems.)
Many mathematicians took up Hilberts challenge: In the US, Alonzo Church
(19031995) analyzed the notion of function and developed the lambda-calculus,
claiming that any function whoses values could be computed in the lambda-calculus
was effectively computable. The Austrian logician Kurt Godel (19061978), who
had previously proved the incompleteness of first-order logic (and thus became the
most respected logician since Aristotle)essentially, the claim that there were true
arithmetic statements that could never be proveddeveloped the notion of recursive
functions, claiming that this was co-extensive with effectively computable functions.
Emil Post, a Polish-born, American logician (18971954), developed production
systems, also claimed to capture the notion of effective computability. And the
Russian A.A. Markov (18561922) developed what are now known as Markov
algorithms. (We will look in more detail at some of these systems in Ch. 7.)
But central to our story was the work of the English mathematician Alan Turing
(19121954), who, rather than trying to develop a mathematical theory of effectively
computable functions in the way that the others approached the subject, gave an
analysis of what human computers did. Based on that analysis, he developed a formal,
mathematical model of a human computer, which he called an a-machine, and which
we now call, in his honor, a Turing machine. In his classic paper published in 1936
[Turing, 1936], Turing presented his informal analysis of human computation, his
formal definition of an a-machine, his claim that functions computable by a-machines
were all and only the functions that were effectively computable, a (negative)
solution to Hilberts Decision Problem (by showing that there was a mathematical
problem that was not decidable computationally, namely, the Halting Problem), a
demonstration that a single Turing machine (a universal Turing machine) could do
the work of all other Turing machines, andas if all that were not enougha proof
that a function was computable by an a-machine if and only if it was computable in
the lambda-calculus. (We will look at Turings work in much greater detail in Ch. 8.
For a very elementary introduction to the lambda-calculus, see [PolR, 2009, Alonzo
Churchs Lambda-Calculus].)
Later, others proved that both methods were also logically equivalent to all of the
others: recursive functions, production systems, Markov algorithms, etc. Because all
of these theories had been proved to be logically equivalent, this finally convinced
9 For more on this notion of effectiveness, see 7.5.
almost everyone that the notion of effective computability (or algorithm) had been
captured precisely. Indeed, Godel himself was not convinced until he read Turings
paper, because Turings was the most intuitive presentation of them all. (But we will
look at the arguments of those who are not convinced in Chs. 10 and 11.)
6.6 The Histories Converge

At this point, the engineering and mathematical histories converge: The development
of programmable, electronic, digital computersespecially the EDVAC, with its
von Neumann architecturebegan to look like Turing machines, and Turing himself
decided to implement a physical computer based on this architecture. (For more on this,
see [Carpenter and Doran, 1977].) But Turing machines are hopelessly inefficient and
cumbersome (register machines, another Turing-equivalent model of computation,
are closer to modern computers):
What is the difference between a Turing machine and the modern

computer? Its the same as that between Hillarys ascent of Everest and the
establishment of a Hilton hotel on its peak. (Alan J. Perlis, Epigrams in
Programming [http://www.cs.yale.edu/homes/perlis-alan/quotes.html].)
To clarify some of this history, it will be necessary for us to look more closely at
the nature of effective computation and algorithms, which we will do in the next
chapter. Armed with the results of that investigation, we will return to the question of
what a computer is (from a philosophical point of view).

1. Make Henkin 1962, required for next chapter, at least recommended here (or
maybe required here and recommended in next chap)
2. Use these quotes from Hilbert; the first might be better in Ch. 7:
It remains to discuss briefly what general requirements may be

justly laid down for the solution of a mathematical problem. I
should say first of all, this: that it shall be possible to establish the
correctness of the solution by means of a finite number of steps
based upon a finite number of hypotheses which are implied in
the statement of the problem and which must always be exactly
formulated. This requirement of logical deduction by means of a
finite number of processes is simply the requirement of rigor in
reasoning. [Hilbert, 1900, 440441]
Occasionally it happens that we seek. . . [a] solution [to a

mathematical problem] under insufficient hypotheses or in an
incorrect sense, and for this reason do not succeed. The problem
then arises: to show the impossibility of the solution under the
given hypotheses, or in the sense contemplated. Such proofs of

impossibility were effected by the ancients, for instance when they
showed that the ratio of the hypotenuse to the side of an isosceles
right triangle is irrational. In later mathematics, the question as to
the impossibility of certain solutions plays a preeminent part, and
we perceive in this way that old and difficult problems, such as
the proof of the axiom of parallels, the squaring of the circle, or
the solution of equations of the fifth degree by radicals have finally
found fully satisfactory and rigorous solutions, although in another
sense than that originally intended. It is probably this important
fact along with other philosophical reasons that gives rise to the
conviction (which every mathematician shares, but which no one
has as yet supported by a proof) that every definite mathematical
problem must necessarily be susceptible of an exact settlement,
either in the form of an actual answer to the question asked, or
by the proof of the impossibility of its solution and therewith
the necessary failure of all attempts. Take any definite unsolved
problem,. . . However unapproachable these problems may seem to
us and however helpless we stand before them, we have, nevertheless,
the firm conviction that their solution must follow by a finite
number of purely logical processes. Is this axiom of the solvability
of every problem a peculiarity characteristic of mathematical thought
alone, or is it possibly a general law inherent in the nature of the
mind, that all questions which it asks must be answerable? This
conviction of the solvability of every mathematical problem is a
powerful incentive to the worker. We hear within us the perpetual
call: There is the problem. Seek its solution. You can find it by pure
reason, for in mathematics there is no ignorabimus. [Hilbert, 1900,
444-445, my boldface]10
3. If computers can be defined historically, then they are machines which

(i) perform calculations with numbers, (ii) manipulate or process data
(information), and (iii) control continuous processes or discrete devices. . . in real
time or pseudo real time [Davis, 1977, 1096]. Note that (ii) can be considered a
generalization of (i), because performing calculations with numbers is a kind of
manipulation of data, and because numbers themselves are a kind of data. (Its
also the case that computers really manipulate numeralsthat is, symbols that
represent numbersnot (abstract) numbers themselves.) And, because being
continuous or being discrete pretty much exhausts all possibilities, criterion (iii)
doesnt really seem to add much. So this characterization comes down to (iii)
alone: A computer is a machine that manipulates or processes data (information).
According to [Davis, 1977, 10961097], computers had evolved to have the
following key characteristics (perhaps among others):
10 For more on impossibility proofs in mathematics, see [Stewart, 2000]; cf. [Toussaint, 1993]. The most
famous impossibility proof in computer science, of course, is the Halting Problem; see 7.8.
(a) digital operation (thus focusing on only the discrete aspect of (iii),
above);
(b) stored program capability (here understood as the notion that the
instructions for the computer be written in the same form as the data being
used by the computer and attributed to von Neumann; we will return to
this issue in Ch. 8);
(c) self-regulatory or self-controlling capability (this is not merely the
automaticity of any machine, but it seems to include the ideas of feedback
and automatic modifiable stored programs);
(d) automatic operation (singled out from the previous characteristic because
of its emphasis on operating independently of human opertors and human
intervention)
(e) reliance on electronics (this is admitted to be somewhat parochial in the
sense that electronic computers were, at the time of writing, the dominant
way of implementing them, but Davis recognized that other kinds of
computers would eventually exist)
So, ignoring the last item and merging the previous two, we come down to
this: A (modern) computer is an automatic, digital, stored-program machine (for
manipulating information).
4. On numerals vs. numbers:
Are computers symbol-manipulation machines or number-manipulations
machines? There are two versions of this question. The first version contrasts
numerical symbols (numerals) with numbers themselves. The second version
contrasts symbols more generally with either numerals in particular or with
numbers.
The first question is closely related to the philosophy of mathematics. Is math
itself more concerned with numerals than with numbers, or the other way
around? Formalists and nominalists suggest that it is only the symbols for
numbers that we really deal with. Platonists suggest that it is numbers that
are our real concern, but at least some of them admit that the only way that
we can directly manipulate numbers is via numerals (although some Platonists,
including Godel, suggest that we have a kind of perceptual ability, called
intution, that allows us to access numbers directly). There are also related
questions about whether numbers exist and, if so, what they are. But these
issues are beyond our scope. Computers, pretty clearly, have to deal with
numbers via numerals. So, The mathematicians and engineers then [in the
1950s] responsible for computers [who] insisted that computers only processed
numbersthat the great thing was that instrutions could be translated into
numbers [Newell, 1980, 137] are probably wrong. But even if we modify such
a claim so that we replace numbers by numerals, we are faced with the second
question above.
An answer to that question will depend in part on how we interpret the symbols
that a computer deals with. Certainly, there are ways to build computers
that, apparently, can deal with more than merely numerical symbols; the Lisp
machines of the 1980s are prime examples: Their fundamental symbols were
Lisp lists. But, insofar as any computer is ultimately constructed from physical
switches that are either in an on/up or off/down position, we are left with
a symbol system that is binaryhence numericalin nature. Whether we
consider these symbols to be numerals or not may be more a matter of taste
or convenience than anything more metaphysical.
5. On Adam Smiths division of labor:

It should be noted that, besides its positive effects on efficiency, the division of
labor has negative ones, too: It would make workers as stupid and ignoratn as
it is possible for a human creature to be. This was because no worker needed to
know how to make a pin, only how to do his part in the process of making a pin
[Skidelsky, 2014, 35], quoting Adam Smith, in The Wealth of Nations, Book V,
Ch. I, Part III, Article II (https://www.marxists.org/reference/archive/smith-
adam/works/wealth-of-nations/book05/ch01c-2.htm)
6. Nice quote, possibly for end of 6.4.4:
There are currently over one hundred computers installed in

American universities. Probably two dozen or more will be added this
year. In 1955 the number was less than twenty-five. . . . [C]onsidering
the costs involved in obtaining, maintaining, and expanding these
machines, the universities have done very well in acquiring hardware
with their limited funds. [Perlis, 1962, 181, my italics]
Of course, a university with, say, 5000 students now probably has at least 5000
computers, if not double that amount (by including smart phones), not to mention
the computers owned by the universities themselves! And each one of those
10,000 computers is at least as powerful, if not more so, than the 100 that there
were a half-century ago.
The basic purpose [of computers at universities], at present [that is,

in 1962], is to do computations associated with and supported by
university research programs, largely government financed. . . . Sad
to state, some uses occur merely because the computer is available,
and seem to have no higher purpose than that. [Perlis, 1962, 182, my
italics]
And I wouldnt be surprised if most uses (Candy Crush?, Skype?, Facebook?,

Twitter?, Amazon?) of the 10,000 computers at an average contemporary
university have no higher purpose!
7. By the Entscheidungsproblem of a system of symbolic logic is here

understood the problem to find an effective method by which, given
any epression Q in the notation of the system, it can be determined
whether or not Q is provable in the system. [Church, 1936a, 41n6]
8. On the Antikythera:
Relevant to questions about whether everything computes anything (the Searle-
Putnam puzzle), we can ask the following questions about the Antikythera, a
real-life example of the computer found in the desert: Does it compute? What
does it compute? Is what it computes determined by its creators? Can we
determine it?
9. The two separate histories of computer science are noted by

[Mahoney and Haigh (ed.), 2011]:
. . . it is really only in von Neumanns collaboration with the
ENIAC team that two quite separate historical strands came together:
the effort to achieve high-speed, high-precision, automatic calculation
and the effort to design a logic machine capable of significant
reasoning.
The dual nature of the computer is reflected in its dual origins:
hardware in the sequence of devices that stretches from the Pascaline
to the ENIAC, software in the series of investigations that reaches
from Leibnizs combinatorics to Turings abstract machines. Until the
two strands come together in the computer, they belong to different
histories. . . . (p. 26)
He also emphasizes the fact that these histories converged but were not
coincident (p. 88)

FOR NEXT DRAFT: SEPARATE OUT THINGS LIKE ALL ARTICLES ON ADA, OR
ON BABBAGE, AND SEPARATE BY ENGG HISTORY VS. LOGICAL HIST, ETC.

Baranger, Walter R. (1995), J. Presper Eckert, Co-Inventor of Early Computer,
Dies at 76, New York Times (7 June): B12; online at:
http://www.nytimes.com/1995/06/07/obituaries/j-presper-eckert-co-inventor-of-
early-computer-dies-at-76.html
Baranger, Walter R. (1995), John V. Atanasoff, 91, Dies; Early Computer
Researcher, New York Times (17 June): 11; online at:
http://www.nytimes.com/1995/06/17/obituaries/john-v-atanasoff-91-dies-early-
computer-researcher.html
Campbell-Kelly, Martin (2009), Origin of Computing, Scientific American
(September): 6269; online at:
http://www.cs.virginia.edu/robins/The Origins of Computing.pdf
See also Readers Respond on the Origin of Computing , Scientific
American (January 2010): 8; online at:
http://www.scientificamerican.com/article.cfm?id=letters-jan-10
Campbell-Kelly, Martin (2010), Be Careful What You Wish For: Reflections on
the Decline of Mathematical Tables, Communications of the ACM 53(4) (April):
2526.
Campbell-Kelly, Martin (2011), In Praise of Wilkes, Wheeler, and Gill ,
Communications of the ACM 54(9) (September): 2527;
http://dx.doi.org/10.1145/1995376.1995386
Reflections on the first textbook on programming.
Campbell-Kelly, Martin (2012), Alan Turings Other Universal Machine,
Communications of the ACM 55(7) (July): 3133.
Reflections on the Turing ACE computer and its influence.
Care, Charles (2007), Not Only Digital: A Review of ACMs Early Involvement
with Analog Computing Technology, Communications of the ACM 50(5) (May):
4245; http://dx.doi.org/10.1145/1230819.1230837
Chaitin, Gregory J. (2002), Computers, Paradoxes and the Foundations of
Mathematics, American Scientist 90 (March-April): 164171; online at:
http://www.umcs.maine.edu/chaitin/amsci.pdf
Chase, George C. (1980), History of Mechanical Computing Machinery, Annals
of the History of Computing 2(3) (July): 198226.
An illustrated history, with a useful introduction by I. Bernard Cohen.
Collen, Morris F. (1994), The Origins of Informatics, Journal of the American

Medical Informatics Association 1(2) (March/April): 91107.
A useful survey of the history of computer science aimed at practitioners of
medical informatics.
Copeland, B. Jack (2004), Computation, in Luciano Floridi (ed.), The Blackwell
Guide to the Philosophy of Computing and Information (Malden, MA: Blackwell).
Esp. pp. 34 on The Birth of the Modern Computer
Copeland, Jack; Flowers, Thomas; & 17 Bletchley Park Codebreakers (2010),
Colossus: The Secrets of Bletchley Parks Codebreaking Computers (New York:
Oxford University Press); online at:
http://www.colossus-computer.com/contents.htm
Davis, Martin (2000), Overheard in the Park, American Scientist 88 (July-
August): 366367.
http://www.americanscientist.org/bookshelf/pub/overheard-in-the-park
A somewhat negative review of David Berlinksi, The Advent of the Algorithm
(Harcourt, 2000), correcting some of the historical errors in that book.
Davis, Martin (2003), Paradoxes in Paradise, American Scientist 91 (May-June):
268269; online at:
http://www.americanscientist.org/bookshelf/pub/paradoxes-in-paradise
Review of Marcus Giaquinto, The Search for Certainty: A Philosophical
Account of Foundations of Mathematics (Oxford University Press, 2002).
Davis, Martin (2004), The Myth of Hypercomputation, in C. Teuscher (ed.),
Alan Turing: The Life and Legacy of a Great Thinker (Berlin: Springer): 195
212.
Accessed 15 May 2014 from:
http://www1.maths.leeds.ac.uk/pmt6sbc/docs/davis.myth.pdf
Early sections contain a good summary of the history of computation.
Dawson, John W. (2001), Review of Davis 2012, Bulletin of Symbolic Logic 7(1)
(March): 6566; online at:
http://www.math.ucla.edu/asl/bsl/0701/0701-003.ps
Dyson, George (2012), Turings Cathedral: The Origins of the Digital Universe
(New York: Pantheon).
Holt, Jim (2012), How the Computers Exploded (review of Dyson 2012),
New York Review of Books (7 June): 3234.
Mauchly, Bill; Bernstein, Jeremy; Dowson, Mark; Adams, David K.; &
Holt, Jim (2012, 27 September), Who Gets Credit for the Computer?: An
Exchange, New York Review of Books: 96, 98.
Ensmenger, Nathan (2011), Building Castles in the Air, Communications of the
ACM 54(4) (April): 2830; http://dx.doi.org/10.1145/1924421.1924432
Reflections on recruiting and training programmers during the early period
of computing.
Forster, E.M. (1909), The Machine Stops

http://archive.ncsa.illinois.edu/prajlich/forster.html (accessed 6 November 2012).
Anybody who uses the Internet should read E.M. Forsters The Machine
Stops. It is a chilling, short story masterpiece about the role of technology
in our lives. Written in 1909, its as relevant today as the day it was published.
Forster has several prescient notions including instant messages (email!) and
cinematophotes (machines that project visual images). Paul Rajlich
(from the above-cited website)
George, Alexander (1983), Philosophy and the Birth of Computer Science,
Robotics Age (January/February): 2631.
Goldstine, Herman H. (1972), The Computer from Pascal to von Neumann
(Princeton, NJ: Princeton University Press).
Green, Christopher D. (2005), Was Babbages Analytical Engine Intended to Be a
Mechanical Model of the Mind?, History of Psychology 8(1): 3545,
http://www.yorku.ca/christo/papers/babbage-HoP.pdf
Briefly, no (at least, not by Babbage).
Grier, David Alan (2005), When Computers Were Human (Princeton, NJ: Princeton
University Press).
Skinner, David (2006), The Age of Female Computers, review of Grier
2005, The New Atlantis (Spring): 96103
http://www.thenewatlantis.com/publications/the-age-of-female-
computers
and http://www.thenewatlantis.com/docLib/TNA12-Skinner.pdf
Hafner, Katie (2002), Happy Birthday :-) to You: A Smiley Face Turns 20, New
York Times (19 September): G4; online at:
http://www.nytimes.com/2002/09/19/technology/typographic-milestones-happy-
birthday-to-you-a-smiley-face-turns-20.html
Haigh, Thomas (2014), Actually, Turing Did Not Invent the Computer,
Discusses the (ir)relevance of the mathematical history of computation to the
engineering history.
Halmos, Paul R. (1973), The Legend of John von Neumann, American
Mathematical Monthly (April): 382394.
A very readable, short biography, with a heavy emphasis on the humorous
legends that have grown up around von Neumann.
Hayes, Brian (1994), The World Wide Web, American Scientist 82 (September-
October): 416420; online at:
http://bit-player.org/wp-content/extras/bph-publications/AmSci-1994-09-Hayes-
www.pdf
Hayes, Brian (2000), The Nerds Have Won, American Scientist 88 (May-June):
200204; onine at:
http://www.americanscientist.org/issues/pub/the-nerds-have-won
On the history of the Internet.
Holt, Jim (2001), The Ada Perplex, The New Yorker (5 March): 8893.
How Byrons daughter came to be celebrated as a cyber-visionary.
Husbands, Philip; Wheeler, Michael; & Holland, Owen (2008), Introduction: The
Mechanical Mind, in Philip Husbands, Owen Holland, & Michael Wheeler (eds.),
The Mechanical Mind in History (Cambridge, MA: MIT Press): 117.
Hyman, Anthony (1982), Charles Babbage: Pioneer of the Computer (Princeton:
Princeton University Press).
Reviewed in:
OHanlon, Redmond (1982), A Calculating Man, New York Times Book
Review (14 November): BR26BR27.
Hyman, Paul (2012), Lost and Found, Communications of the ACM 55(7) (July):
21.
Researchers discover computer pioneer Konrad Zuses long-forgotten Z9, the
worlds first program-controlled binary relay calculator using floating-point
arithmetic.
See also:
Winkler, Jurgen F.H. (2012), Konrad Zuse and Floating-Point Numbers,
letter to the editor, Communications of the ACM 55(10) (October): 67.
Johnson, George (2001), Claude Shannon, Mathematician, Dies at 84, New York
Times (27 February): B7; online at:
http://www.nytimes.com/2001/02/27/nyregion/27SHAN.html
Johnson, Mark (2002), Review of Davis 2012, MAA Reviews
http://mathdl.maa.org/mathDL/19/?pa=reviews&sa=viewBook&bookId=68938
Keats, Jonathan (2009), The Mechanical Loom, in The Start of Everything,
Scientific American (September): 88; online at:
http://www.readcube.com/articles/10.1038/scientificamerican0909-88
The story of the Jacquard loom.
Kernan, Michael (1990), The Object at Hand, Smithsonian 21 (May): 22, 24, 26.
About the Nazis code-making machine, the Enigma.
Kim, Eugene Eric; & Toole, Betty Alexandra (1999), Ada and the first Computer,
Scientific American (May): 7681; online at:
http://people.cs.kuleuven.be/danny.deschreye/AdaTheFirst.pdf
Kleene, Stephen C. (1981), Origins of Recursive Function Theory, Annals of the
History of Computing 3(1) (January): 5267.
Levy, Steven (2013), The Brief History of the ENIAC Computer, 101 Objects
that Made America, Smithsonian (November): 6264; online at:
http://www.smithsonianmag.com/history-archaeology/The-Brief-History-of-the-
ENIAC-Computer-228879421.html
Lohr, Steve (1996), The Face of Computing 50 Years and 18,000 Tubes Ago,
New York Times (19 February): D3; online at:
http://www.nytimes.com/1996/02/19/business/the-face-of-computing-50-years-
and-18000-tubes-ago.html
Lohr, Steve (2001), Frances E. Holberton, 84, Early Computer Programmer, New
York Times (17 December): F5; online at:
http://www.nytimes.com/2001/12/17/business/frances-e-holberton-84-early-
computer-programmer.html (accessed 12 November 2013).
Mrs. Holberton. . . was one of the six young women recruited by the Army to
program. . . Eniac.
Lohr, Steve (2002), Scientist at Work: Frances Allen; Would-Be Math Teacher
Ended Up Educating a Computer Revolution, New York Times (6 August): F3;
http://tinyurl.com/mqhbpey11 (accessed 12 November 2013).
Lohr, Steve (2010), Inventor Whose Pioneer PC Helped Inspire Microsoft Dies,
New York Times (3 April): A1A3.
Obituary of H. Edward Roberts, inventor of the Altair, the first personal
computer.
MacFarlane, Alistair (2013), Ada Lovelace (18151852), Philosophy Now 96;
online at: http://philosophynow.org/issues/96/Ada Lovelace 1815-1852
Mahoney, Michael Sean (2011), Histories of Computing, ed. by Thomas Haigh
(Cambridge, MA: Harvard University Press).
Not so much a history of computing or computers as a history of computer
science. Chs. 10 and 11 are especially good on some of the recent
mathematical history.
Markoff, John (1992), Rear Adm. Grace M. Hopper Dies; Innovator in Computers
Was 85, New York Times (3 January); online at:
http://tinyurl.com/lthvdy412 (accessed 11 November 2013).
Markoff, John (2000), A Tale of the Tape from the Days When It Was Still Micro
Soft, New York Times (18 September): C1, C4; online at:
http://tinyurl.com/ox9upbh13 (accessed 12 November 2013).
About the day someone stole Bill Gatess software.
Markoff, John (2002), Edsger Dijkstra, 72, PHysicist Who Shaped Computer Era,
New York Times (10 August): A11; online at:
http://www.nytimes.com/2002/08/10/obituaries/10DIJK.html
Markoff, John (2005), What the Dormouse Said: How the Sixties Counterculture
Shaped the Personal Computer Industry (New York: Viking).
11 http://www.nytimes.com/2002/08/06/science/scientist-work-frances-allen-would-be-math-teacher-
ended-up-educating-computer.html
12 http://www.nytimes.com/1992/01/03/us/rear-adm-grace-m-hopper-dies-innovator-in-computers-was-
85.html
13 http://www.nytimes.com/2000/09/18/business/technology-a-tale-of-the-tape-from-the-days-when-it-
was-still-micro-soft.html
Reviewed in [Lanier, 2005].

Markoff, John (2011), It Started Digital Wheels Turning, New York Times Science
Times (8 November): D1, D4; online at:
http://tinyurl.com/dxqt6ca14 (accessed 12 November 2013).
In the age of steam,. . . [Charles Babbage] brainstormed a contraption that
inspired dreams of computers. Experts in London intend to spend a decade
piecing it together. Article contains a diagram of the Analytical Engine,
online at:
http://www.nytimes.com/interactive/2011/11/07/science/before-its-time-
machine.html
Martin, C. Dianne (1993), The Myth of the Awesome Thinking Machine,
Communications of the ACM 36(4) (April): 120133;
http://dx.doi.org/10.1145/255950.153587
[T]he thesis of this article [is] that early public attitudes toward computers
were shaped by the press.
Martin, Douglas (2008), David Caminer, 92, Dies; A Pioneer in Computers, New
York Times (29 June): 28;
http://www.nytimes.com/2008/06/29/technology/29caminer.html
Caminer worked for the first company to computerize its operationsthe
Lyons tea companyin 1951(!).
Martin, Douglas (2013), Mavis Batey, 92, Allied Code Breaker in World War II,
New York Times (24 November): 32;
http://www.nytimes.com/2013/11/23/world/europe/mavis-batey-world-war-ii-
code-breaker-dies-at-92.html
Batey worked with Turing at Bletchley Park.
Mathews, Walter M.; & Reifers, Kaye (1984), The Computer in Cartoons: A
Retrospective from The Saturday Review, Communications of the ACM 27(11)
(November): 11141119.
ORegan, Gerard (2008), A Brief History of Computing (Springer).
Papadimitriou, Christos H. (2001), The Sheer Logic of IT, review of
[Davis, 2012] (2000 original edition), American Scientist 89 (March-April): 168
171; online at:
http://www.americanscientist.org/bookshelf/pub/the-sheer-logic-of-it
Park, Edwards (1996), The Object at Hand, Smithsonian 26(11) (February): 20
23.
The incredible world of computers was born some 150 years ago, with a
clunky machine dreamed up by a calculating genius named Charles Babbage.
14 http://www.nytimes.com/2011/11/08/science/computer-experts-building-1830s-babbage-analytical-
engine.html
Press, Larry (1993), Before the Altair: The History of Personal Computing,
Communications of the ACM 36(9) (September): 2733; online at:
http://bpastudio.csudh.edu/fac/lpress/articles/hist.htm
Robinson, J. Alan (1994), Logic, Computers, Turing, and von Neumann, in
K. Furukawa, D. Michie, & S. Muggleton (eds.), Machine Intelligence 13: Machine
Historical comments, by the developer of the resolution method of automated
theorem proving, on the development of computers and the related history of
logic.
Ryan, Bob (1991), Dynabook Revisited with Alan Kay, Byte 16(2) (February):
203204, 206208.
Is the reign of the desktop computer about to end? A prediction of the future
of what is now known as laptop computers.
Sammet, Jean E. (1992), Farewell to Grace HopperEnd of an Era!,
Communications of the ACM 35(4) (April): 128131;
http://dx.doi.org/10.1145/129852.214846
Samuel, Arthur L. (1953), Computing Bit by Bit, or Digital Computers Made
Easy, Proceedings of the IRE [Institute of Radio Engineers] 41(10) (October):
12231230.
An introduction to computers, aimed at (radio) engineers who might not be
familiar with them; written by an IBM researcher who later became famous
for his work on computer checkers-players.
Shapiro, Stewart (1983), Remarks on the Development of Computability, History
and Philosophy of Logic 4: 203220.
On the mathematical history of computing.
Sieg, Wilfried (1994), Mechanical Procedures and Mathematical Experience, in
Alexander George (ed.), Mathematics and Mind (New York: Oxford University
Press): 71117;
http://repository.cmu.edu/cgi/viewcontent.cgi?article=1248&context=philosophy
See 1 for remarks on the mathematical history of computation.
Smith, Adam (1776), Of the Division of Labor, Book I, Ch. I, of The Wealth of
Nations, online at: http://www.econlib.org/library/Smith/smWNCover.html
(from that page, link to I.1. Of the Division of Labor)
Stein, Dorothy (1985), Ada: A Life and a Legacy (Cambridge, MA: MIT Press).
See also:
Kidder, Tracy (1985), Less (and More) than Meets the Eye, review of Stein
1985, New York Times Book Review (29 December): 67; online at:
http://www.nytimes.com/1985/12/29/books/less-and-more-than-meets-the-
eye.html
Stoll, Cliff (2006), When Slide Rules Ruled, Scientific American 294(5) (May):
8087.
Before electronic calculators, the mechanical slide rule dominated scientific

and engineering computation.
Note: The slide rule is an analog calculator!
Vardi, Moshe Y. (2011), Solving the Unsolvable, Communications of the ACM
54(7) (July): 5; online at:
http://www.inf.unibz.it/calvanese/teaching/tc/material/solving-the-unsolvable-
CACM-2011-07.pdf
Vardi, Moshe Y. (2013), Who Begat Computing?, Communications of the ACM
56(1) (January): 5; online at:
http://cacm.acm.org/magazines/2013/1/158780-who-begat-computing/fulltext
For a reply, see:
Copeland, Jack (2013), What Apple and Microsoft Owe to Turing,
Huff[ington] Post Tech/The Blog (12 August),
http://www.huffingtonpost.com/jack-copeland/what-apple-and-microsoft-
b 3742114.html
For further discussion of what a stored-program computer is, and whether it
was Turing or von Neumann (or someone else) who invented it, see Ch. 8.
von Neumann, John (1945), First Draft of a Report on the EDVAC, ed. by
Michael D. Godfrey, in IEEE Annals of the History of Computing 15(4) (1993):
2775; online at:
http://virtualtravelog.net.s115267.gridserver.com/wp/wp-content/media/2003-08-
TheFirstDraft.pdf (accessed 30 October 2013).
Waldrop, M. Mitchell (2001), The Origins of Personal Computing, Scientific
American (December): 8491.
Wheeler, David L. (1997), An Ancient Feud: Who Invented the Computer?,
Chronicle of Higher Education (3 October): B2.
On the ENIAC-ABC controversy, with a discussion of an attempt to replicate
the ABC.
Wright, Alex (2008), The Web Time Forgot, New York Times (17 June): F1, F4;
online at:
http://www.nytimes.com/2008/06/17/science/17mund.html
A 1934(!) version of a World Wide Web.
2. Websites:
Three pictures showing the evolution, from the ENIAC to the iPad, of computers
and their users: http://www.cse.buffalo.edu/rapaport/111F04/summary.html
On Charles Babbage:
(a) Charles Babbage Institute, Center for the History of Information Technology,
University of Minnesota, http://www.cbi.umn.edu/ (accessed 5 November
2012).
(b) Greenemeier, Larry (2008, April 24), 150-Year-Old Computer Brought to

Life [Slideshow],
http://www.scientificamerican.com/article.cfm?id=150-year-old-computer-
babbage (accessed 5 November 2012).
(c) Johnstone, Adrian (2014), Babbages Language of Thought, Plan 28 Blog
(28 November),
http://blog.plan28.org/2014/11/babbages-language-of-thought.html
i. From a website devoted to Charles Babbages many inventions.
ii. Plan 28 is working towards building Charles Babbages Analytical
Engine
(http://plan28.org/, accessed 5 January 2015)
(d) Lee, J.A.N. (1994), Charles Babbage
http://ei.cs.vt.edu/history/Babbage.html (accessed 5 November 2012).
(e) Sydell, Laura (2009, December 10), A 19th-Century Mathematician Finally
Proves Himself,
http://www.npr.org/templates/story/story.php?storyId=121206408&ft=1&f=1001
(f) Also see [Swade, 1993], Green 2005, and Campbell-Kelly 2010, above.
Computer Programming Used to Be Womens Work, Smart News Blog,
http://blogs.smithsonianmag.com/smartnews/2013/10/computer-programming-
used-to-be-womens-work/
Generations of Computers (blog),
http://generationsofcomputers.blogspot.com/2007/12/generation-of-
computers.html (accessed 5 November 2012).
Contains some good images of early calculating machines, including Pascals
and Leibnizs.
IBM Antique Attic (three-volume, illustrated exhibit of computing artifacts);
links located at: http://www-03.ibm.com/ibm/history/exhibits/index.html
Joyce, David E. (1997), The Mathematical Problems of David Hilbert
http://aleph0.clarku.edu/djoyce/hilbert/ (accessed 5 November 2012).
OConnor, J.J.; & Robertson, E.F. (2012), The MacTutor History of Mathematics
Archive, http://www-gap.dcs.st-and.ac.uk/history/ (accessed 6 November 2012),
including:
OConnor, J.J.; & Robertson, E.F. (1997), Gaspard Clair Fran cois Marie
Riche de Prony,
http://www-gap.dcs.st-and.ac.uk/history/Mathematicians/De Prony.html
and OConnor & Robinson 1998, above.
Sale, Tony, The Colossus Rebuild Project,
http://www.codesandciphers.org.uk/lorenz/rebuild.htm
(accessed 30 October 2013)
Seabrook, John (2007), Pieces of History (slideshow)
http://www.newyorker.com/online/2007/05/14/slideshow 070514 antikythera
Standage, Tom (1998), The Victorian Internet: The Remarkable Story of the
Telegraph and the Nineteenth Centurys On-Line Pioneers (New York: Walker
Publishing).
See review by John R. Alden in Smithsonian 29(10) (January 1999): 126127;
online at:
http://www.smithsonianmag.com/history-
archaeology/bookreview jan99 a.html (accessed 12 November 2013).
Tout, Nigel (20002012), Vintage Calculators Web Museum
http://www.vintagecalculators.com/ (accessed 6 November 2012).
Wells, Benjamin (2003), The Architecture of Colossus, the First PC (abstract)
http://www.stanford.edu/class/ee380/Abstracts/040204.html
See also Copeland et al. 2010, above.
Figure 6.2: 2011,

c Hilary B. Price
Chapter 7
What Is an Algorithm?

Thou must learne the Alphabet, to wit, the order of the Letters as they
stand.. . . Nowe if the word, which thou art desirous to finde, begin with (a) then
looke in the beginning of this Table, but if with (v) looke towards the end. Againe,
if thy word beginne with (ca) looke in the beginning of the letter (c) but if with
(cu) then looke toward the end of that letter. And so of all the rest. &c. Robert
Cawdrey, A Table Alphabeticall, conteyning and teaching the true writing, and
understanding of hard usuall English wordes (1604), cited in [Gleick, 2008, 78].
Editor:
We are making this communication intentionally short to leave as much room as
possible for the answers. 1. Please define Algorithm. 2. Please define Formula.
3. Please state the difference. T. WANGSNESS, J. FRANKLIN
TRW Systems, Redondo Beach, California
[Wangsness and Franklin, 1966].
This nation is built on the notion that the rules restrain our behavior. . . .
[Editorial, 2006]
Figure 7.1: 2004

c by ??
223
224 CHAPTER 7. WHAT IS AN ALGORITHM?

1. Required:
Henkin, Leon (1962), Are Logic and Mathematics Identical?, Science 138(3542)
(November 16): 788794.
Read pp. 788791; skim the rest
An excellent, brief overview of the history of logic and the foundations of
mathematics that led up to Turings analysis.
Davis, Martin (1987), Mathematical Logic and the Origin of Modern Computers,
Studies in the History of Mathematics; reprinted in Rolf Herken (ed.), Universal
Turing Machine: A Half-Century Survey; Second Edition (Vienna: Springer-Verlag,
1995): 135158.
Overlaps and extends Henkins history, and provides a useful summary of
[Turing, 1936].
Herman, Gabor T. (1993), Algorithms, Theory of, in Anthony S. Ralston &
Edwin D. Riley (eds.), Encyclopedia of Computer Science, 3rd edition (New York:
Van Nostrand Reinhold): 3739.
Discussion
of the informal notions of algorithm and effective computability;
good background for Turing 1936.
Either:
Soare, Robert I. (1996), Computability and Recursion, Bulletin of
Symbolic Logic 2(3) (September): 284321.
or:
Soare, Robert I. (1999), The History and Concept of Computability,
in E.R. Griffor (ed.), Handbook of Computability Theory (Amsterdam:
Elsevier): 336.
a revised and shortened form of Soare 1996.
Read 13, 4.54.6; skim the rest
Browse through the Examples of Algorithms at:
http://www.cse.buffalo.edu/rapaport/584/whatisanalg.html
Haugeland, John (1981), Semantic Engines: An Introduction to Mind Design, in
John Haugeland (ed.), Mind Design: Philosophy, Psychology, Artificial Intelligence
(Cambridge, MA: MIT Press): 95128.
A good description of the syntax and semantics of formal systems and their
relationship to Turing Machines.
Boehm, C.; & Jacopini, G. (1966), Flow Diagrams, Turing Machines, and
Languages with Only Two Formation Rules, Communications of the ACM 9(5):
366371.
Uses flowcharts to prove that go to statements are eliminable from computer

programs in favor of sequence, selection, and loops. An important paper, but
not for the faint of heart!
7.2 Introduction
We are examining the questions of whether computer science is a science (or something
else, such as a branch of engineering or some combination of both science and
engineering) and what its subject matter is: Does it study computers: (physical)
devices that computeor computing: the algorithmic processes that computers do?
(Or perhaps it studies something else, such as information, or information processing.)
In the former case, we need to ask what computers are; in the previous chapter, we
began that investigation by looking at the history of computers. In this chapter, we ask
what computing is. Then we will be in a better position to return to our question of
what a computer is, looking at it from a philosophical, rather than a historical, point of
view. And after that, we will return to the question of what computing is, again looking
at some philosophical issues.
7.3 What Is Computation?

Many now view computation as a fundamental part of nature, like atoms
or the integers. [Fortnow, 2010, 2]
Although I have been talking about computing, other terms that are used to
describe more or less the same territory are computation and algorithms. It may
be worth a brief detour into the etymology of these and some related terms.
7.3.1 compute
According to the Oxford English Dictionary (OED)1 , the verb compute comes from
the Latin verb computare, meaning to calculate, reckon, to count up. But when
people talk about computing today, they mean a great deal more than mere counting;
computing has come to include everything we can do with computers, from text
processing to watching videos and playing games, so, clearly, the meaning has been
extended to include non-numerical reckoning.
The Latin word computare itself comes from the Latin morpheme com, meaning
together with, and the Latin word putare, meaning to cleanse, to prune, to reckon,
to consider, think (and putare came from a word meaning clean, pure). So, in
ancient Rome at least, to compute seems to have meant, more or less, something
like: to consider or think about things together, to clear things up together, or
maybe to reckon with (something).
7.3.2 reckon
The verb reckon originally meant to give an account of, recount; to tell; to describe,
and later came to mean to count, to calculate. Reckon is from an Indo-European2
root rek, possibly meaning to reach or to tell, to narrate, to say (as in to recite
1 http://www.oed.com/view/Entry/37974
2 http://en.wikipedia.org/wiki/Indo-European languages
7.4. WHAT IS COMPUTATION? 227
or to recount). These meanings, in turn, may derive from an earlier meaning to

arrange, to put right, to move in a straight line.3
7.3.3 count, calculate, figure

The origins of count, calculate, and figure are also interesting.
Count also came from computare and originally meant to enumerate, to
recite a list (and, as we just saw, recite is probably related to reckon). Note that
when you count, you recite a list of number words.
Calculate came from Latin calculus. This certainly did not mean the contents
of a certain high school or college course that studies the branch of mathematics
concerned with differentiation and integration, and invented by Newton and Leibniz
in the 17th century! Rather, it meant pebble or small stone(!), since counting was
done with stones originally! (Even today, a calculus in medicine is an accumulation
of minerals in the body, forming a small, stone-like object. The root calc came from
calx, meaning chalk, limestone, and is related to calcium.
The verb to figure means to use figures to reckon, where the noun figure seems
originally to have meant numerical symbol. The earliest citation for figure in the
OED is from 1225, when it meant numerical symbol. A citation from 1250 has the
meaning embodied (human) form. And a citation from 1300 has the more general
meaning of shape. (This conversion of the noun figure to a verb is an example of
what the computer scientist Alan Perlis meant when he joked, In English, every word
can be verbed.)
7.3.4 computation
The bottom line seems to be this: Computation originally meant something very
closely related to our modern notion of symbol (i.e., shape) manipulation, which is
another way of describing syntax. (See Ch. 22, as yet unwritten.)
Now that we know how we got the word computation, well turn to what the
object computation is.
7.4 What Is Computation?

To understand what computation is, we first need to understand what a (mathematical)
function is.
7.4.1 What Is a Function?

7.4.1.1 Two Meanings
The English word function has at least two, very different meanings. The ordinary,
everyday meaning is, roughly, purpose. Instead of asking, What is the purpose of
this button?, we might say, What is the function of this button? What does it do?.
3 http://etymonline.com/?term=reckon. Cf. http://www.utexas.edu/cola/centers/lrc/ielex/R/P1613.html
But we will be interested in its other, mathematical meaning, as when we say that
some dependent variable is a function of i.e., depends onsome independent
variable. (As it turns out, this technical sense of the word was, first of all, initiated
by our old friend Leibniz and, second, was an extension of its other meaning; for the
history of this, see the OED entry on the noun function, in its mathematical sense
(sense 6; http://www.oed.com/view/Entry/75476. On the history of the concept of
function, see [OConnor and Robertson, 2005].)
7.4.1.2 Functions Described Extensionally

Many introductory textbooks define a (mathematical) function as an assignment of
values to inputs, but they never define what an assignment is.4 A much more rigorous
way of defining a function is to give a definition based on set theory.
A function (described extensionally)5 is a set of input-output pairs such that no
two of them have the same input (or first element). In other words, a function is what
is called a binary relation: a set of ordered pairs of elements from two sets. (The
two sets can be the same one; you can have a relation among the members of a single
set.) But it is a special kind of binary relation in which no two, distinct members of
the relation have the same first element (but different second elements). That is, if you
have what you think are two different members of a function, and if they have the same
first element, then they also have the same second element, and so it only seemed as if
they were two members; they were really one and the same member, not two different
ones. As a rule of thumb, a binary relation is a function if same input implies same
output.
The logically correct way to say this, in mathematical English, is:
Let A, B be sets.
Then f is a function from A to B
=def (i.e., means by definition)
1. f is a binary relation from A to B,
and
2. for all members a of A,
and for all members b of B
and for all members b0 of B6
if : (a, b) is a member of f
and if : (a, b0 ) is also a member of f
then: b = b0
Because we are considering a binary relation as a set of ordered pairs, lets write the
members of a binary relation from A to B as ha, bi, where a A and b B. Here are
some examples of functions in this extensional sense:
4 Itis something like a mapping. Such an assignment is not quite the same thing as an assignment of
a value to a variable in a programming language or in a system of logic.
5 http://www.cse.buffalo.edu/rapaport/intensional.html
6 Keep in mind that b is allowed to be the same as b0 !
1. f = {h0, 1i, h1, 2i, h2, 3i, . . .}

This is sometimes written: f (0) = 1, f (1) = 2, . . .
2. g = {h0, 0i, h1, 2i, h2, 4i, h3, 6i, . . .}

This is sometimes written: g(0) = 0, g(1) = 2, . . .
3. E(y) = hm, di
=h
((((19 (y mod 19)) + (y/100) ((y/100)/4) (((8 (y/100)) + 13)/25) +
15) mod 30) (((y mod 19) + (11 (((19 (y mod 19)) + (y/100)
((y/100)/4) (((8 (y/100)) + 13)/25) + 15) mod 30)))/319) + (((2
((y/100) mod 4))+(2((y mod 100)/4))((y mod 100) mod 4)(((19
(y mod 19)) + (y/100) ((y/100)/4) (((8 (y/100)) + 13)/25) + 15) mod
30) (((y mod 19) + (11 (((19 (y mod 19)) + (y/100) ((y/100)/4)
(((8 (y/100)) + 13)/25) + 15) mod 30)))/319) + 32) mod 7) + 90)/25 ,
((((19 (y mod 19)) + (y/100) (y/100)/4 (((8

(y/100))+13)/25)+15) mod 30)(((y mod 19)+(11(((19(y mod 19))+
(y/100)(y/100)/4(((8(y/100))+13)/25)+15) mod 30)))/319)+(((2
((y/100) mod 4)) + (2 ((y mod 100)/4)) ((y mod 100) mod 4) (((19
(y mod 19)) + (y/100) (y/100)/4 (((8 (y/100)) + 13)/25) + 15) mod
30) (((y mod 19) + (11 (((19 (y mod 19)) + (y/100) (y/100)/4 (((8
(y/100)) + 13)/25) + 15) mod 30)))/319) + 32) mod 7) + ((((19 (y mod
19)) + (y/100) (y/100)/4 (((8 (y/100)) + 13)/25) + 15) mod 30)
(((y mod 19) + (11 (((19 (y mod 19)) + (y/100) (y/100)/4 (((8
(y/100)) + 13)/25) + 15) mod 30)))/319) + (((2 ((y/100) mod 4)) + (2
((y mod 100)/4)) ((y mod 100) mod 4) (((19 (y mod 19)) + (y/100)
(y/100)/4 (((8 (y/100)) + 13)/25) + 15) mod 30) (((y mod 19) + (11
(((19 (y mod 19)) + (y/100) (y/100)/4 (((8 (y/100)) + 13)/25) +
15) mod 30)))/319) + 32) mod 7) + 90)/25 + 19) mod 32
i
This function takes as input a year y and outputs an ordered pair consisting of
the month m and day d that Easter falls on in year y.[Stewart, 2001]7
4. Here is a finite function (i.e., a function with a finite number of members

remember: a function is a set, so it has members):
h = {hyes, print helloi, hno, print byei, hinput 6= yes & input 6= no, print sorryi}
The idea behind this function is that it prints hello, if the input is yes;
it prints bye, if the input is no;
and it prints sorry, if the input is neither yes nor no.
7 See also http://techsupt.winbatch.com/TS/T000001048F35.html (accessed 11 December 2013).
5. A partial function (i.e., a function that has no outputs for some possible inputs):
1 1 1 1
k = {. . . , h2, i , h1, i , h1, i , h2, i , . . .}
2 1 1 2
Here, k(0) is undefined.
7.4.1.3 Functions Described as Machines

Sometimes, functions are characterized as machines that take input into a
black box with a crank that mysteriously converts the input into output:
In this picture, f is a machine into which you put a; you then turn a crank (clockwise,
lets suppose); f then grinds away at the input by means of some mysterious
mechanism; and finally the machine outputs b (i.e., f (a)). But this view of a function
as being something active or dynamic that changes something is incorrect.8
8 See http://www.askphilosophers.org/question/1877.
Despite what you may have been told elsewhere (I was told this in high school),
this is NOT what a function is! A function, as we have seen, is merely the set of input-
output pairs. So, what is the machine? It is a computer! (And the mysterious gears
hidden inside the black box implement an algorithm that computes the function.) In
fact, one problem with this machine metaphor for a function, as we will see, is that
not all functions can be computed by algorithms; that is, there are functions for which
there are no such function machines.9
7.4.1.4 Functions Described Intensionally

Sometimes, functions are described (intensionally)10 but not uniquely described
by formulas. Here are some examples:
1. f (i) = 2i
2. g(i) = i + 1
3. g0 (i) = 2i i + 7/(3 + 4)
Note that both g and g0 describe the same function; that is, g = g0 , even though
their formulas are different.

hello if i = yes
4. h(i) = bye if i = no
sorry, otherwise

5. if i 6= 0, then k(i) = 1i .
Although formulas may look a lot like algorithms, they are not the same thing.
Consider, for example, the formula 2 + 4 5: Without an explicit statement of a
rule telling you whether to multiply first or to add first, there is no way of knowing
whether the number expressed by that formula is 30 or 22. Such a rule, however,
would constitute an algorithm telling you how to calculate the value of the formula. Or
consider the formula 2x + 1: Should you first calculate 2x and then add 1? Or should
you store 1 somewhere (say, by writing it on a piece of paper), then calculate 2x, and
finally add 2x to 1? Of course, the commutative law of addition tells us that, in this
case, it doesnt matter in which order you compute the value of the formula, but here
we have a clear case of only one formula but (at least) two distinct algorithms.
Perhaps an even clearer example is function E, abovethe one that tells you when
Easter occurs. I dare you to try to use this formula to find out when Easter will occur
9 Interestingly, Godel made this observation in the 1930s in an unpublished paper!
[Turing] has shown that the computable functions defined in this way [that is, in terms of
Turing machines] are exactly those for which you can construct a machine with a finite
number of parts which will do the following thing. If you write down any number n1 ,. . . ,nr on
a slip of paper and put the slip into the machine and turn the crank, then after a finite number
of turns the machine will stop and the value of the function for the argument n1 ,. . . ,nr will be
printed on the paper. [Godel, 930s, 168], cited in [Shagrir, 2006a, 3] and in [Soare, 2009,
3.2, p. 374].
So, the machine pictured in Fig. 7.4.1.3 is a Turing machine!
10 See http://www.cse.buffalo.edu/rapaport/intensional.html
next year! Where would you even begin? (A related matter is knowing whether the
formula is even correct! Well explore this issue in Ch. 16.)11
(For a good discussion of the difference between formulae and algorithms, see the
question asked in [Wangsness and Franklin, 1966] (quoted in full at the beginning of
this chapter) and the answers in [Huber, 1966] and [Knuth, 1966]; Knuths answer is
a commentary on Hubers. Hubers answer, roughly, is that an algorithm is a set of
instructions for computing the value of a function by executing (or carrying out, or
following) the instructions, whereas a formula is an expression describing the value of
a function; it can be evaluated (that is, the value of the function can be determined
from the formula) but not executed (because a formula does not come equipped with an
algorithm for telling you how to evaluate it). In a sense, a formula is static, whereas
an algorithm is (potentially) dynamic.)
Functions describable by formulas are not the only kind of functions. There are
functions without formulas for computing them; that is, there are functions such that
the gears of their machines work by magic! For example, there are table look-
up functions, where the only way to identify the correct output for a given input is to
look it up in a table (rather than to compute it); usually, this is the case when there is no
lawlike pattern relating the inputs and the outputs. Of course, there are non-computable
functions, such as the Halting Problem (well have more to say on what this in 7.8).
And there are random functions.
In fact, one of the central purposesperhaps the central quesionof computer
science (or, at least, of the non-engineering portion of the field) is to figure out which
functions do have algorithms for computing them! This includes non-mathematical
functions, such as the (human) cognitive functions that take as input sensory
information from the environment and produce as output (human, cognitive) behavior.
To express this another way, the subfield of computer science known as AI can be
considered as having as its purpose figuring out which such cognitive functions are
computable.
7.4.1.5 Computable Functions

So we have two concepts: function and algorithm. We have given a careful definition
of the mathematical notion of function. We have not yet give a careful definition of the
mathematical notion of algorithm, but we have given some informal characterizations
(and we will look at others in 7.5, below). We can combine them as follows:
A function f is computable will mean, roughly, that there is an

algorithm that computes f .
This is only a rough definition or characterization because, for one thing, we havent yet
defined algorithm. But, assuming that an algorithm is, roughly, a set of instructions
for computing the output of the function, then it makes sense to define a function as
being computable if we can. . . well. . . compute it!
11 I created this formula by working backwards from the algorithm given in [Stewart, 2001], so its quite
possible that I introduced a typographical error! Even if I didnt, I am assuming that [Stewart, 2001]s
algorithm is correct. And that could be a big assumption.
So, a function f is computable if there is an algorithm A f such that, for all inputs
i, A f (i) = f (i). That is, a function f is computable by an algorithm A f if both f and A f
have the same input-output behavior (that is, if both define the same binary relation,
or set of input-output pairs). Moreover, A f must specify how f s inputs and outputs are
related. So, whereas a function only shows its input-output pairs but is silent about how
they are related, an algorithm for that function must say more. It must be a procedure,
or a mechanism, or a set of intermediate steps or instructions that either transforms the
input into the output or else shows you explicitly how to find the output by starting with
the input, or how to get from the input to the output. Algorithms shouldnt be magic or
merely arbitrary.
It seems easy enough to give examples of algorithms for some of the functions
listed earlier:
1. A f (i) =
input i;
add 1 to i;
output result.
2. Ag (i) =
input i;
multiply i by 2;
output result.
3. A0f (i) =
input i;
mulitply i by 2;
call the result x;
subtract i from x;
add 3 to 4;
call the result y;
divide 7 by y;
add x to y;
output result.
4. For E(m, d), see the English algorithm in [Stewart, 2001] or the computer
program online at
http://techsupt.winbatch.com/TS/T000001048F35.html.
Note that, even though that algorithm may not be easy to follow, it is certainly
much easier than trying to compute the output of E from the formula. (For one
thing, the algorithm tells you where to begin!)
Can this notion of algorithm be made more precise? How?

7.5 Algorithms Made Precise

The meaning of the word algorithm, like the meaning of most other words
commonly used in the English language, is somewhat vague. In order to
have a theory of algorithms, we need a mathematically precise definition
of an algorithm. However, in giving such a precise definition, we run
the risk of not reflecting exactly the intuitive notion behind the word.
[Herman, 1983, 57]
Before anyone attempted to define algorithm, many algorithms were in use

both by mathematicians (e.g., ancient Babylonian procedures for finding lengths
and for computing compound interestsee [Knuth, 972a], Euclids procedures for
construction of geometric objects by compass and straightedge [Toussaint, 1993],
Euclids algorithm for computing the greatest common divisor of two integers) as
well as by ordinary people (e.g., the algorithms for simple arithmetic with Hindu-
Arabic numerals [Robertson, 1979]).12 In fact, the original, eponymous use of the
word referred to those arithmetic rules as devised by Abu Abdallah Muhammad ibn
Musa Al-Khwarizmi, a Persian mathematician who lived ca. 780850 AD, i.e., about
1200 years ago.13
When David Hilbert investigated the foundations of mathematics, his followers
began to try to make the notion of algorithm precise, beginning with discussions of
effectively calculable, a phrase first used by Jacques Herbrand in 1931 [Gandy, 1988,
68] and later taken up by Alonzo Church [Church, 1936b] and Stephen Kleene
[Kleene, 1952], but left largely undefined, at least in print.14
J. Barkley Rosser [Rosser, 1939, 55 (my italics and enumeration)] made an effort
to clarify the contribution of the modifier effective:
Effective method is here used in the rather special sense of a method
each step of which is [1] precisely predetermined and which is [2] certain
to produce the answer [3] in a finite number of steps.
But what, exactly, does precisely predetermined mean? And does finite number
of steps mean that the written statement of the algorithm has a finite number of
instructions, or that, when executing them, only a finite number of tasks must be
performed? (In other words, what gets counted: written steps or executed instructions?
One written stepfor i := 1 to 100 do x := x + 1can result in 100 executed
12 This section is adapted from [Rapaport, 2012b, Appendix].
13 What sounds like his last name, Al-Khwarizmi, really just means something like the
person who comes from Khwarizm; the Aral Sea. . . was once known as Lake Khwarizm
[Knuth, 1985, 171]. See http://www-groups.dcs.st-and.ac.uk/history/Mathematicians/Al-Khwarizmi.html
and [Crossley and Henry, 1990].
14 [Church, 1956] calls it an informal notion: See pp. 50 (and our 10.5.1 below), 52 (including notes
118 [an effective method of calculating, especially if it consists of a sequence of steps with later steps
depending on results of earlier ones, is called an algorithm] and 119 [an effective method of computation,
or algorithm, is one for which it woud be possible to build a computing machine, by which he means a
Turing machine]), 83, 99n.183 [a procedure. . . should not be called effective unless there is a predictable
upper bound of the number of steps that will be required], and 326 (note 535); cf. [Manzano, 1997, 219
220]. See also Sieg 1997; another explication of effective is in [Gandy, 1980, 124], which well return to
in Ch. 10.
7.5. ALGORITHMS MADE PRECISE 235
instructions. And one written stepwhile true do x := x + 1can result in an

infinite number of executed instructions!)
Much later, after Turings, Churchs, Godels, and Posts CITE? precise
formulations and during the age of computers and computer programming,
A.A. Markov, Stephen Kleene, and Donald Knuth also gave slightly less vague, though
still informal, characterizations.
According to Markov [Markov, 1954, 1], an algorithm is a computational process
satisfying three (informal) properties:
1. being determined (carried out according to a precise prescription. . . leaving

no possibility of arbitrary choice, and in the known sense generally understood),
2. having applicability (The possibility of starting from original given objects

which can vary within known limits), and
3. having effectiveness (The tendency of the algorithm to obtain a certain result,

finally obtained for appropriate original given objects).
These are a bit obscure: Being determined may be akin to Rossers precisely
predetermined. But what about being applicable? Perhaps this simply means that
an algorithm must not be limited to converting one, specific input to an output, but must
be more general. And Markovs notion of effectiveness seems restricted to only the
second part of Rossers notion, namely, that of producing the answer. There is no
mention of finiteness, unless that is implied by being computational.
In his undergraduate-level, logic textbook, Kleene [Kleene, 1967] elaborates on
the notions of effective and algorithm that he left unspecified in his earlier,
classic, graduate-level treatise on metamathematics. He continues to identify effective
procedure with algorithm [Kleene, 1967, 231], but now he offers a characterization
of an algorithm as
1. a procedure (i.e., a finite set of rules or instructions) that
2. in a finite number of steps answers a question, where
3. each instruction can be followed mechanically, like robots; no insight or

ingenuity or invention is required,
4. each instruction tell[s] us what to do next, and
5. the algorithm enable[s] us to recognize when the steps come to an end

[Kleene, 1967, 223].
Knuth [Knuth, 1973, Basic Concepts: 1.1: Algorithms, pp. xiv9, esp. pp. 19]
goes into considerably more detail, albeit still informally. He says that an algorithm is
a finite set of rules which gives a sequence of operations for solving a specific type of
problem, with five important features [Knuth, 1973, 4]:
1. Finiteness. An algorithm must always terminate after a finite number of steps

[Knuth, 1973, 4].
Note the double finiteness: A finite number of rules in the text of the
algorithm and a finite number of steps to be carried out. Moreover,
algorithms must halt. (Halting is not guaranteed by finiteness; see
point 5, below.) Interestingly, Knuth also says that an algorithm is a
finite computational method, where a computational method is a
procedure that only has the next four features [Knuth, 1973, 4].
2. Definiteness. Each step. . . must be precisely defined; the actions to be carried
out must be rigorously and unambiguously specified. . . [Knuth, 1973, 5].
This seems to be Knuths analogue of the precision that Rosser and
Markov mention.
3. Input. An algorithm has zero or more inputs [Knuth, 1973, 5].
Curiously, only Knuth and Markov seem to mention this explicitly,
with Markovs applicability property suggesting that there must be
at least one input. Why does Knuth say zero or more? Presumably,
he wants to allow for the possibility of a program that simply outputs
some information. On the other hand, if algorithms are procedures
for computing functions, and if functions are regular sets of input-
output pairs (regular in the sense that the same input is always
associated with the same output), then algorithms would always have
to have input. Perhaps Knuth has in mind the possibility of the
input being internally stored in the computer rather than having to
be obtained from the external environment. Or perhaps this is how
constant functions (functions whose output is constant, no matter
what their input is) are handled. It is worth noting, however, that
[Hartmanis and Stearns, 1965, 288]the founding document of the
field of computational complexityallows their multi-tape Turing
machines to have at most one tape, which is an output-only tape; there
need not be any input tapes.
4. Output. An algorithm has one or more outputs [Knuth, 1973, 5].
That there must be at least one output echoes Rossers property
(2) (certain to produce the answer) and Markovs notion (3)
of effectiveness (a certain result). But Knuth characterizes
outputs as quantities which have a specified relation to the inputs
[Knuth, 1973, 5]; the relation would no doubt be the functional
relation between inputs and outputs, but, if there is no input, what
kind of a relation would the output be in? Very curious!15
5. Effectiveness. This means that all of the operations to be performed in the
algorithm must be sufficiently basic that they can in principle be done exactly and
in a finite length of time by a man [sic] using pencil and paper [Knuth, 1973,
6].
15 For an example of an algorithm that has an input but no output, see 17; see also
[Copeland and Shagrir, 2011, 230231].
7.5. ALGORITHMS MADE PRECISE 237
Note, first, how the term effective has many different meanings
among all these characterizations of algorithm, ranging from
it being an unexplained term, through being synonymous with
algorithm, to naming very particularand very different
properties of algorithms.
Second, it is not clear how Knuths notion of effectiveness differs
from his notion of definiteness; both seem to have to do with the
preciseness of the operations.
Third, Knuth brings in another notion of finiteness: finiteness
in time. Note that an instruction to carry out an infinite sequence
of steps in a finite time could be accomplished by doing each step
twice as fast as the previous step; or each step might only take a finite
amount of time, but the number of steps required might take longer
than the expected life of the universe (as in computing a perfect, non-
losing strategy in chess; cf. [Zobrist, 2000, 367]). These may have
interesting theoretical implications (which we will explore in Ch. 11)
but do not seem very practical. Knuth [Knuth, 1973, 7] observes that
we want good algorithms in some loosely-defined aesthetic sense.
One criterion of goodness is the length of time taken to perform the
algorithm. . . .
Finally, the gold standard of a [hu]man using pencil
and paper seems clearly to be an allusion to Turings analysis
[Turing, 1936], which we will examine in great detail in the next
chapter.
We can summarize these informal observations as follows:
An algorithm (for executor E to accomplish goal G) is:
1. a procedure, i.e., a finite set (or sequence) of statements (or rules, or
instructions), such that each statement is:
(a) composed of a finite number of symbols (or marks) from a finite
alphabet
(b) and unambiguous for Ei.e.,
i. E knows how to do it
ii. E can do it
iii. it can be done in a finite amount of time
iv. and, after doing it, E knows what to do next
2. which procedure takes a finite amount of time, i.e., halts,
3. and that ends with G accomplished
But the important thing to note is that the more one tries to make precise these
informal requirements for something to be an algorithm, the more one recapitulates
Turings motivation for the formulation of a Turing machine. We will look at exactly
what Turing did in Ch. 8. But first we are going to look a bit more deeply into the
current view of computation.
7.6 Four Great Insights of Computer Science

In this section, we will look at four great insights of computer science: The first
three help make precise the vague notion of algorithm that we have been looking
at. The fourth links the vague notion with a precise one. Together, they define the
smallest possible language in which you can write any procedure for any computer.
(And by computer here, I mean anythingmachine or humanthat can execute an
algorithm.)
7.6.1 Bacons, Leibnizs, Morses, Booles, Ramseys, Turings, and

Shannons Insight
The first great insight is this:
All the information about any computable problem can be

represented using only two nouns: 0 and 1
Rather than enter into a complicated and controversial historical investigation of who
is responsible for this insight, I have chosen instead to simply list the names of some
of the people who contributed to it:
Sir Francis Bacon, around 1605, developed an encoding of the alphabet by

any objects capable of a twofold difference.16 And, of course, once youve
represented the alphabet in a binary coding, then anything capable of being
represented in text can be similarly encoded.
Our old friend, Leibniz gave an Explanation of Binary Arithemetic in 1703.17
Famously, Samuel F.B. Morse not only invented the telegraph but also (following
in Bacons footsteps) developed his eponymous, binary code in the mid-1800s.18
The philosopher Frank P. Ramsey, in a 1929 essay on a language

for discussing. . . factsperhaps something like Leibnizs characteristica
universalis (which we discussed in 6.5 [and possibly also in Ch. 3])
suggested that all [of the terms of the language] may be best symbolized
by numbers. For instance, colours have a structure, i which any given
colour maybe assigned a place by three numbers. . . . Even smells may be so
treated. . . [Ramsey, 1929, 101102]. PERHAPS ADD MORE EXAMPLES?
SEE http://www.cse.buffalo.edu/rapaport/111F04/greatidea1.html
In 1936, as we will see in Ch. 8, Turing made essential use of 0 and 1 in the
development of Turing machines.
16 [Cerf, 2015, 7] and http://home.hiwaay.net/paul/bacon/advancement/book6ch1.html
17 http://www.leibniz-translations.com/binary.htm
18 http://en.wikipedia.org/wiki/Morse code. Arguably, however, Morse code (traditionally conceived as
having only two symbols, dot and dash) is not strictly binary, because there are blanks, or time-lapses,
between those symbols; CITE GLEICK.
7.6. FOUR GREAT INSIGHTS OF COMPUTER SCIENCE 239
Finally, the next year, Claude Shannon (in his development of the mathematical
theory of information) used The symbol 0. . . to represent. . . a closed circuit,
and the symbol 1. . . to represent. . . an open circuit [Shannon, 1937, 4] and then
showed how propositional logic could be used to represent such circuits.
There is nothing special about 0 or 1. As Bacon emphasized, any other bistable19

pair suffices, as long as they can flip-flop between two easily distinguishable states,
such as on/off, magnetized/de-magnetized, high voltage/low voltage, etc.
Strictly speaking, these can be used to represent discrete things; continuous things
can be approximated to any desired degree, however. This limitation to two nouns is
not necessary: Turings original theory had no restriction on how many symbols there
were. There were only restrictions on the nature of the symbols (they couldnt be too
close to each other; that is, they had to be distinguishable) and that there be only
finitely many.
But, if we want to have a minimal language for computation, having only
two symbols suffices, and making them 0 and 1 (rather than, say, A and B) is
mathematically convenient.
7.6.2 Turings Insight

So we have two nouns for our minimal language. Turing is also responsible for
providing the verbs of our minimal language. Our second great insight is this:
Every algorithm can be expressed in a language for a computer

(namely, a Turing machine) consisting of:
an arbitrarily long, paper tape divided into squares (like a roll of

toilet paper, except you never run out [Weizenbaum, 1976]),
with a read/write head,
whose only nouns are 0 and 1,
and whose only five verbs (or basic instructions) are:
1. move-left-1-square
2. move-right-1-square
(These could be combined into a single verb (that is, a
function) that takes a direct object (that is, a single input
argument): move(location).)
3. print-0-at-current-square
4. print-1-at-current-square
5. erase-current-square
(These could be combined into another, single, transitive
verb: print(symbol), where symbol could be either 0, 1,
19 I.e., something that can be in precisely one of two states; http://en.wikipedia.org/wiki/Bistability.
or blank (here, erasing would be modeled by printing a

blank).)20
So, we can get by with either only two (slightly complex) verbs or five (slightly
simpler) verbs. But either version is pretty small.
7.6.3 Boehm and Jacopinis Insight

We have two nouns and perhaps only two verbs. Now we need some grammatical rules
to enable us to put them together. The software-engineering concept of structured
programming does the trick. This is a style of programming that avoids the use of
the go to command. In early programming languages, programmers found it useful
to jump or go to another location in the program, sometimes with the ability to
return to where the program jumped from, but sometimes not. This resulted in what
was sometimes called spaghetti code, because, if you looked at a flowchart of the
program, it consisted of long, intertwined strands of code that were hard to read and
harder to ensure that they were correct. Edsger W. Dijkstra wrote a letter to the editor
of the Communications of the ACM, that was headlined Go To Statement Considered
Harmful [Dijkstra, 1968], arguing against the use of such statements. This resulted in
an attempt to better structure computer programs so that the use of go to could be
minimized; Corrado Bohm and Giuseppe Jacopini proved that it could be completely
eliminated [Bohm and Jacopini, 1966]. This gives rise to the third insight (and the third
item needed to form our language):
Only three rules of grammar are needed to combine any set of basic
instructions (verbs) into more complex ones:
1. sequence:
first do this; then do that
2. selection (or choice):
IF such-&-such is the case,
THEN do this
ELSE do that
3. repetition (or looping):
WHILE such & such is the case
DO this
. . . where this and that can be:

20 Deciding how to count the number of actions is an interesting question. [Gurevich, 1999, 99100] points
out that at one step, a Turing machine may change its control state, print a symbol at the current tape cell[,]
and move its head. . . . One may claim that every Turing machine performs only one action a time, but that
action [can] have several parts. The number of parts is in the eye of the beholder. You counted three parts.
I can count four parts by requiring that the old symbol is erased before the new symbol is printed. Also,
the new symbol may be composed, e.g. 12. Printing a composed symbol can be a composed action all by
itself.
any of the basic instructions, or

any complex instruction created by application of any grammatical
rule.
Digression:
For readers who are familiar with the structure of recursive definitions,
what we have here is a recursive definition of the notion of an
instruction: The base cases of the recursion are the primitive verbs of the
Turing machine (move(location) and print(symbol)), and the recursive
cases are given by sequence, selection, and repetition.
More generally, a recursive definition of some concept C consists
of two parts. The first part, called the base case, gives you some
explicit examples of C. These are not just any old examples, but are
considered to be the simplest, or most basic or atomic, instances of C
the building blocks from which all other, more complex instances of C can
be constructed.
The second part of a recursive definition of C consists of rules
(algorithms, if you will!) that tell you how to construct those more complex
instances of C. But these rules dont simply tell you how to construct
the complex instances from the base cases. Rather, they tell you how to
construct the complex instances from any instances that have already been
constructed. The first complex instances, of course, will be constructed
directly from the base cases. But others, even more complex, will be
constructed from the ones that were constructed directly from the base
cases, and so on. What makes such a definition recursive is that simpler
instances of C recur in the definitions of more complex instances.
So, the base case of a recursive definition tells you how to begin.
And the recursive case tells you how to continue. But recursive definitions
sometimes seem to be circular: After all, we seem to be defining C in terms
of C itself! But really we are defining new (more complex) instances
of C in terms of other, older (that is, already constructed), or simpler
instances of C, which is not circular at all. (It would only be circular if the
base cases were somehow defined in terms of themselves. But they are
not defined; they are given, by fiat.)21 End digression
There are optional, additional instructions and grammatical rules:

1. An exit instruction, allowing the program to exit from a loop under certain
conditions, before the body of the loop is completed; this is provided for
simplicity or elegance.
21 Not only is recursion considered to be the core of the concept of computation, but it has been argued that
it is also the core of the concept of language [Hauser et al., 2002]: The faculty of language. . . in the narrow
sense. . . only includes recursion and is the only uniquely human component of the faculty of language.
However, this is a highly controversial claim; to follow the debate, see [Pinker and Jackendoff, 2005,
Fitch et al., 2005, Jackendoff and Pinker, 2005]. Recursion is discussed more fully in those three follow-
up papers than in the original one.
2. named procedures (or procedural abstraction): Define new (typically, complex)

actions by giving a single name to a (complex) action. (This is even more
optional than exit, but it is very powerful in terms of human readability and
comprehension, and even in terms of machine efficiency; see [Pylyshyn, 1992].)
3. recursion (an elegant replacement for repetition): A recursive instruction tells
you how to compute the output value of a function in terms of previously
computed output values instead of in terms of its input value; of course, the
base case (that is, the output value of the function for its initial input) has to be
given to you in a kind of table-lookup.
We now have our minimal language: Any algorithm for any computable problem
can be expressed in this language (for a Turing machine) that consists of the two
nouns 0 and 1, the two verbs move(location) and print(symbol), and the three
grammatical rules of sequence, selection, and repetition.
Is that really all that is needed? Can your interaction with, say, a spreadsheet
program or Facebook be expressed in this simple (if not simple-minded!) language?
Theres no doubt that a spreadsheet program, for example, that was written in this
language would be very long and very hard to read. But thats not the point. The
question is: Could it be done? And the answer is our final great insight.
7.6.4 The (Church-)Turing Thesis

Our final great insight is the answer to the question at the end of the previous section:
in one word, yes:
Effective computability can be identified with (anything logically
equivalent to) Turing-machine computability.
That is, an algorithm is definable as a program expressible in (anything equivalent
to) our minimal language.
This idea was almost simultaneously put forth both by Church [Church, 1936b]
and by Turing [Turing, 1936]. Some people call it Churchs Thesis; others call it
Turings Thesis; and, as you might expect, some call it the Church-Turing Thesis.
For this reason, Robert Soare [Soare, 2009] has advocated calling it, more simply and
more perspicuously, the Computability Thesis.
But it is only a proposed definition of computable or algorithm: It proposes
to identify any informal, intuitive notion of effective computability or algorithm with
the formal, mathematically precise notion of a Turing machine. How do we know
that Turing-machine computability captures (all of) the intuitive notion(s) of effective
computability?
There are several lines of evidence:
1. There is Turings analysis of computation, which we will examine in detail in
Ch. 8. But, of course, there is also Churchs analysis in terms of the lambda
calculus. Should one of these be preferred over the other? There are two reasons
for preferring Turings over Churchs: First, Turings is easier to understand,
because it follows from his analysis of how humans compute. Secondand
this is merely an appeal to authorityGodel preferred Turings analysis, not

only to Churchs but also to his own! (For discussion, see [Soare, 2009].)
PROBABLY NEED TO SAY A BIT MORE HERE, ESPECIALLY ABOUT
GOEDEL AND WHY HE IS AN AUTHORITY
2. All of the formalisms that have been proposed as precise, mathematical analyses
of computability are not only logically equivalent (that is, any function that
is computable according to one analysis is also computable according to each
of the others), but they are constructively equivalent (that is, they are inter-
compilable, in the sense that you can write a computer program that will translate
(or compile) a program in any of these languages into an equivalent program in
any of the others): GIVE CITATIONS
Turing machines
Post machines (like TMs, but treating the tape as a queue)
Churchs lambda calculus (which John McCarthy later used as the basis of
the Lisp programming language)
Markov algorithms (later used as the basis of the Snobol programming

language)
Post productions (later used as the basis of production systems)

[Post, 1941, Post, 1943], [Soare, 2012, 3293].
Herbrand-Godel recursion equations (later used as the basis of the Algol

family of programming languages)
-recursive functions
register machines [Shepherdson and Sturgis, 1963], etc.
3. Finally, there is empirical evidence: All algorithms so far translate to Turing-

machines; that is, there are no intuitively effective-computable algorithms that
are not Turing-machine computable.
But this has not stopped some philosophers and computer scientists from challenging
the Computability Thesis. Some have advocated forms of computation that exceed
Turing-machine computability. We will explore some of these options in Ch. 11.
What we will now turn our attention to is a somewhat more mathematical outline
of structured programming and recursive functions, after which we will ask whether
there are functions that are non-computable.
7.7 Structured Programming

and Recursive Functions22
7.7.1 Structured Programming
In 7.6.3, I mentioned structured programming, a style of programming that avoids
the use of jump commands, or the go to command, and that Boehm & Jacopini
proved that it could be completely eliminated. Lets see how this can be done:
We can begin with a (recursive) definition of structured program: As with all
recursive definitions, we need to give a base case and a recursive case. The base case
is the notion of a basic program. Using the capital and lower-case Greek letters pi
(, ) to represent programs, we can define the basic programs as follows:
1. Basic programs:
(a) If = begin end., then is a basic (structured) program.

This is called the empty program.
(b) Let F be a primitive operation that is (informally) computablesuch as
an assignment statement (which will have the form y c, where y is a
variable and c is a value that is assigned to y).
Then, if = begin F end., then is a basic (structured) program.

Such programs are called 1-operation programs. Note that it
is not specified which operations are primitive; they will vary
with the programming language. That means that structured
programming is a style of programming, not a particular
programming language. It is a style that can be used with
any programming language. As we will see when we examine
Turings paper in Ch. 8, he spends a lot of time justifying his
choice of primitive operations.
The recursive case for structured programs specifies how to construct more
complex programs from simpler ones. The simplest ones, of course, are the
basic programs: the empty program and the 1-operation programs. So, in
good recursive fashion, we begin by constructing slightly more complex
programs from these. Once we have both the basic programs and the
slightly more complex programs constructed from them, we can combine
themusing the recursive constructs belowto form more complex ones.
2. Program constructors:
Let , 0 be (simple or complex) programs, each of which contains exactly 1

occurrence of end.
22 The material in this section is based on lectures by John Case. WHICH IN TURN ARE BASED ON
CLARK & COWELL; SO COMPARE MY PRESENTATION WITH THEIRS!!

7.7. STRUCTURED PROGRAMMING AND RECURSIVE FUNCTIONS23 245
Let P be a Boolean test. (A Boolean test, such as x > 0, is sometimes

called a propositional function or open sentence. The essential feature of
a Boolean test is that either its output value is true or else it is false.) P must also
be (informally) computable (and, again, Turing spends a lot of time justifying
his choices of tests).
And let y be an integer-valued variable.
Then the following are also (more complex) programs:
(a) = begin ; 0 end. is a (complex) structured program.

Such a is the linear concatenation of followed by 0 .
(b) = begin
if P
then
else 0
end.
is a (complex) structured program.
Such a is a conditional branch: If P is true, then is executed,
but, if P is false, then 0 is executed.
(c) = begin
while y > 0 do
begin
;
y y1
end
end.
Such a is called a count loop (or for-loop, or bounded
loop): The simpler program is repeatedly executed while (i.e.,
as long as) the Boolean test y > 0 is true (i.e., until it becomes
false). Eventually, it will become false, because each time the
loop is executed, y is decremented by 1, so eventually it must
become equal to 0. Thus, an infinite loop is avoided.
(d) = begin
while P do
end.
is a (complex) structured program.
Such a is called a while-loop (or free loop, or unbounded
loop): The simpler program is repeatedly executed while (i.e.,
as long as) the Boolean test P is true (i.e., until P is false). Note
that, unlike the case of a count loop, a while loop can be an infinite
loop, because there is no built-in guarantee that P will eventually
become false (because, in turn, there is no restriction on what P
can be, as long as it is a Boolean test; in particular, if the Boolean

test is the constantly-true test trueor the constantly-true test
1=1then the loop will be guaranteed to be infinite).
We can classify structured programs based on the above recursive definition:
1. is a count-program (or a for-program, or a bounded-loop program) =def
(a) is a basic program, or

(b) is constructed from count-programs by:
linear concatenation, or
conditional branching, or
count looping
(c) Nothing else is a count-program.
2. is a while-program (or a free-loop program, or an unbounded-loop

program) =def
(a) is a basic program, or

(b) is constructed from while-programs by:
linear concatenation, or
conditional branching, or
count-looping, or
while-looping
(c) Nothing else is a while-program.
The inclusion of count-loop programs in the construction-clause for while-

programs is not strictly needed, because all count-loops are while-loops (just
let the P of a while-loop be y > 0 and let the of the while-loop be the linear
concatenation of some other 0 followed by y y 1).
So count-programs are a proper subclass of while-programs: While-programs
include all count-programs plus programs constructed from while-loops that are
not also count-loops.
7.7.2 Recursive Functions

Now lets look at one of the classic analyses of computable functions: a recursive
definition of non-negative integer functions that are intuitively computable
that is, functions whose inputs are non-negative integers, also known as natural
numbers.
First, what is a natural number? The set N of natural numbers = {0, 1, 2, . . .}.
They are the numbers defined (recursively!) by Peanos axioms [Peano, 1889],
[Dedekind, 1890]:
P1 0 is a natural number. (This is the base case.)

P2 Let S be a one-to-one (or injective) function that satisfies the following
two properties:
(a) S takes as input a natural number and returns as output its successor
(b) For any natural number n, S(n) 6= 0
(so, S is not an onto, or surjective, function).
Then, if n is a natural number, then so is S(n).
(This is the recursive case.)
P3 Nothing else is a natural number.
That is, consider a subset M of N. Suppose that:

(a) 0 M
and that
(b) for all n N, if n M, then S(n) M.
Then S = N.
(This is the axiom that underlies the logical rule of inference known as
mathematical induction.)
To define the recursive functions, we once again begin with basic functions
that are intuitively, clearly computable, and then we construct more complex
functions from them.
(a) Basic functions:

i. successor: S(x) = x + 1
ii. predecessor: P(x)= x1 (sometimes pronounced x monus 1)
a b, if a b
(where ab =def )
0, otherwise
j
iii. projection:24 Pk (x1 , . . . , x j , . . . , xk ) = x j
These can be seen to correspond in an intuitive way to the basic

operations of a Turing machine: The successor function corresponds
to move(right), the predecessor function corresponds to move(left)
(where you cannot move any further to the left than the beginning of
the Turing-machine tape), and the projection function corresponds to
reading the current square of the tape.25
(b) Function constructors:
Let g, h, h1 , . . . , hm be (basic or complex) recursive functions.
Then the following are also (complex) recursive functions:
24 Sometimes called identity [Kleene, 1952, 220], [Soare, 2012, 3280].
25 For an analogous comparison, in the context of register machines, see [Shepherdson and Sturgis, 1963,
220].
i. f is defined from g, h1 , . . . , hm by generalized composition =def

f (x1 , . . . , xk ) = g(h1 (x1 , . . . , xk ), . . . , hm (x1 , . . . , xk ))
This can be made a bit easier to read by using the symbol x for the
sequence x1 , . . . , xk . If we do this, then generalized composition can
be written as follows:
f (x) = g(h1 (x), . . . , hm (x))

and this itself can be further simplified to:
f (x) = g(h(x))
Note that g(h(x)), called function composition, is sometimes written

g h. So, roughly, if g and h are recursive functions, then so is their
(generalized) composition g h.
This should be compared to structured programmings notion of linear

concatenation (i.e., sequencing): First compute h; then compute g.
ii. f is defined from g, h, i by conditional definition =def

g(x1 , . . . , xk ), if xi = 0
f (x1 , . . . , xk ) =
h(x1 , . . . , xk ), if xi > 0
Using our simplified notation, we can write this as:

g(x), if xi = 0
f (x) =
h(x), if xi > 0
This should be compared to structured programmings notion of

conditional branch (i.e., selection): If a Boolean test (in this case,
xi = 0) is true, then compute g, else compute h.
iii. f is defined from g, h1 , . . . , hk , i by while-recursion =def

g(x1 , . . . , xk ), if xi = 0
f (x1 , . . . , xk ) =
f (h1 (x1 , . . . , xk ), . . . , hk (x1 , . . . , xk )), if xi > 0
Again, using our simplified notation, this can be written as:

g(x), ifxi = 0
f (x) =
f (h(x), ifxi > 0
And, again, this should be compared to structured programmings

notion of while-loop (i.e., repetition): While a Boolean test (in this
case, xi > 0) is true, compute h, but, when the test becomes false,
then compute g).
An example of a while-recursive function might help to clarify the notion. The

Fibonacci sequence is:
0, 1, 1, 2, 3, 5, 8, 13, . . .
where each term after the first two terms is computed as the sum of the previous
two terms. This can be defined recursively:
First, write down 0, then 1.

Each subsequent term in the sequence is the sum of the previous two terms.
This can be defined using a while-recursive formula as follows:

0, if x = 0
f (x) = 1, if x = 1
f (x 1) + f (x 2), if x > 1

We can make this look a bit more like the official definition of while-recursion
by taking h1 (x) = x1 and h2 (x) = x2. In other words, the two base cases of
f are projection functions, and the recursive case uses the predecessor function
twice (the second time, it is the predecessor of the predecessor).
Given these definitions, we can now classify computable functions:
(a) f is a while-recursive function =def

i. f is a basic function, or
ii. f is defined from while-recursive functions by:
generalized composition, or
conditional definition, or
while-recursion
iii. Nothing else is while-recursive.
This is the essence of the Boehm-Jacopini Theorem: Any computer

program (that is, any algorithm for any computable function) can be
written using only the three rules of grammar: sequence (generalized
composition), selection (conditional definition), and repetition (while-
recursion).
(b) i. f is defined from g, h by primitive recursion =def

g(x1 , . . . , xk ), if y = 0
f (x1 , . . . , xk , y) =
h(x1 , . . . , xk , f (x1 , . . . , xk , y 1)), if y > 0
Using our simplified notation, this becomes:

g(x), if y = 0
f (x, y) =
h(x, f (x, y 1), if y > 0
We can now more clearly see that this combines conditional definition
with a computation of f based on f s value for its previous output.
This is the essence of recursive definitions of functions: Instead of

computing the functions output based on its current input, the output
is computed on the basis of the functions previous output.27
This can be usefully compared to structured programmings notion of

a count-loop: while y > 0, decrement y and then compute f .
ii. f is a primitive-recursive function =def

f is a basic function, or
f is defined from primitive-recursive functions by:
primitive recursion
Nothing else is primitive-recursive.
The primitive-recursive functions and the while-recursive functions
overlap: Both include the basic functions and functions defined
by generalized composition (sequencing). The primitive-recursive
functions also include the functions defined by primitive recursion
(a combination of selection and count-loops), but nothing else. The
while-recursive functions include (along with the basic functions and
generalized composition) functions defined by conditional definition
(selection) and those defined by while-recursion (while-loops).
(c) i. f is defined from h by the -operator [pronounced: mu-operator]
=def
f (x1 , . . . , xk ) = z[h(x1 , . . . , xk , z) = 0]
where:
z[h(x1 , . . . , xk , z) = 0] =def

h(x1 , . . . , xk , z) = 0

min{z : and }, if such z exists

(y < z)[h(x , . . . , x , y) has a value]
1 k

undefined, if no such z exists
This is a complicated notion, but one well worth getting an intuitive

understanding of. It may help to know that it is sometimes called
unbounded search [Soare, 2012, 3284]. Let me first introduce a
useful notation. If f (x) has a value (that is, if it defined, or, in other
words, if an algorithm that computes f halts), then we will write:
f (x) ; and if f (x) is undefined (that is, if it is only a partial function,
or, in other words, if an algorithm for computing f goes into an infinite
loop), then we will write: f (x) .
Now, using our simplfied notation, consider the sequence

h(x, 0), h(x, 1), h(x, 2), . . . , h(x, n), h(x, z), . . . , h(x, z0 )
27 For a useful discussion of this, see [Allen, 2001].
Suppose each of the first n terms of this sequence halts with a non-zero
value, but thereafter each term halts with value 0; that is:
h(x, 0) 6= 0
h(x, 1) 6= 0
h(x, 2) 6= 0
...
h(x, n) 6= 0
but:
h(x, z) = 0
...
h(x, z0 ) = 0
What the -operator does is give us a description of that smallest z
(i.e., the first z in the sequence) for which h halts with value 0. So the
definition of says, roughly:
z[h(x, z) = 0] is the smallest z for which h(x, y) has a non-0 value

for each y < z
but h(x, z) = 0, if such a z exists;
otherwise, if no such z exists, then z is undefined.
So, f is defined from h by the -operator if you can compute f (x)

by computing the smallest z for which h(x, z) = 0.
If h is intuitively computable, then, to compute z, we just have to

compute h(x, y), for each successive natural number y until we find
z. So definition by -operator is also intuitively computable.
ii. f is a partial-recursive function =def

f is a basic function, or
f is defined from partial-recursive functions by:
primitive recursion, or
the -operator
Nothing else is partial-recursive.
iii. f is a recursive function =def

f is partial-recursive, and
f is a total function
(i.e., defined for all elements of its domain)
How are all of these notions related? As we saw, there is an overlap between
the primitive-recursive functions and the while-recursive functions, with the basic
functions and generalized composition in their intersection.
The partial-recursive functions are a superset of the primitive-recursive functions.
The partial-recursive functions consist of the primitive-recursive functions together
with the functions defined with the -operator.
The recursive functions are partial-recursive functions that are also total functions.
Finally, f is primitive-recursive if and only if f is count-program-computable. If
f is primitive-recursive (i.e., if f is count-program-computable), then f is partial-
recursive, which is logically equivalent to being while-program-computable. And
both of these are logically equivalent to being Turing-machine computable, lambda-
definable, Markov-algorithmic, etc.
7.8 The Halting Problem

Have we left anything out? That is, are there any other functions besides these? Yes!
The Halting Problem provides an example of a non-computable function. Recall
that a function is computable if and only if there is an algorithm (that is, a computer
program) that computes it. So, is there a computer program (for example, a program
for a Turing machine) that takes as input:
1. a computer program C, and
2. Cs input (suppose that Cs input is some integer, i)
and that outputs:
yes, if C halts on i
no, otherwise (that is, if C loops on i)?
Call this program H. Because Hs input is C and i, we can refer to H and its input as:
H(C, i)
And our question is: Does H exist? Can we write such a program?
In terms of the function machine illustration, above, we are asking whether there
is a function machine (i.e., a computer) whose internal mechanism (i.e., whose
program) is H. When you input the pair hC, ii to this function machine and turn
its crank, it should output yes if the function machine for C successfully outputs a
value when you give it input i, and it should ouput no if the function machine for C
goes into an infinite loop and never outputs anything.
(Heres another way to think about this: H is a kind of super-machine that takes
as input, not only an integer i, but also another machine C. When you turn Hs crank,
H first feeds i to C, and then H turns Cs crank. If H detects that C has successfully
output a value, then H outputs yes; otherwise, H outputs no.)
This would be very useful for introductory computer-programming teachers (or
software engineers in general)! One thing you never want in a computer program is an
7.8. THE HALTING PROBLEM 253
unintentional infinite loop. (Sometimes, you might want an intentional one: You dont
want an ATM machine, for instance, to haltyou do want it to behave in an infinite
loop so that it is always ready to accept new input from a new customer.) It would
be very useful to have a single, handy program that could quickly check to see if any
program that someone writes has an infinite loop in it. But no such program can be
written!
It is important to be clear that it can be possible to write a program that will check
if another program has an infinite loop. In other words, given a program C1 , there
might be another program H1 that will check whether C1 but not necessarily any
other programhas an infinite loop. What cant be done is this: To write a single
program H that will take any program C whatsoever and tell you whether C will halt
or not.30
Note that we cant answer the question whether C halts on i by just running C on
i: If it halts, we know that it halts. But if it loops, how would we know that it loops?
After all, it might just be taking a long time to halt.
There are two ways that we might try to write H.
1. You can imagine that H(C, i) works as follows:
H gives C its input i, and then runs C on i.

If C halts on i, then H outputs yes;
otherwise, H outputs no.
So, we might write H as follows:
algorithm H1 (C, i):

begin
if C(i)
then output yes
else output no
end.
2. But heres another way to write H:

begin
output no; {i.e., make an initial guess that C loops}
if C(i)
then output yes; {i.e., revise your guess}
end.
30 To be more precise, the difference between these two situations has to do with the order of the logical
quantifiers: The Halting Problem asks this question: Does there exist a single program H, such that, for any
program C, H outputs yes if C halts, else it outputs no? The answer to this question is no; no such
program H exists. The other question, for which there can be a positive answer, is this: Given any program
C, does there exist a program H (which will depend on what C is!) that outputs yes if C halts, else it outputs
no? Note that different Cs might have different Hs. The answer to this question can be yes.
Programs like this will prove useful in our later discussion of hypercomputation
(that is, computation that, allegedly, cannot be modeled by Turing machines). But it
wont work here, because were going to need to convert H to another program called
H , and H2 cant be converted that way, as well see. More importantly, H2 doesnt
really do the required job: It doesnt give us a definitive answer to the question of
whether C halts, because its initial answer is not really no, but something like not
yet.
The answer to our question about whether such an H exists or can be written is
negative: There is no such program as H(C, i). In other words, H(C, i) is a non-
computable function. Note that it is a function: There exists a perfectly good set of
input-output pairs that satisfies the definition of function and that looks like this:
{hC1 (i1 )i, yesi , . . . , , hC j (ik )i, noi , . . .}
Here is a sketch of a proof that H is not computable. The proof takes the
form of a reductio ad absurdum: We will assume that H is computable, and
derive a contradiction. If an assumption implies a contradiction, thenbecause no
contradiction can be truethe assumption must have been wrong. So, our assumption
that H is computable will be shown to be false.
So, assume that H is computable. Now consider another program, H , that works
as follows: H is just like H, except that:
if C halts on i, then H loops
(remember: if C halts on i, H does not loop; it outputs yes)
and if C loops on i, then H outputs no (just like H does).
Here is how we might write H , corresponding to the version of H that we called H1
above:
begin
if C(i)
then
begin
while true do begin end
end
else output no
end.
Here, true is a Boolean test that is always true. (As we noted earlier, you could replace
it by something like 1=1, which is also always true.)
Note that we cannot write a version of H2 that might look like this:
begin
output no; {i.e., make an initial guess that C loops}
if C(i)
7.8. THE HALTING PROBLEM 255
then {i.e., if C halts, then loop}

begin
while true do begin end
end
end.
Why not? Because if C halts, then the only output we will ever see is the message that
says that C loops! That initial, incorrect guess is never revised. So, well stick with H
(that is, with H1 ) and with H (that is, with H1 ).
Note, by the way, that if H exists, so does H . That is, we can turn H into H as
follows: If H were to output yes, then let H go into an infinite loop. That is, replace
Hs output yes by H s infinite loop.
Returning to our proof sketch, the next step is to code C as a number, so that it can
be treated as input to itself.
What? Why do that? Because this is the way to simulate the idea of putting the C
machine into the H machine and then having the H machine turn Cs crank.
So, how do you code a program as a number? This is an insight due to Kurt
Godel. To code any text (including a computer program) as a number in such a way
that you could also decode it, begin by coding each symbol in the text as a unique
number (for example, using the ASCII code). Suppose that these numbers, in order,
are L1 , L2 , L3 , . . . , Ln , where L1 codes the first symbol in the text, L2 codes the second,
. . . , and Ln codes the last symbol.
Then compute the following number:
2L1 3L2 5L3 7L4 . . . pLn n
and where pn is the nth prime number, and where the ith factor in this product is the ith
prime number raised to the Li th power.
By the Fundamental Theorem of Arithmetic,31 the number that is the value of
this product can be uniquely factored, those exponents can be recovered, and then they
can be decoded to yield the original text.32
Now consider H (C,C). (This step is called diagonalization.) That is, suppose
that you (1) code up program C as a Godel number, (2) use it as input to the program C
itself (after all, the Godel number of C is an integer, and thus it is in the domain of the
function that C computes, so it is a legal input for C), and (3) then let H do its job on
that pair of inputs.
By the definition of H :
if program C halts on input C, then H (C,C) loops, and

if program C loops on input C, then H (C,C) outputs no.
31 http://mathworld.wolfram.com/FundamentalTheoremofArithmetic.html (accessed 19 November 2013).
32 Godelnumbering is actually a bit more complicated than this. For more information, see Godel
Number [http://mathworld.wolfram.com/GoedelNumber.html] (accessed 19 November 2013) or Godel
Numbering [http://en.wikipedia.org/wiki/Godel numbering] (accessed 13 November 2013). Turing has an
even simpler way to code symbols; see [Turing, 1936, 5 (Enumeration of Computable Sequences)]. For
a comparison of the two methods, see [Kleene, 1987, 492].
Now code H by a Godel number! And consider H (H , H ). (This is another

instance of diagonalization. It may look like some kind of self-reference, but it really
isnt, because the first occurrence of H names a function, but the second and third
occurrences are just numbers that happen to be the code for that function.)
In other words, (1) code up H by a Godel number, (2) use it as input to the program
H itself, and then let H do its job on that pair of inputs.

Again, by the definition of H :

if program H halts on input H , then H (H , H ) loops, and
if program H loops on input H , then H (H , H ) outputs no.
But outputting no means that it halts!
So, if H halts (outputting no), then it loops, and, if H loops, then it halts. In
other words, it loops if and only if it halts; that is, it does loop if and only if it does not
loop!
But thats a contradiction. So, there is no such H . So, there is no such program as
H. In other words, the Halting Problem is not computable.
7.9 Summary
Lets take stock of where we are. We asked whether computer science is the science
of computing (rather than the science of computers. In order to answer that, we asked
what computing, or computation, is. We have now seen one answer to that question:
Computation is the process of executing an algorithm to determine the output value
of a function, given an input value. We have seen how to make this informal notion
precise, and we have also seen that it is an interesting notion in the sense that not all
functions are computable.
But this was a temporary interruption of our study of the history of computers
and computing. In the next chapter, we will return to the history by examining Alan
Turings formulation of computability.

1. Need a section on heuristic. Be sure to cite its first mention back in
Ch. 3, and be sure to include a discussion of [Simon and Newell, 1958] and
[Newell and Simon, 1976].
Romanycia & Pelletier 1985 is probably the best survey of the varying meanings
of heuristic. Shapiro 1992 suggests that a heuristic is any problem solving
procedure that fails to be an algorithm, or that has not been shown to be an
algorithm. It can fail to be an algorithm because it doesnt terminate, or
because it has not been proved correct, and so on (depending, of course, on
how you define algorithm).
I prefer the following:
A heuristic for problem P can be defined as an algorithm for some
problem P0 , where the solution to P0 is good enough as a solution
to P [Rapaport, 1998, 406].
Being good enough is, of course, a subjective notion, but the idea behind it is
related to Simons notion of bounded rationality: We might not be able to solve
P because of limitations in space, time, or knowledge. But we might be able
to solve P0 algorithmically within the required spatio-temporal-epistemic limits.
And if the algorithmic solution to P0 gets us closer to a solution to P, then it
is a heuristic solution to P. But it is still an algorithm. Oommen & Rueda (in
their abstract, p. 1) call the good enough solution a sub-optimal solution that,
hopefully, is arbitrarily close to the optimal.
2. For 7.6:
Add as verb: read (?)
Add as verb??: P? (i.e., Boolean test)
3. Question for discussion:
Can scientific experiments be considered as (mathematical) functions? Harry
Collins described an experimenters regress:
[Y]ou can say an experiment has truly been replicated only if the
replication gets the same result as the original, a conclusion which
makes replication pointless. Avoiding this, and agreeing that a
replication counts as the same procedure even when it gets a
different result, requires recognising the role of tacit knowledge and
judgment in experiments. [The Economist, 2013]
If you consider the experiment as a mathematical binary relation whose input is,
lets say, the experimental set-up and whose output is the result of the experiment,
then, if a replication of the experiment always gets the same result, then the
relation is a function. In that case, does it make any sense to replicate an
experiment in order to confirm it?
4. Cartoons from website for algorithms:
willy n ethel, for recursion

shoe, for named procedures
hagar, for imprecision
nancy and/or shoe and/or real-life one, for ambiguity
nancy, for infinite loop
zits, for detail
5. A function described extensionally is like a black box; we know the inputs and
outputs, but not what (if anything) goes on inside it. A function described
intensionally (via a formula) is less opaque and gives us more understanding
of the relationship between the input and the outputs. A function described
algorithmically gives us even more understanding, telling us not only what the
relationship is, but giving explicit instructions on how to make the conversion
from input to output. (For more on this notion of understanding in terms of the
internal workings of a black box, see [Strevens, 2013]; this article also suggests
an analogy between computation and causationcf. Chs. 10, 12, 14, 16, and
21.)
6. On algorithmic thinking:
I think it is generally agreed that mathematicians have somewhat

different thought processes from physicists, who have somewhat
different thought processes from chemists, who have somewhat
different thought processes from biologists. Similarly, the respective
mentalities of lawyers, poets, playwrights, historians, linguists,
farmers, and so on, seem to be unique. Each of these groups
can probably recognize that other types of people have a different
approach to knowledge; and it seems likely that a person gravitates
to a particular kind of occupation according to the mode of thought
that he or she grew up with, whenever a choice is possible. C.P. Snow
wrote a famous book about two cultures, scientific vs. humanistic,
but in fact there seem to be many more than two. [Knuth, 1985, 171]
There is a saying that, to a hammer, everything looks like a nail
(http://en.wikipedia.org/wiki/Law of the instrument). This can be taken two
ways: as a warning not to look at things from only one point of view, or as
an observation to the effect that everyone has their own point of view. I take
Knuths remarks along the latter lines. And, of course, his eventual observation
is going to be that computer scientists look at the world algorithmically. Given
the wide range of different points of view that he mentions, one conclusion could
be that, just as students are encouraged to study many of those subjects, so as to
see the world from those points of view, so we should add algorithmic thinking
computer scienceto the mix, because of its unique point of view.
Comparing mathematical thinking to algorithmic thinking, [Knuth, 1985, 181]
finds several areas of overlap and two areas that differentiate the latter from the
former. The areas of overlap include manipulating formulas, representing reality,

problem solving by reduction to simpler problems (which, by the way, is a form
of recursion), abstract reasoning, dealing with information structures, and, of
course, dealing with algorithms (presumably in a narrower sense). The two areas
unique to algorithmic thinking are the notion of complexity or economy of
operation. . . (presumably, what is studied under the heading of computational
complexity [Loui, 1996, Aaronson, 2013]) andof most significancethe
dynamic notion of the state of a process: How did I get here? What is true
now? What should happen next if Im going to get to the end? Changing
states of affairs, or snapshots of a computation, seem to be intimately related
to algorithms and algorithmic thinking.
7. On the difference between algorithm and program:

One way to consider the difference between and algorithm and a program
(written in some programming language) is that the former is an abstract object
and the latter is a concrete implementation of an algorithm. Heres a slightly
different take on this:
A program specifies in the exact syntax of some programming

language the computation one expects a computer to perform. . . . An
algorithm may specify essentially the same computation as a specific
program. . . . Yet, the purpose of an algorithm is to communicate a
computation not to computers but to humans. [Dewdney, 1989, 1]
Dewdney seems to distinguish three things: a computation, an algorithm (for

that computation), and a program (for that computation). What he calls a
computation seems to be more like what I would call an algorithm. (There
might also be a fourth thing: a computation being performed, or what is
sometimes called a process.) His notion of algorithm seems to be more
concrete than mine, because it has to be communicated to a human, hence needs
to be an expression in some humanly understandable language.
Further evidence that his notion differs from mine are the following remarks:
A recipe, for example, is an algorithm for preparing food (assuming,

for the moment, that we think of cooking as a form of computation).
[Dewdney, 1989, 1]
Before continuing, heres something to think about that will help prepare you
for the discussion in Ch. 10: Is it reasonable to think of cooking as a form
of computation? Unlike other forms of computation that involve numbers or,
more generally, symbols,33 cooking involves physical objects, in particular,
food. Arithmetic and word processing are typical kinds of things we do with
computers. (So are drawing pictures and making music.) Do we expect
computers to be able to prepare dinner for us?
But lets continue with the evidence I mentioned:
33 A related question: Do computers deal with numbers? Or with numerals (that is, symbols for numbers)?
A major feature shared by. . . [recipes] with all algorithms is the

looseness of its description. . . . Such things do not bother experienced
cooks. Their judgment and common sense fill the obvious gaps in the
recipe. [Dewdney, 1989, 3]
But algorithms, as standardly conceived, are never loose! They never require
judgment on the part of the agent that executes them.
Another consideration raised by the previous quotation is this: If cooking is a
form of computation, what plays the role of the computer? Dewdney makes it
sound like the cook is the computer.
8. On the intenTionality of an algorithm:

Somewhere, need to discuss this problem, possibly in Ch. 17 if not here: Must
algorithms be intentional in the sense that they are for some purpose, designed
to solve some particular problem? Is an algorithm that is designed to allow
a robot to solve a blocks-world problem but that has no feedback to enable it
to determine if it has dropped a block (and has therefore failed to accomplish its
task) therefore not an algorithm? Is a recipe (lets assume for the moment that its
an algorithm) for making Hollandaise sauce but that is executed on the moon and
therefore fails for lack of oxygen (or whatever) not therefore an algorithm? Is an
algorithm for computing GCD that is given base-2 numerals as input instead of
base-10 numerals therefore not an algorithm? [Rescorla, 2013]. Is an algorithm
for playing a particular chess game but that also simulates a particular Civil War
battle one algorithm or two? [Fodor, 1978, 232]. Is an algorithm for analyzing
x-ray diffraction data that. . . also solves Sudoku puzzles [Elser, 2012, 420]
one algorithm or two? Is this any different from the situation in (applied)
mathematics, where a mathematical structure designed for one purpose is found
useful elsewhere? (E.g., the application of group theory in physics.) Perhaps all
that means is that the two application situations share a common (mathematical
strucure). So, perhaps, it is a common, i.e., abstract, algorithmic structure that
is the real algorithm, and not some specific interpretation or use of it (which
might require a different input-output encoding or interpretation). This is also
related to the more general question of the nature and role of (mathematical
or other scientific) models. (For other ideas along these linesput in Further
Sources of Infosee:
Hayes, Brian (2004), Small-Town Story, American Scientist (March-
April): 115119; https://www.americanscientist.org/issues/pub/small-town-
story (accessed 14 March 2014)
For some ideas on the intentionality of algorithms, see Hayess comments

on human vs. ant behavior models.
Re: [Rescorla, 2013]:
[Stewart, 2000, 98] suggests that 3 is not an integer power of 2 is true
in base-10 arithmetic but false in the set of integers modulo 5 . More
importantly, he says that This doesnt mean that [the first statement] is
wrong, because the context has been changed. It just means that I have to
be careful to define what Im talking about. Similarly, we must distinguish

between these two statements:
(a) The number represented by the numeral 3 is an integer power of the
number represented by the numeral 2.
(b) The number canonically represented as {{{}}} is an integer power
of the number canonically represented as {{}}.
The first of these depends on the base that the symbols are in; the second
doesnt. Or are neither about symbols, and both are about numbers?
9. In [Soare, 2012, 3289]s characterization, the output of a Turing machine is the

total number of 1s on the tape. So, the key to determining what is computable
(that is, what kinds of tasks are computable) is finding a coding scheme that
allows a sequence of 1sthat is, an integerto be interpreted as a symbol, a
pixel, a sound, etc.
10. Not sure if this goes here, elsewhere, or nowhere, because it has more to do
with causation than computation, but it may have some relationship to ethical
issues in robotics, and it is certainly related to issues in (math) education: If I
use a calculator (or a computer, or if a robot performs some action), who or what
is really doing the calcuation (or the computation, or the action)Is it the
calculator (computer, robot)? Or me? Compare this real-life story: I was making
waffles from scratch on a waffle iron. The 7-year-old son of friends who were
visiting was watching me and said, Actually, youre not making it; its the thing
[what he was trying to say was that it was the waffle iron that was making the
waffles]. But you set it up, so youre the cook.
11. The Halting Problem is not the only non-computable function. There are many
more; in fact, there are an infinite number of them. Two other famous ones are
Hilberts 10th Problem and the Busy Beaver function.
The first of these was the 10th problem in the famous list of math problems that
Hilbert presented in his 1900 speech as goals for 20th century mathematicians
to solve [Hilbert, 1900]. It concerns Diophantine equations, that is, equations
of the form p(x1 , . . . , xn ) = 0, where p is a polynomial with integer coefficients.
Hilberts 10th Problem says:
Given a diophantine equation with any number of unknown quantities
and with rational integral numerical coefficients: To devise a process
according to which it can be determined by a finite number of
operations whether the equation is solvable in rational integers.
[Hilbert, 1900, 458]
In the early 1970s, Yuri Matiyasevich, along with the mathematicians Martin
Davis and Julia Robinson, and the philosopher Hilary Putnam, proved that there
was no such process (that is, no such algorithm). (For more information, see:
http://en.wikipedia.org/wiki/Hilberts tenth problem
and Hilberts Tenth Problem page at:
http://logic.pdmi.ras.ru/Hilbert10/ (both accessed 16 December 2013).)
The Busy Beaver function has been described as follows:

(n) is defined to be the largest number which can be computed by
an n-state Turing machine. . . . [Chaitin, 1987, 2]
GIVE DEWDNEY DESCRIPTION, TOO. It was first described, and proved to
be non-computable, by Tibor Rado [Rado, 1962].34 (For more information, see
[Suber, 1997b] and
http://en.wikipedia.org/wiki/Busy beaver
(accessed 16 December 2013).)
12. On inputs and outputs (relevant to discussion above about whether the informal
notion of algorithm requires these; also relevant to discussions in later chapters
on relation of program to world, Clelands views, etc.):
Do computations have to have inputs and outputs? The
mathematical resources of computability theory can be used to
define computations that lack inputs, outputs, or both. But
the computations that are generally relevant for applications are
computations with both inputs and outputs. [Piccinini, 2011, 741,
note 11]
See comment in Ch. 17 at NOTES6.
13. While the official Turing thesis is that every computable function
is Turing computable, Turings informal argument in favor of his
thesis justifies a stronger thesis: every algorithm can be simulated
by a Turing machine. [Gurevich, 2000, 77]
But note that Gurevich says simulated: The actual algorithm running on the
TM will typically not be the same as the algorithm being simulated, because:
TMs deal with single bits. Typically there is a huge gap between
the abstraction level of the program and that of the TMs; even if you
succeed in the tedious task of translating the program to the language
of TMs, you will not see the forest for the trees. Can one generalize
Turing machines so that any algorithm, never mind how abstract, can
be modeled by a generalized machine very closely and faithfully?
The obvious answer seems to be no. The world of algorithms is
too diverse even if one restricts attention to sequential algorithms. . . .
[Gurevich, 2000, 78]
Therefore, there is more to an algorithm than the function it computes
[Gurevich, 2000, 80].
Given a computable function f , there will be an algorithm that computes
it. A Turing machine for f will be an example of such an algorithm (or, if you
34 This paper, by the way, is a wonderfully readable introduction to the Busy Beaver game. II gives a
very simple example of a Turing machine aimed at students, and IIII and, especially, VIII are amusing
and well worth reading!
prefer, a universal Turing machine with a program for computing f ). But there
will be many algorithms that compute f , and not all of them will be expressible in
the language of Turing machines. So, theres a difference between an algorithm
and a program (which can be thought of as an algorithm expressed in some
programming language).
An algorithm is something abstract that can be implemented as different
programs in different programming languages. Any program that implements
a particular algorithm at best simulates any other program that implements that
algorithm.
Moreover, Turing-machine algorithms (according to [Gurevich, 2011,
6]) are symbolic sequential algorithms. What about ruler-and-compass
algorithms? [Toussaint, 1993]: [T]he functions computed by ruler-and-
compass algorithms are not Turing computable [Gurevich, 2011, 3]. Why
not? One possible way to reduce the latter to the former might be by way
of an output plotter that can simulate ruler-and-compass constructions. Or,
more simply (in one sense of simply!), if all such (geometric) constructions
can be expressed in terms of analytic geometry, then they would seem to be
able to be expressed in terms of functions (over real numbers), and hence be
(Turing-)computable. But lets suppose that Gurevich is right: Surely, theres
something algorithmic about ruler-and-compass constructions. They would
seem to satisfy the informal definition of algorithm given in 7.5, differing
only in what counts as a statement, rule, or instruction. And, given the
Boehm-Jacopini Theorem and the third insight of 7.6.3, the core difference
is what counts as a primitive or basic statement, rule, or instruction.
For Turing machines, those are print and move; for recursive functions,
they are successor, predecessor, and projection; for ruler-and-compass
constructions, they would be something like drawing an arc with a certain radius,
drawing a line of a certain length, etc. Under this view, are all algorithms
expressible in the language of Turing machines? (see also [Goodman, 1987, 4])
Gurevich distinguishes pretty clearly algorithms from Turing
machines, pointing out that, although for any (intuitively) computable function
there is an algorithmin particular, a TMthat computes it, it doesnt follow
that a TM can compute every algorithm; at best, TMs can simulate any
algorithm. One example is Euclids GCD algorithm, which comes in both a
subtraction and a division form (division just being repeated subtraction). A
TM probably cant implement the latter (in any direct way), because its only
way of (indirectly) implementing division would be by implementing repeated
subtraction.
He also says that ruler-and-compass constructions are not Turing
computable, nor is Gaussian elimination. He gives no justification for either
statement. On the face of it, the former might be TM-computable if you hook a
plotter, or even a ruler and a compass, up to it and let the TM control them. Even
easier, translate every ruler-and-compass algorithm into the language of analytic
geometry, which, in turn, is expressible in terms of (computable) functions (on
the reals). Seems to me that that makes ruler-and-compass constructions TM-

computable. (Although, consistent with the remarks in the previous paragraph,
the actual ruler-and-compass algorithms are only being simulated.) As for
Gaussian elimination (an algorithm for dealing with matrices), I cant find any
source other than Gurevich who says that they arent (TM-)computable.
But ruler and compass was their computer. It was their true computer.
They really computed that way. [Gurevich, 2013]
([Feferman, 1992, 318] makes a similar remark.)
Heres another statement of the issue:
There may be a huge difference between an algorithm and the
simulating Turing machine. The poor Turing machine may work
long and hard, with its head creeping painfully slowly back and
forth on that long narrow tape, in order to simulate just one step
of the given algorithm. . . . The problem arises whether there is a
machine model such that any algorithm, however abstract, can be
simulated step-for-step by a machine of that model. Let me call such
a model universal with respect to algorithms. Turings machine is
universal with respect to computable functions, but not with respect
to algorithms. [Gurevich, 1999, 9596, my italics]
14. In a section titled The flaw in Churchs thesis, Robert I. Soare says:
If the basic steps are stepwise recursive, then it follows easily by the
Kleene normal form theorem. . . that the entire process is recursive.
The fatal weakness in Churchs argument was the core assumption
that the atomic steps were stepwise recursive, something he did not
justify. [Soare, 2012, 3286, my italics]
Churchs theory, the -calculus, had as its basic, or atomic, steps formal
operations on function formulas that some peopleGodel in particulardid not
find to be intuitively computable.35 The same could be said even for Godels own
theory of recursive functions. But Turings basic operations were, by design,
simple things that any human could easily do: put a mark at specific location on
a piece of paper, and shift attention to a different location.
15. For statements of equivalence of general recursive, -recursive, -definable, etc.,
see [Soare, 2012, 3284].
16. One of the essential features of an algorithm is that it is an explicit set
of instructions (for doing something). (Exactly what counts as being an
instruction and as being explicit is whats at the heart of the formal notion
of algorithm.) Consider this observation:
35 For an elemetary introduction to the lambda-calculus, see [PolR, 2009, Alonzo Churchs Lambda-
Calculus].
Computer-driven machines can now carry out a huge range of

very highly skilled tasks, from navigating and landing aircraft,
to manufacturing and assembling a wide range of products.
[MacFarlane, 2013, 18]
What makes this possible is that we have figured out how to express the detailed
instructions for how to do these thingswe have figured out algorithms for them;
we not only know how to do them, we know how to tell someone or something
how to do them.
And one way of characterizing computer science is as the discipline that is
concerned with determining which tasks are such that we can express how to
accomplish them in the language of an algorithm; that is, computer science is
concerned with determining which tasks are computable.
17. On the difference between functions given by formulas and functions given by
algorithms:
[Pearl, 2000] considers the difference between two representations of simple
algebraic relations, which he terms equations versus diagrams, contrasting:
Y = 2X
Z =Y +1
with
X 2[Y ] +1[Z]
The former describes relations between variables; the latter specifies a simple
computer program, in the form of a flowchart, indicating the order in which
operations are carried out [Chater and Oaksford, 2013, 1172].36 (The former
could even be simplified to Z = 2X + 1.) The flowchart, or algorithmic, version
allows for interventionschanges that can be inserted in between steps of the
algorithm; the equation, or formula, version does not allow for that, being merely
a static description of the output of the function.
18. In general[,] algorithms perform tasks, and computing functions is a
rather special class of tasks. [Gurevich, 2012, 2]
Unless every task can be expressed as (that is, coded as) a (mathematical)
function. This is also relevant to Clelands concerns; see Ch. 10. Even Gurevich
himself alludes to this, when he says:
The validity relation on first-order formulas can be naturally
represented as a real number, and the Entscheidungsproblem becomes
36 The interpretation of the flowchart version is something like this:
(a) input X
(b) multiply X by 2; store in Y
(c) add 1 to Y ; store in Z
Note that this algorithm does not appear to have an output; see 7.5.
whether this particular real number is computable. [Gurevich, 2012,

2]
But, by our first great insight (7.6.1), anything that we want to process
computationally can be encoded as a real number. So, perhaps computing
functions is not so special after all.
19. To my list of features of algorithms in 7.5, above, [Gurevich, 2012, 4] adds

isolation:
Computation is self-contained. No oracle is consulted, and nobody

interferes with the computation either during a computation step or in
between steps. The whole computation of the algorithm is determined
by the initial state.
Does anyone else cited above mention this? How does (the lack of) this relate to
interactive computing/hypercomputation? (See Ch. 11.)
Does the use of such an oracle mean that the computation is no longer purely
automatic or mechanical? (But why not?)
Also, while Im thinking about it, what about Gurevichs first constraint:
Computation is symbolic (or digital, symbol-pushing) (p. 4). That is, the
syntactic nature of computation. Is that part of my characterization above? (For
more on this, see papers by Rescorla.)
20. [C]onflating algorithms with Turing machines is a misreading of

Turings 1936 paper. . . . Turings aim was to define computability, not
algorithms. His paper argued that every function on natural numbers
that can be computed by a human computer. . . can also be computed
by a Turing machine. There is no claim in the paper that Turing
machines offer a general model for algorithms. [Vardi, 2012]
Vardi goes on to cite Gurevichs idea that algorithms are abstract state
machines, whose key requirement is that one step of the machine can cause
only a bounded local change on. . . [a] state (which is an arbitrary data
structure). Note that this corresponds to Turings analysis and to our analysis in
7.5.
He also cites Moschovakiss idea that algorithms are recursors: recursive
description[s] built on top of arbitrary operations taken as primitives. Note that
this corresponds to recursive functions and the lambda-calculus, as we discussed
in 7.7.
And he then observes, in line with the proofs that Turing machines, the lambda-
calculus, recursive functions, etc., are all logically equivalent, that these distinct
notions are analogous to the wave-particle duality in quantum mechanics: An
algorithm is both an abstract state machine and a recursor, and neither view by
itself fully describes what an algorithm is. This algorithmic duality seems to be
a fundamental principle of computer science.
21. Another informal notion of algorithm:
(a) [A] method for answering any one of a given infinite class of
questions. . . is given by a set of rules or instructions, describing
a procedure that works as follows. (b) After the procedure has
been described, [then] if we select any question from the class, the
procedure will then tell us how to perform successive steps, so that
after a finite number of them we will have the answer to the question
selected. (c) In particular, immediately after selecting the question
from the class, the rules or instructions will tell us what step to
perform first, unless the answer to the question selected is immediate.
(d) After our performing any step to which the procedure has led us,
the rules or instructions will either enable us to recognize that now
we have the answer before us and to read it off, or else that we do
not yet have the answer before us, in which case they will tell us what
step to perform next. (e) In performing the steps, we simply follow
the instructions like robots; no ingenuity or mathematical invention is
required of us. [Kleene, 1995, 18]
So, an algorithm (informally) is:

a A set of rules or instructions that describes a procedure. The procedure is one
thing; its description is another; the latter is a set of imperative sentences.
b Given a class of questions CQ and a procedure PCQ for answering any
member: q CQ[PCQ gives a finite sequence of steps (described by its rules)
that answers q]
So, the finiteness occurs in the execution of PCQ (not necessarily in PCQ
itself). And PCQ does not depend on q, only on CQ.
That does suggest that an algorithm is intentional in the sense of having a
goal or purpose.
c The algorithm inputs question q, and either outputs qs answer (base case), or
else outputs the first step to answer q (recursive case).
d If it is the recursive case, then, presumably, q has been reduced to a simpler
question and the process repeats until the answer is produced as a base case.
Moreover, the answer is immediately recognizable. That does not necessarily
require an intelligent mind to recognize it. It could be merely that the
algorithm halts with a message that the output is, indeed, the answer. (The
difference between this and a trial-and-error machine is that here the output is
of the form: The answer to q is qi , where qi is the answer and the algorithm
halts, or qi is a one-step-simpler question and the algorithm tells us what the
next step is. In a trial-and-error machine, the output is of the form: My current
guess is that the answer to q is qi and the algorithm tells us what the next step
is.
e The algorithm is complete in the sense that it contains all information for
executing the steps. We, the executor, do not (have to) supply anything else. In
particular, we do not (have to) accept any further, unknown/unforeseen input

from any other source (that is, no oracle or interaction)
22. On the relation of algorithm, decidability, and computability:

Questions come in two kinds: true-false (yes-no) questions and what
questions. For example, Is 2+2=4? is a true-false question, and What is 2+2?
is a what question. A relation R(a1 , . . . , an )that is, a Boolean proposition
that is either true or else false (a yes-no question)is decidable means that
there is an algorithm (a decision procedure) that decides its truth value (that
answers the question). A function f (a1 , . . . , an )that is, a what questionis
computable means that there is an algorithm (a computation procedure) that
computes its output value (that answers the question). [Kleene, 1995, 17]
23. [Kleene, 1995] shows how to compile (or translate the language of) recursive
functions into (the language of) Turing machines, that is, how a Turing machine
can compute recursive functions.
24. Nice terminology (due to Bill Duncan? Or did he get it from me?):
Functions relate inputs to outputs;
Algorithms (describe how to) transform inputs into outputs.
25. [A function is a set of ordered pairs. . . [satisfying the unique output

condition]. . . . A computable function. . . is a mapping [that is, a
function] that can be specifed in terms of some rule or other,
and is generally characterized in terms of what you have to do
to the first element to get the second [where the rule must satisfy
the constraints of being an algorithm]. . . . [A] noncomputable
function. . . is an infinite set of ordered pairs for which no rule
can be provided, not only now, but in principle. Hence its
specification consists simply and exactly in the list of ordered
pairs. [Churchland and Sejnowski, 1992, 62, their italics, my
boldface]
Note that this is almost the same as Chaitins definition of a random number.
26. On input and output:
There remains, however, the necessity of getting the original

definitory information from outside into the device, and also of
getting the final information, the results, from the device to the
outside. [von Neumann, 1945, 2.6, p. 3].
In other words, there has to be input and output. The computer has to have
something to work on (definitory information, or input), and it has to let the
human user know what it has computed (the final information, the results, or
output). It cant just sit there silently computing.
27. A function from a set A to a set B is a regularity between the two sets (I borrow
the term from [Shagrir, 2006b, 402]. It is not random, because the output of a
given input must always be the same; a random relation between A and B would
not guarantee this. And a non-random, but non-functional, relation would give
different (if not random) outputs for some inputs (as in the square-root relation,
which, for a given positive integer input gives both positive and negative square
roots as output).
28. David Marr CITE analyzed information processing into three levels:
computational (what a system does), algorithmic (how it does it), and physical
(how it is implemented). But a different terminology, more consistent with what
we are using in this book, would identify four levels: functional (a description
of a system in input-output terms; this would seem to align with Marrs
computational level), formula (or algebraic) (a description using an (algebraic)
formula that relates the inputs and outputs), algorithmic (a description of how to
compute the formula; this might also be called the computational level, but
it seems to align more with Marrs algorithmic level), and implementional (a
description of a specific physical device).
29. The notion that algorithms and Turing machines are not exactly the same thing
(as Gurevich argues) is also explored in [Ross, 1974]:
Does it [Turings analysis] tell us what an algorithm is? [Ross, 1974,

517]
He points out that this depends on what we would mean by equivalent when
we ask whether a given program (expressing an algorithm) is equivalent to a
Turing-machine program. Does this mean mere input-output equivalence? Or
something stronger (some kind of procedural equivalence)?
And, relevant to hypercomputation, he observes that:
Ultimately, the truth or falsity fo Churchs thesis seems to rest heavily

upon what powers are attributable to the idealized agents who are
to perform the computations specified by algorithms. For example,
if we were to consider God as calculator it might not be improper
to consider as algorithmic instructions involving the performance of
infinitely many operations. [Ross, 1974, 519]
And, relevant to Turings strange inversion, he observes:
Almost all intuitively written algorithms require the agent to know

what he is doing. That is, most intuitive algorithms require the agent
to understand the meanings of the symbols he is manipulating and
usually a great deal more as well. . . . Doubt about whether powers of
comprehension used by human agents give no computational powers
beyond those of a pigeon leads to skepticism concerning Churchs
thesis. [Ross, 1974, 520]
On pigeons, Ross says, in a footnote, that

It has been shown by P. Suppes that a pigeon can be taught to perform

all the tasks necessary to follow the program of a universal Turing
machine, hence any Turing program. [Ross, 1974, 520n1].
Shades of Searles Chinese Room Argument!
30. On whether an algorithm requires output:
Read is the companion process to write, each being necessary

to make the other useful. Read only obtains what was put into
expressions by write at an earlier time; and a write operation whose
result is never read subsequently might as well not have happened.
The last clause suggests that output is not necessary.
31. Nice quotes relevant to Turings strange inversion:
One may be a mathematician of the first rank without being able to
compute. It is possible to be a great computer without having the
slightest idea of mathematics.
Novalis (Georg Philipp Friedrich Freiherr von Hardenberg), cited
in [Ralston, 1999, 173]
Every cell in my body knows how to replicate DNA yet Im not in on

it so I have to spend hours studying it.
anonymous meme found on the Web, 2015.
32. Leibniz isolated some general features of algorithmic procedures. . . .

. . . an algorithmic procedure must determine completely what actions
have to be undertaken by the computing agent. . . . the instructions
of a calculation procedure can be viewed as prescribing operations
on symbolic expressions in general, and not just on numerical
expressions. . . . only physical properties of symbolssuch as their
shape and arrangementand not, e.g., their meaning, play a role in
a calculation process. . . . only elementary intellectual capabilities are
required on the part of the executor of a calculation procedure. . . .
[Spruit and Tamburrini, 1991, 78].
33. Computability is a relative notion, not an absolute one. All
computation, classical or otherwise, takes place relative to some set or
other or primitive capabilities. The primitives specified by Turing in
1936 occupy no privileged position. One may ask whether a function
is computable relative to these primitives or to some superset of
them. [Copeland, 1997, 707]; see also [Copeland and Sylvan, 1999,
4647].
Compare the situation with Euclidean geometry: An angle cannot be trisected
using only compass and straightedge; but, of course, using a protractor to
measure its size and then calculating one-third of that will do the trick.
34. [Copeland, 2000] identifies effective with mechanical, and defines a method,
or procedure, M, for achieving some desired result to be effective iff:
1 M is set out in terms of a finite number of exact instructions (each instruction

being expressed by means of a finite number of symbols);
2 M will, if carried out without error, always produce the desired result in a finite
number of steps;
3 M can (in practice or in principle) be carried out by a human being unaided by
any machinery save paper and pencil;
4 M demands no insight or ingenuity on the part of the human being carrying it
out.
35. Following email discussions with Peter Denning, Im inclined to add a fifth
insight to my previous list: Turing machines can be physically implemented.
(See also Ch. 14, Notes for Next Draft No. 8.)
This brings in one engineering aspect of computer science. In fact, there are
two engineering aspects. One might be called the software-engineering aspect:
Although only two nouns, two transitive verbs, and three rules of grammar are
needed, other nouns (data structures), other verbs (named subroutines), and other
rules of grammar (other control structures such as exits from loops, controlled
gotos, etc.) can be added to allow for greater expressivity and control of
complexity.
The other aspect of engineering is the physical kind (the fifth insight). But it
brings in its train limitations imposed by physical reality: limitations of space,
time, memory, etc. Issues concerning what is feasibly or efficiently computable
in practice (as opposed to what is theoretically computable in principle)
complexity theory, worst-case analyses, etc.and issues concerning the use of
heuristics come in here.
36. On formulas vs. algorithms:

Some functions expressed as formulas might be seen as containing an implicit
algorithm for how to compute them:
[A] term in the series for arctan 1/5 can be written either as (1/5)m /m
or as 1/(m5m ). Mathematically these expressions are identical, but
they imply different computations. In the first case you multiply and
divide long decimal fractions; in the second you build a large integer
and then take its reciprocal. [Hayes, 2014a, 344]
But these formulas can only be interpreted as algorithms with additional

information about the order of operations (roughly, do things in innermost
parentheses first, then do exponentiations, then multiplication and division (from
left to right), then addition and substraction (from left to right)).
37. On the intentionality of algorithms (again):

First, recall Robin Hills point about this: Algorithms need to be expressed in the
form To do A, do the following. . . .
Second, consider these observations:
The syntactic behaviour of a Turing machine is fully described by

its machine table. But we need also an explanation of why the
Turing machine computes the arithmetic function that it does. We
have an explanation of that only when we attribute various semantic
properties to collections of the quadruples. [Peacocke, 1999, 197, my
italics]
This is related to my experience of being told how to use a spreadsheet by

following meaningless instructions that turned out to be instructions for
adding. On this, see Dennetts Turings Strange Inversion.
Why do we need an explanation of why the Turing machine is computing the

function that does? Is that different from merely asking what it is doing? I think
it does go beyond that. Given a Turing machine that does something, we can ask
what it is doing. We can answer that by describing its machine table (a purely
syntactic answer). We can also answer by saying that it is adding two numbers.
(Compare: When I add two numbers using a calculator, we can describe what I
am doing either by saying that I am pressing certain buttons on the calculator in
a certain sequence or by saying that I am adding two numbers.) But it is a further
question to ask why printing and moving in that way (or pushing those buttons)
results in adding two numbers.
Peacockes emphasis on the semantic properties (the semantic interpretation of

what the Turing machine (or we) are doing) as being attributed by us is certainly
in line with Searles view of things.
But, in line with Fodors chess-Civil War example, the single thing that the
Turing machine is doing might be interpreted in two different (albeit isomorphic)
ways. Is one thing being done, or two?
One way in which the external semantic interpretation can be seen as being given
is in terms of identifying certain groups of operations as constituting subroutines
that can be named. Identifying certain groups as accomplishing a subtask seems
to me to be syntactic. Giving such a group a name borders on being semantic.
But not if the name is something like operation 7Bonly if the name is
something like adding two numbers and then only if that name is associated
with other information (see my arguments about the ability of a computer to
understand what it is doing, in, for example, [Rapaport, 1988]).
Egan says that a description of a device as computing a

mathematical function is an environment-independent description.

[Peacocke, 1999, 198].
Egans statement is certainly in line with objections to Clelands views about the
Church-Turing Computability Thesis (see 10.5.1). But Peacocke objects:
The normal interpretation of a Turing machine assigns the number 0
to a single stroke |, the number 1 to ||, the number 2 to |||, and
so on. But there will equally be an interpretation which assigns 0 to a
single stroke |, and then assigns the number 2 to ||, the number 4
to |||, and generally assigns 2n to any symbol to which the previous
interpretation assigns n. Under the second interpretation, the Turing
machine will still be computing a function. . . . What numerical value
is computed, and equally which function is computed, by a given
machine, is not determined by the purely formal characterization of
the machine. There is no such thing as purely formal determination
of a mathematical function. . . . [W]e can say that a Turing machine
is really computing one function rather than another only when it is
suitably embedded in a wider system. [Peacocke, 1999, 198199].
The description of this single Turing machine as either computing n or 2n is
like the chess-Civil War situation. Again, we can say that the function that
this Turing machine is computing is a function whose outputs are strokes, or
a function whose outputs are n, or a function whose outputs are 2n. There
is no single, correct answer to the question What is this Turing machine
computing?. We can fix an answer by embedding the Turing machine in a
wider system; this is what I advocate in my argument that a computer can
understand what it is computing.
It is also reminiscent of a point made in [Frenkel, 2013]: He considers a simple

equation, like y2 + y = x3 + x2 , and asks
But what kind of numbers do we want x and y to be? There are several
choices: one possibility is to say that x and y are natural numbers or
intergers. Anotehr possibility is to say rational numbers. We can
also look for solutions x, y that are real numbers, or even complex
numbers. . . (p. 83).
He continues:
[W]hen we talk about solutions of such an equation, it is important
to specify to what numerical system they belong. There are many
choices, and different choices give rise to different mathematical
theories. (p. 99)
Heres a simpler example: What are the solutions to the equation x2 = 2? In
the rational numbers, there is no solution; in the positive real numbers, there is
one solution; in the (positive and negative) real numbers, there are two solutions.
Similarly, x2 = 1 has no solution in the real numbers, but two solutions in the
complex numbers. Deciding which wider number system the equation should
be embedded in gives different interpretations of it.
But does this mean that content or interpretation is necessarily part of

computation? Surely, a Turing machine that outputs sequences of strokes does
just that, and whether those strokes should be
interpreted as n or 2n is a separate
matter. Similarly, the solution to x2 = c is 2 c; whether we assign a rational,
real, or complex number to that solution is a separate matter.
[Sprevak, 2010, 260, col. 1, my italics] discusses the received view. . . that
. . . a necessary condition on any physical process counting as a computation is
that it possess representational content. A physical process is what it is, so to
speak, independent of anything that it might mean, be about, or represent.
But, according to Sprevaks understanding of what he calls the received view,
whatever it is by itself, so to speak (syntactically?), it isnt a computation unless
it is about something. Hence Hills insistence on the intentionality: Do X might
be a mere physical process, but To accomplish A, do X is a computation.
One reason in support of this received view that Sprevak offers, and against
Egans interpretation of Marr, is that mathematical computation theory does not,
by itself, have the resources to explain how the visual system works. . . . Marr
needs some way of connecting the abtract mathematical descriptions to the nuts
and bolts of physical reality [Sprevak, 2010, 263, col. 1]. But the mathematical
theory does have the resources; thats the point of [Wigner, 1960]. (It may still
be a puzzle how it does, but theres no question that it does.)
[Sprevak, 2010, 3.2, p. 268] goes on to argue that

one cannot make sense of I/O equivalence without requiring that
computation involves representational content.
Consider two computational systems that perform the same
numerical calculation. Suppose that one system takes ink-marks
shaped like Roman numerals (I, II, III, IV, . . . ) as input and yields ink-
marks shaped like Roman numerals as output. Suppose that the other
system takes ink-marks shaped like Arabic numerals (1, 2, 3, 4, . . . )
as input and yields ink-marks shaped like Arabic numerals as output.
Suppose that the two systems compute the same function, say, the
addition function. What could their I/O computational equivalence
consist in? Again, there may be no physical or functional identity
between their respective inputs and outputs. The only way in which
the their inputs and outputs are relevantly similar seems to be that
they represent the same thing.
But there is a difference between what a system is doing and whether two systems
are doing the same thing: The two addition algorithms (Roman and Arabic)
arent doing the (same) identical thing, but they are doing the same (equivalent)
thing. It is the sameness that depends on the semantics.
He goes on to say:
Two physical processes that are intrinsic physical duplicates may

have different representational contents associated with them, and
hence different computational identities. One physical process may
calculate chess moves, while a physical duplicate of that process
calculates stock market predictions. We seem inclined to say that, in
a sense, the two processes compute different functions, yet in another
sense they are I/O equivalent. Appeal to representational content can
accommodate both judgements. [Sprevak, 2010, 268, col. 2]
But this is a different case: identical algorithm but different task. The other case
is different algorithm but same (input-outputequivalent) task.
38. . . . [R]ealism about computation. . . is the view that whether or not a

particular physical system is performing or implementing a particular
computation is at least sometimes a fact that obtains independently
of human beliefs, desires and intentions. Unless we have a precise
account of implementation it will not be possible to decide whether
or not realism about computation is correct just because it will not be
clear what computation means. [Ladyman, 2009, 377]
First, this is relevant to the Fodorian chess-Civil War example: Is it independent
of human beliefs, desires, and intentions that a particular computer is playing
chess rather than rehearsing a Civil War battle? It certainly seems to be the case
that the internal computations are thus independent even if the input-output is
not.
Second, the reason that the meaning of computation is relevant has to do with
this observation:
The idea of computation bridges the abstract and the concrete worlds.
For example, suppose somebody claims: The computation took ten
minutes. He or she is clearly referring to a concrete property of a
particular process in space and time. On the other hand, suppose
somebody claims: The computation is logically irreversible. He
or she is clearly referring to a general abstract notion of the
computation and an abstract property of it, namely that knowledge
of the computation and its output is not sufficient for knowledge of its
input. [Ladyman, 2009, 376]
Here, the point is that computation might refer to a concrete, physical process
or to an abstract, logical-mathematical process. (Is process even the correct
term here? Perhaps something less dynamic, such as structure, might be
more appropriate. [Ladyman, 2009, 377] suggests the term transformation,
which has a satisfactory feel of abstract dynamics to it.)
So, whether our computer is performing or implementing a chess game may

depend on whether the computation that it is performing is a physical one (in
which case, perhaps it is dependent on human intentions) or a theoretical one (in
which case, perhaps it is independent, possibly because it is uninterpreted, hence
non-intentional).
But those perhapss in the previous paragraph need to be read with caution.
Realism, according to [Ladyman, 2009, 377], is the view that what a physical
computation is ought to be independent of human intentions:
If whether or not the human nervous system implements particular
computations is not a natural fact about the world that is independent
of whether we represent it as doing so, then the computational theory
of mind fails to naturalise the mind. Similarly, claims that the universe
is a computer, that computation occurs in bacteria and cells, and so on
must be understood merely as claims about how we think about these
systems if realism about computation is false.
Searles and Putnams arguments are arguments against realism in computation:

Searle. . . argues against realism about computation by reductio: if it
were sufficient for some physical system to implement a particular
computation that within it there be a pattern of activity with a
particular structure, then since most macroscopic physical systems
are hugely complex, there will be some pattern of activity within
almost every macroscopic system of almost any particular structure,
and so almost every macroscopic physical system would implement
any given computation. Hence, it cannot be purely in virtue of the
patterns of physical activity that systems implement computations,
and since these are the only features of physical systems about
which we should be realists, realism about computation fails.
Putnam. . . argues that every ordinary open system realises every finite
automaton because it is always possible to take a sequence of the
successive temporal states of such a system as representing the
successive states of the automaton. Clearly, the basic idea of both
arguments is that provided a physical system has sufficient complexity
it can be said to implement any computation and therefore there is no
fact of the matter about systems implementing one computation in
particular. [Ladyman, 2009, 378, my italics]
So its the focus on the internal computation, so to speak, that provides
the slippage allowing every physical system to implement every theoretical
computation.
The analysis in [Ladyman, 2009, 378] supports this idea that it is the
interpretation of the inputs and outputs that seems to matter:
A common response to these arguments (for example,
[Chalmers, 1996b] is to claim that for a computation to be
implemented depends not only on what a system does but also on

what it would do were its initial state different in an appropriate way.
For example, a handheld calculator can only be said to calculate
the sum of 7 and 5 when the appropriate buttons are pressed and
the screen outputs 12, because had different buttons been pressed it
would have given a different output. Similarly, a logic gate that gives
the correct output 1 when the inputs represent 1 and 1, is only
computing AND if, had the inputs represented 1 and 0 it would
have given the output 0. In general, functions are maps from many
different values in the domain of the function to values in its range,
but in general implementations are only of a particular instance of the
function, in others words, they are processes that evolve from a state
that represents one value in the domain to the state that represents the
corresponding value in the range.
Of course, the interpretation of the inputs and outputs is dependent on human

intentions, as Ladyman later observes. But he also observes that even this
is debateable, because there are theories of teleosemantics (purposeful
semantics) that allow certain semantic interpretations to be independent of
intentions. This, however, is beyond our scope. We explore the notion of
implementation in Ch. 14.
39. On the great insights, see also Ch. 14, Notes for Next Draft No. 8.
40. On the BJ insight (7.6.3, No. 2), for more on the power of abstraction, see the
first few paragraphs of [Antoy and Hanus, 2010].
41. [Rescorla, 2012a] argues that computation is not sensitive to meaning or

semantic properties (1, p. 703). More precisely, he argues that if a
computational process were to be sensitive to semantic properties, then it would
have to violate either a condition that he calls Syntactic Rules or a condition
that he calls Freedom, and that such a semantically sensitive computation
would have to have an indigenous semantics, not an inherited semantics. He
defines these terms as follows:
SYNTACTIC RULES: Computation is manipulation of syntactic

entities according to mechanical rules. We can specify those rules
in syntactic terms, without mentioning semantic properties such as
meaning, truth, or reference. (3, p. 707).
FREEDOM: We can offer a complete syntactic description of

the systems states and the mechanical rules governing transitions
between states (including any interaction with sensory and motor
transducers), while leaving semantic interpretation unconstrained.
More precisely, it is metaphysically possible for the system to satisfy
our syntactic description while instantiating arbitrarily different
semantic properties. (3, p. 708).
Inherited meanings arise when the systems semantic properties are

assigned to it by other systems, through either explicit stipulation
or tacit convention. Nothing about the system itself helps generate
its own semantics. For instance, words in a book have inherited
meanings. Indigenous meanings arise when a system helps generate
its own semantics. Indigenous meanings do not arise merely from
external assignment. They arise partly from the systems own
activity, perhaps with ample help from other factors, such as causal,
evolutionary, or design history. Virtually all commentators agree that
the mind has indigenous meanings. (3, pp. 707708)
(He notes that inherited meanings are similar to what others call derived
meanings or derived intentionality, and that indigenous meanings are similar to
what others call original meanings or intentionality. Extrinsic and intrinsic
might be other terms that are relevantly similar.)
42. On Peacocke and on the Fodorian chess-war example:

[Rescorla, 2012a, 2.2, p. 707] agrees with Peacockes analysis, but points out
that One can consistently hold the following combination of views: a physical
systems semantic properties inform its computational nature, so they help type-
identify its computational states; yet the transitions between computational states
are not sensitive to semantic properties. Whether a computer is playing chess
or simulating a Civil War battle depends on a (derived, extrinsic, or inherited)
semantic interpretation assigned to it by an external agent. But the fact that
a single computer can be doing both (playing chess and simulating a battle),
simultaneously, depends on the fact that the computations are purely syntactic
and that the two different semantic interpretations are interpretations of that self-
same syntax): What the computer is doing is the realm of semantics; how it is
doing it is in the realm of syntax.
43. The halting theorem is often attributed to Turing in his

1936 paper. In fact, Turing did not discuss the halting
problem, which was introduced by Martin Davis in about 1952.
[Copeland and Proudfoot, 2010, 248, col. 2]
The authors dont elaborate on this. Perhaps it has something to do with the
difference between halting and circle-free?
44. This point might be equally appropriate in Ch. 19. The unsolvability of the
Entscheidungsproblem due to the non-computability of the Halting Problem
can be used as a response to the Lady Lovelace objection to the possibility
of AI, that is, the objection that computers can only do what they have
been programmed to do: According to [Abramson, 2014, his italics, my
interpolation], Turings 1936 paper shows
that the concepts of determinism and predictability fall apart.

Computers, which can be understood as the finite unfoldings of a
particular Turing machine, are completely deterministic. But there is
no definite procedure for figuring out, in every case, what theyll do

[because of the Halting Problem]: if you could, then you would have
a definite procedure for deciding whether any statement of arithmetic
is true or not. But there is no such procedure for the one, so there is
no such procedure for the other. Computers are, in the general case,
unpredictable, even by someone who knows exactly how they work.
45. Were the ancient Babylonians really creating algorithms? Insofar as what
they were doing fits our informal notion of algorithm, the answer looks to be:
yes. But [Mahoney and Haigh (ed.), 2011, 39] cautions against applying 20th-
century insights to ancient times:
When scientists study history, they often use their modern tools
to determine what past work was really about; for example,
the Babyonian mathematicians were really writing algorithms.
But thats precisely what was not really happening. What was
really happening was what was possible, indeed imaginable, in the
intellectual environment of the time; what was really happening
was what the linguistic and conceptual framework then would allow.
The framework of Babylonian mathematics had no place for a
metamathematical notion such as algorithm.
He even cites the computer scientist Richard Hamming as making the same
point in an essay on the history of computing, that we would [like to] know
they thought when they did it: What were Babylonian mathematicians thinking
when they created what we now call algorithms?
But is that fair? Yes, it would be nice to know what they were really thinking,
but isnt it also the case that, whatever they thought they were doing, we can
describe it in terms of algorithms? (This is the difference between a de dicto and
a de re approach to history!)
46. . . . every physical process instantiates a computation insofar as it
progresses from state to state according to dynamics prescribed by
the laws of physics, that is, by systems of differential equations.
[Fekete and Edelman, 2011, 808]
This suggests the following very odd and very liberal definition: Something is a
computation =def it is a progression from state to state that obeys a differential
equation. This definition is liberal, because it seems to go beyond the limitations
of a Turing-machine-like algorithm. Thats not necessarily bad; for one thing, it
subsumes both analog and discrete computations under one rubric. But its odd,
because its not clear (to me; check on this!) that Turing-machine algorithms are
describably by differential equations; if they are, so much the better.

A Books and Essays:
1 Blass, Andreas; & Gurevich, Yuri (2003), Algorithms: A Quest for Absolute
Definitions, Bulletin of the European Association for Theoretical Computer Science
(EATCS), No. 81 (October): 195225.
Pre-print accessed 22 April 2014 from:
http://research.microsoft.com/en-us/um/people/gurevich/opera/164.pdf
2 Carey, Kevin (2010), Decoding the Value of Computer Science, Chronicle of Higher
Education (7 November);
http://chronicle.com/article/Decoding-the-Value-of-Computer/125266/
An essay arguing for the value of algorithmic thinking in fields other than computer
science (including finance and journalism.
3 Cerf, Vinton G. (2014), Virtual Reality Redux, Communications of the ACM 57(1)
(January): 7; accessed 21 January 2014 from
http://cacm.acm.org/magazines/2014/1/170860-virtual-reality-redux/fulltext
Has some interesting comments that are relevant to the insight about binary
representation of information.
4 Chaitin, Gregory (2006), How Real Are Real Numbers?, International Journal of
Bifurcation and Chaos; online at:
2006 version: http://www.cs.auckland.ac.nz/chaitin/olympia.pdf
2009 version: http://www.umcs.maine.edu/chaitin/wlu.html
(both accessed 6 November 2012).
On the halting problem, the non-existence of real numbers(!), and the idea that
everything is software, God is a computer programmer,. . . and the world is. . . a giant
computer(!!).
On the non-reality of real numbers, see also [Knuth, 2001, 174175].
5 Chaitin, Gregory (2006), The Limits of Reason, Scientific American (March): 7481;
online at: http://www.umcs.maine.edu/chaitin/sciamer3.pdf
Ideas on complexity and randomness originally suggested by Gottfried W. Leibniz
in 1686, combined with modern information theory [in particular, implications of the
Halting Problem], imply that there can never be a theory of everything for all of
mathematics.
6 Chisum, Donald S. (1985-1986), The Patentability of Algorithms, University of
Pittsburgh Law Review 47: 9591022.
7 Cook, Byron; Podelski, Andreas; & Rybalchenko, Andrey (2011), Proving Program
Termination, Communications of the ACM 54(5) (May): 8898.
In contrast to popular belief, proving termination is not always impossible.. . . [M]any
have drawn too stong of a conclusion [from Turings proof that Halting is non-
computable]. . . and falsely believe we are always unable to prove termination, rather
than more benign consequence that we are unable to always prove termination.. . . In
our new problem statement we will still require that a termination proving tool always
return answers that are correct, but we will not necessarily require an answer (my
italics).
8 Copeland, B. Jack (1996), What Is Computation?, Synthese 108: 335359.

9 Copeland, B. Jack (1997), The Broad Conception of Computation, American Behavioral
Scientist 40(6) (May): 690716.
An essay on hypercomputation (or non-classical computation), but the introductory
section (pp. 690698) contains an enlightening discussion of the scope and limitations
of Turings accomplishments.
10 Copeland, B. Jack (2004), Computation, in Luciano Floridi (ed.), The Blackwell Guide
to the Philosophy of Computing and Information (Malden, MA: Blackwell).
11 Corballis, Michael C. (2007), The Uniqueness of Human Recursive Thinking, American
Scientist 95 (May-June): 240248.
12 Crossley, John N.; & Henry, Alan S. (1990), Thus Spake Al-Khwarizmi: A Translation
of the Text of Cambridge University Library ms. Ii.vi.5, Historia Mathematica 17(2):
103131.
13 Farkas, David K. (1999), The Logical and Rhetorical Construction of Procedural
Discourse, Technical Communication 46(1) (February): 4254; online at:
http://www.hcde.washington.edu/sites/default/files/people/docs/proceduraldiscourse.pdf
14 Feldman, Yishai A. (1993), Finite-State Machines, in Allen Kent & James G. Williams
(eds.), Encyclopedia of Computer Science and Technology (New York: Marcel Dekker)
28(13): 387418.
15 Floridi, Luciano (1999), The Digital Workshop, Ch. 2 of Philosophy and Computing:
An Introduction (London: Routledge).
16 Folina, Janet (1998), Churchs Thesis: Prelude to a Proof, Philosophia Mathematica 6:
302323.
17 Gurevich, Yuri (2011), What Is an Algorithm?, Technical Report MSR-TR-2011-116
(Redmond, WA: Microsoft Research); online at:
http://research.microsoft.com/pubs/155608/209-3.pdf (accessed 6 November 2012).
18 Harel, David (1980), On Folk Theorems, Communications of the ACM 23(7) (July):
379389,
http://www.univasf.edu.br/marcus.ramos/pc-2008-2/p379-harel.pdf
(accessed 30 September 2015).
On the history of the Boehm-Jacopini Theorem.
19 Harnad, Stevan (ed.) (1994), Special Issue on What Is Computation?, Minds and
Machines 4(4), containing:
Harnad, Stevan, Preface, pp. 377378.
Harnad, Stevan, Computation Is Just Interpretable Symbol Manipulation; Cog-
nition Isnt, pp. 379390; online at:
http://users.ecs.soton.ac.uk/harnad/Papers/Harnad/harnad94.computation.cognition.html
Chalmers, David J., On Implementing a Computation, pp. 391402.
Chrisley, Ronald L., Why Everything Doesnt Realize Every Computation,
pp. 403420.
MacLennan, Bruce J., Words Lie in Our Way, pp. 421437.
Kentridge, Robert W., Symbols, Neurons, Soap-Bubbles, and the Neural
Computation Underlying Cognition, pp. 439449.
Boyle, C. Franklin, Computation as an Intrinsic Property, pp. 451467.

Bringsjord, Selmer, Computation, among Other Things, Is beneath Us,
pp. 489490; portions online at:
http://homepages.rpi.edu/brings/SELPAP/beneath.ab.html
20 Hershenov, David B. (2002), Scattered Artifacts, Southern Journal of Philosophy 60:
211216.
Accessed 1 August 2014 from:
http://www.acsu.buffalo.edu/dh25/articles/scattered.pdf
Is a watch undergoing repair, its parts spread actross a craftsmans table still a
watch? This essay raises some interesting considerations about where in a computer
the computation happens (that is, where is the input converted to output?).
For more on the topic of scattered objects, see:
http://theotodman.com/Notes/Notes 1/Notes 123.htm
(accessed 1 August 2014).
21 Hill, Robin (2013), What an Algorithm Is, and Is Not, Communications of the ACM,
56(6) (June): 89;
http://mags.acm.org/communications/june 2013/?pg=11#pg11
22 Horgan, John (1990), Profile: Claude E. Shannon; Unicyclist, Juggler and Father of
Information Theory, Scientific American (January): 22, 22A22B.
23 Knuth, Donald (1973), Basic Concepts: 1.1: Algorithms, The Art of Computer
Programming, Second Edition (Reading, MA: Addison-Wesley): xiv9.
24 Knuth, Donald (1974), Computer Science and Its Relation to Mathematics, American
Mathematical Monthly 81(4) (April): 323343.
25 Korfhage, Robert R. (1993), Algorithm, in Anthony Ralston & Edwin D. Reilly (eds.),
Encyclopedia of Computer Science, 3rd Edition (New York: Van Nostrand Reinhold): 27
29.
26 Kreisel, Georg (1987), Churchs Thesis and the Ideal of Informal Rigour, Notre Dame
Journal of Formal Logic 28(4): 499519.
A paper by a well-known logician arguing that Churchs Thesis can be proved.
27 Leiber, Justin (2006), Turings Golden: How Well Turings Work Stands Today,
Philosophical Psychology 19(1): 1346.
28 Manzano, Mara (1997), Alonzo Church: His Life, His Work and Some of His Miracles,
History and Philosophy of Logic 18: 211232.
29 Mendelson, Elliott (1990), Second Thoughts about Churchs Thesis and Mathematical
Proofs, Journal of Philosophy 87(5) (May): 225233.
For a response, see:
Bringsjord, Selmer (1993), The Narrational Case against Churchs Thesis, accessed
17 April 2014 from:
http://homepages.rpi.edu/brings/SELPAP/CT/ct/
30 Moschovakis, Yiannis N. (1998), On Founding the Theory of Algorithms, in H.G. Dales
& G. Oliveri (eds.), Truth in Mathematics (Oxford: Clarendon Press): 71104; online at:
http://www.math.ucla.edu/ ynm/papers/foundalg.ps
31 Moschovakis, Yiannis N. (2001), What Is an Algorithm?, in B. Engquist & W. Schmid

(eds.), Mathematics Unlimited: 2001 and Beyond (Berlin: Springer): 918936.
32 Piccinini, Gualtiero (2007), Computationalism, the Church-Turing Thesis, and the
Church-Turing Fallacy, Synthese 154: 97120.
33 Rescorla, Michael (2007), Churchs Thesis and the Conceptual Analysis of
Computability, Notre Dame Journal of Formal Logic 48(2): 253280.
Rescorla argues that Churchs Thesis (that a number-theoretic function is intuitively
computable iff it is recursive) is not the same as Turings Thesis (that a number-
theoretic function is intuitively computable iff a corresponding string-theoretic
function [that represents the number-theoretic one] is computable by a Turing
machine).
See also:
Shapiro, Stuart C. (1977), Representing Numbers in Semantic Networks:
Prolegomena, in Proceedings of the 5th International Joint Conference on Artificial
Intelligence (Los Altos, CA: Morgan Kaufmann): 284.
34 Schulman, Ari N. (2009), Why Minds Are Not Like Computers, The New Atlantis
(Winter): 46-68.
35 Shagrir, Oron (1999), What Is Computer Science About?, The Monist 82(1): 131149.
Despite its title, this paper is more about what computers are and what computation
is.
36 Shapiro, Stewart (1983), Remarks on the Development of Computability, History and
Philosophy of Logic 4: 203220.
Discusses the notion of computability from a historical perspective; contains a
discussion of Churchs thesis.
37 Shapiro, Stewart (1993), Understanding Churchs Thesis, Again, Acta Analytica 11:
S.5977.
tries to show that Churchs thesis can be proved mathematically. . .
38 Sieg, Wilfried (1994), Mechanical Procedures and Mathematical Experience, in George
Alexander (ed.), Mathematics and Mind (New York: Oxford University Press): 71117.
Accessed 10 December 2013 from:
Contains a detailed history and analysis of the development of the formal notions of
algorithm in the 1930s and after.
39 Sieg, Wilfried (1997), Step by Recursive Step: Churchs Analysis of Effective
Calculability, Bulletin of Symbolic Logic 3(2) (June): 154180.
40 Sieg, Wilfried (2008), Church without Dogma: Axioms for Computability, in
S.B. Cooper, Benedikt Lowe, & Andrea Sorbi (eds), New Computational Paradigms:
Changing Conceptions of What Is Computable (Springer): 139152; preprint (2004) online
at:
http://www.hss.cmu.edu/philosophy/sieg/Church%20without%20dogma.pdf
41 Sieg, Wilfried (2006), Godel on Computability, Philosophia Mathematica 14 (June):
189207; online at:
http://www.hss.cmu.edu/philosophy/sieg/A%20G del%20on%20Computability.pdf
42 Smith, Brian Cantwell (2002), The Foundations of Computing, in Matthias Scheutz

(ed.), Computationalism: New Directions (Cambridge, MA: MIT Press): 2358.
43 Soare, Robert I. (2009), Turing Oracle Machines, Online Computing, and Three
Displacements in Computability Theory, Annals of Pure and Applied Logic 160: 368
399.
Preprint accessed 2 July 2014 from:
http://www.people.cs.uchicago.edu/ soare/History/turing.pdf
and published version accessed 2 July 2014 from:
http://ac.els-cdn.com/S0168007209000128/1-s2.0-S0168007209000128-
main.pdf? tid=8258a7e2-01ef-11e4-9636-
00000aab0f6b&acdnat=1404309072 f745d1632bb6fdd95f711397fda63ee2
2, Origins of Computabilty and Incomputablity, contains a good summary of the
history of both Turings and Godels accomplishments.
44 Vardi, Moshe Y. (2011), Solving the Unsolvable, Communications of the ACM 54(7)
(July): 5; online at:
http://www.inf.unibz.it/calvanese/teaching/tc/material/solving-the-unsolvable-CACM-
2011-07.pdf
See also:
Ledgard, Henry; & Vardi, Moshe Y. (2011), Solved, for All Practical Purposes
(letter to the editor, and reply), Communications of the ACM 54(9) (September): 7;
online at: http://tinyurl.com/LedgardVardi2011
45 Wangsness, T.; & Franklin, J. (1966), Algorithm and Formula , Communications of
the ACM 9(4) (April): 243.
The published answers:
i. Huber, Harmut G.M. (1966), Algorithm and Formula, Communications of the
ACM 9(9) (September): 653654.
ii. Knuth, Donald (1966), Algorithm and Program: Information and Data,
Communications of the ACM 9(9) (September): 654.
46 On heuristics vs. algorithms:
i. Findler, Nicholas V. (1993), Heuristic, in Anthony Ralston & Edwin D. Reilly
(eds.), Encyclopedia of Computer Science, 3rd Edition (New York: Van Nostrand
Reinhold): 611612.
ii. Korf, Richard E. (1992), Heuristics, in Stuart C. Shapiro (ed.), Encyclopedia of
Artificial Intelligence, 2nd Edition (New York: John Wiley & Sons): 611615.
iii. Newell, Allen, & Simon, Herbert A. (1976), Computer Science as Empirical
Inquiry: Symbols and Search, Communications of the ACM 19(3) (March): 113
126.
iv. Oommen, B. John; & Rueda, Luis G. (2005), A Formal Analysis of Why Heuristic
Functions Work, Artificial Intelligence 164(1-2) (May): 122.
v. Rapaport, William J. (1998), How Minds Can Be Computational Systems, Journal
of Experimental and Theoretical Artificial Intelligence 10: 403419.
vi. Romanycia, Marc H.J., & Pelletier, Francis Jeffry (1985), What Is a
Heuristic?, Computational Intelligence 1: 4758.
vii. Shapiro, Stuart C. (1992), Artificial Intelligence, in Stuart C. Shapiro (ed.),

Encyclopedia of Artificial Intelligence, 2nd Edition (New York: John Wiley &
Sons): 5457; revised version appears in Anthony Ralston & Edwin D. Reilly (eds.),
Encyclopedia of Computer Science, 3rd Edition (New York: Van Nostrand Reinhold,
1993): 8790.
viii. Simon, Herbert; & Newell, Allen (1958), Heuristic Problem Solving: The Next
Advance in Operations Research, Operations Research 6(1) (January-February): 1
10.
In addition to its discussion of the division of labor (see 6.4.3, above) and its
infamous prediction that a computer would become the world chess champion
by 1968, this classic paper also distinguishes algorithmic problem solving of
well-structured problems from heuristic problem solving of ill-structured
problems.
ix. Thagard, Paul (2007), Coherence, Truth, and the Development of Scientific
Knowledge, Philosophy of Science 74 (January): 2847.
Not about heuristics, but presents a theory of approximate truth that bears a
close resemblance to the idea that a heuristic is an algorithm that computes an
approximately correct answer.
B Websites:
1 Chazelle, Bernard (2006), The Algorithm: Idiom of Modern Science
http://www.cs.princeton.edu/chazelle/pubs/algorithm.html
2 Etymology of Compute
http://www.cse.buffalo.edu/rapaport/584/computetymology.html
3 OConnor, John J.; & Robertson, Edmund F. (1999), Abu Jafar Muhammad ibn Musa
Al-Khwarizmi
http://www-groups.dcs.st-and.ac.uk/history/Mathematicians/Al-Khwarizmi.html
4 PolR (2009), An Explanation of Computation Theory for Lawyers, Groklaw
(11 November),
http://www.groklaw.net/article.php?story=20091111151305785,
http://www.groklaw.net/staticpages/index.php?page=20091110152507492,
and
http://www.groklaw.net/pdf2/ComputationalTheoryforLawyers.pdf
C Rajagopalan, Bala (ca. 2011), Halting Problem Is Solvable (joke),
http://www.netfunny.com/rhf/jokes/89q4/halting.760.html
Figure 7.2: source: https://xkcd.com/844/

Figure 7.3: source: http://abstrusegoose.com/432

Chapter 8
Turings Analysis of
Computation
Turings Machines. These machines are humans who calculate.

[Wittgenstein, 1980, 191e, 1096]1
[A] human calculator, provided with pencil and paper and explicit
instructions, can be regarded as a kind of Turing machine. [Church, 1937]
You can build an organ which can do anything that can be done, but
you cannot build an organ which tells you whether it can be done.
[von Neumann, 1966], cited in [Dyson, 2012a].
Figure 8.1: 2009

c by ??
1 The quotation, though insightful, is somewhat confusingly put. Better would have been: these machines
are Turings mechanical model of humans who calculate. [Kripke, 2013, 96fn12]
289
290 CHAPTER 8. TURINGS ANALYSIS OF COMPUTATION
Figure 8.2: YEAR

c by ??
8.1 Required Reading:

Turing, Alan M. (1936), On Computable Numbers, with an Application to the
Entscheidungsproblem, Proceedings of the London Mathematical Society, Ser. 2,
Vol. 42 (1937): 230265.
Reprinted, with corrections, in Martin Davis (ed.), The Undecidable: Basic Papers
on Undecidable Propositions, Unsolvable Problems and Computable Functions
(New York: Raven Press, 1965): 116154.
Online at:
http://www.cs.virginia.edu/robins/Turing Paper 1936.pdf (accessed 6 November
2012).
1. Concentrate on the informal expository parts; the technical parts are, of course,
of interest, but are rather difficult to follow, incorrect in many parts, and can be
skimmed.
2. In particular, concentrate on:
16
(study the simple examples of Turing machines carefully;
skim the complex ones)
and 9, part I
(which elaborates on what it is that a human computer does).
3. 7 describes the universal Turing machine;
8 describes the Halting Problem.
You can skim these sections (thats skim, not skip!)

8.2 Introduction
If there is a single document that could be called the foundational document of
computer science, it would be Alan Mathison Turings 1936 article, On Computable
Numbers, with an Application to the Entscheidungsproblem, which appeared in the
journal Proceedings of the London Mathematical Society, Series 2. In this paper, Turing
(who was only about 24 years old at the time) accomplished 4 major goals:
1. He gave what is considered to be the clearest and most convincing mathematical

analysis of computation (what is now called, in his honor, a Turing machine).
2. He described a universal Turing machine, which is a mathematical version of

a programmable computer.
3. He proved that there were some functions that were not computable, thus
showing that computation was not a trivial property. After all, if all
functions were computablewhich no doubt would be a very nice featurethen
computability would not really be a very interesting or special property. But,
because some functions are not computable, computability is a property that
only some (but not all) functions have, and so it becomes more interesting.
4. And (I will suggest) he wrote the first AI program (see 8.14, below).
Thus, arguably, in this paper, he created the modern discipline of computer science.
Because this paper was so important and so influential, it is well worth reading.
Fortunately, although parts of it are tough going (and it contains some errors),2 much
of it is very clearly written. It is not so much that the tough parts are difficult or hard
to understand, but they are full of nitty, gritty details that have to be slogged through.
Fortunately, Turing has a subtle sense of humor, too.
In this chapter, I will provide a guide to reading parts of Turings paper slowly and
carefully, by actively thinking about it. (A wonderful guide to reading all of it slowly
and carefully is [Petzold, 2008].)3
8.3 The Entscheidungsproblem

First, lets look at the title, in particular its last word: Entscheidungsproblem. This is
a German noun that was well known to mathematicians in the 1930s; Entscheidung
means decision, -s represents the possessive,4 and problem means problem. So,
an Entscheidungsproblem is a decision problem, and the Decision Problem was the
problem of finding an algorithm that would take two things as input: (1) a formal logic
L and (2) a proposition L in the language for that logic, and that would output either
yes, if L was a theorem of that logic, or else no, if L was a theorem of that logic
2 Some of which Turing himself corrected [Turing, 1938]. For a more complete debugging, see
[Davies, 1999].
3 On the value of slow and active reading in general, see my website [Rapaport, 2014], and
[Fletcher, 2007, Blessing, 2013].

4 Just as in English, so Entscheidungs means decisions.
(that is, if L was not a theorem of L). In other words, the Decision Problem was the
problem of finding a general algorithm for deciding whether any given proposition was
a theorem.
Wouldnt that be nice? Mathematics could be completely automated: Given any
mathematical proposition, one could apply this general algorithm to it, and you would
be able to know if it were a theorem or not. Turing was fascinated by this problem, and
he solved it. Along the way, he invented computer science! He solved the Decision
Problem in the negative, by showing that no such algorithm existedweve already
seen how: He showed that there was at least one problem (the Halting Problem) for
which there was no such algorithm.5
8.4 Paragraph 1
One of the best ways to read is to read slowly and actively. This is especially
true when you are reading a technical paper, and even more especially when you
are reading mathematics. Reading slowly and actively menas reading each sentence
slowly, thinking about it actively, and making sure that you understand it before reading
the next sentence. (One way to make sure that you understand it is to ask yourself why
the author said it, or why it might be true.) If you dont understand it (after reading
it slowly and actively), then you should re-read all of the previous sentences to make
sure that you really understood them. Whenever you come to a sentence that you really
dont understand, you should ask someone to help you understand it.
(Of course, it could also be the case that you dont understand it because it isnt
true, or doesnt follow from what has been said, or is confused in some wayand not
because its somehow your fault that you dont understand it!)
8.4.1 Sentence 1
So, lets begin our slow and active reading of Turings paper with the first sentence of
the first paragraph:
The computable numbers may be described briefly as the real

numbers whose expressions as a decimal are calculable by finite means.
[Turing, 1936, 230, my italics]6
The word computable occurs in quotes here, because Turing is using it in an

informal, intuitive sense. It is the sense that he will make mathematically precise in the
rest of the paper.
Real numbers are all of the numbers on the continuous number line. They consist
of:
1. the rational numbers, which consist of:
(a) the integers, whichin turnconsist of:

5 For more on the decision problem, see 6.5.
6 In the rest of this chapter, citations from [Turing, 1936] will just be cited by section or page number.
8.4. PARAGRAPH 1 293
i. the non-negative natural numbers (0, 1, 2, . . . ), and

ii. the negative natural numbers (1, 2, . . .), and
(b) all other numbers that can be expressed as a ratio of integers (or that can be
expressed in decimal notation with repeating decimals), and
2. the irrational numbers(that is, those numbers that cannot be expressed as a ratio
of integers, such as , 2, etc.)

But the real numbers do not include the complex numbers, such as 1.
Every real number can be expressed as a decimal, that is, in decimal notation.
For instance:
1 = 1.0 = 1.00 = 1.000 (etc.),
1
2 = 0.5 = 0.50 = 0.500 = 0.5000 (etc.),
1
3 = 0.33333. . .
These are all rational numbers and examples of repeating decimals. But the reals also
include the irrational numbers, which have non-repeating decimals:7
= 3.1415926535 . . .,

2 = 1.41421356237309 . . .
Given a real number, is there an algorithm for computing its decimal

representation? If so, then its decimal [is] calculable by finite means (because
algorithms must be finite, as we saw in 7.5).8
8.4.2 Last Sentence

Now, if we were really going to do a slow (and active!) reading, we would next move
on to sentence 2. But, in the interests of keeping this chapter shorter than a full book
(and so as not to repeat everything in [Petzold, 2008]), well skip to the last sentence
of the paragraph:
According to my definition, a number is computable if its decimal can be

written down by a machine. (p. 230, my italics.)
This is probably best understood as an alternative way of expressing the first

sentence: To be calculable by finite means is to be capable of being written down
by a machine. Perhaps this way of putting it extends the notion a bit, because it
7 Decimal notation is also called base-10 notation. It is merely one example (another being binary
or base-2notation) of what is more generally known as radix notation or positional notation; see
http://en.wikipedia.org/wiki/Radix and http://math.comsci.us/radix/radix.html (both accessed 22 November
2013).
8 For another discussion of the computation of real numbers, see [Hartmanis and Stearns, 1967]. For a
commentary on the reality of real numbers, see [Knuth, 2001, 174175].

suggests that if a number is calculable by finite means, then that calculation can be
done automatically, that is, by a machinewithout human intervention. And that, after
all, was the goal of all of those who tried to build calculators or computing machines,
as we saw in Ch. 6. So, Turings goal in this paper is to give a mathematical analysis
of what can be accomplished by any such machine (and then to apply the results of this
analysis to solving the Decision Problem).
8.5 Paragraph 2
8.5.1 Sentence 1
In 9, 10 I give some arguments with the intention of showing that
the computable numbers include all numbers which could naturally be
regarded as computable. (p. 230.)
We will look at some of those arguments later, but, right now, lets focus on
the phrase naturally be regarded as computable. This refers to the same informal,
intuitive, pre-theoretical notion of computation that his quoted use of computable
referred to in the first sentence. It is the sense in which Hilbert wondered about which
mathematical problems were decidable, the sense in which people used the phrase
effective computation, the sense in which people used the word algorithm, and so
on. It is the sense in which people (mathematicians, in particular) can compute. And
one of its crucial features is that it be finite.
8.5.2 Last Sentence

Once again, well skip to the last sentence of the paragraph:
The computable numbers do not, however, include all definable numbers,

and an example is given of a definable number which is not computable.
(p. 230, my italics)
As we noted before (8.2), it is much more interesting if not all functions

or numbersare computable. Any property that everything has is not especially
interesting. But if there is a property that only some things have (and others lack), then
we can begin to categorize those things and thus learn something more about them.
So Turing is promising to show us that computability is an interesting (because not
a universal) property. And hes not going to do that by giving us some abstract (or
transcendental) argument; rather, hes actually going to show us a non-computable
number (and, presumably, show us why its not computable). Weve already seen what
this is: Its the (Godel) number for an algorithm (a Turing machine) for the Halting
Problem. So, in this chapter, well skip that part of Turings paper. Well also skip the
rest of the introductory section of the paper, which simply lays out what Turing will
cover in the rest of the paper.
8.6. SECTION 1, PARAGRAPH 1 295
8.6 Section 1, Paragraph 1

Lets move on to Turings Section 1, Computing Machines. Well look at the first
paragraph and then digress to Turings Section 9 before returning to this section.
Here is the first paragraph of Section 1:
We have said that the computable numbers are those whose decimals are
calculable by finite means. This requires rather more explicit definition.
No real attempt will be made to justify the definitions given until we reach
9. (p. 231.)
This is why we will digress to that section in a moment. But first lets continue with
the present paragraph:
For the present I shall only say that the justification [of the definitions]
lies in the fact that the human memory is necessarily limited. (p. 231, my
interpolation and italics.)
Turings pointfollowing Hilbertis that we humans do not have infinite means at
our disposal. We all eventually die, and we cannot work infinitely fast, so the number
of calculations we can make in a single lifetime is finite.
But how big is finite? Lets suppose, for the sake of argument, that a typical
human (named Pat) lives as long as 100 years. And lets suppose that from the time
Pat is born until the time Pat dies, Pat does nothing but compute. Obviously, this is
highly unrealistic, but I want to estimate the maximum number of computations that
a typical human could perform. The actual number will, of course, be far fewer.
How long does a computation performed by a human take? Lets suppose that the
simplest possible computation (following our notion of a basic function in 7.7.2) is
computing the successor of a natural number, and lets suppose that it takes as long as
1 second. In Pats lifetime, approximately 3,153,600,000 successors can be computed
(because thats approximately the number of seconds in 100 years). Are there any
problems that would require more than that number of computations? Yes! It has been
estimated that the number of possible moves in a chess game is 10125 , which is about
10116 times as large as the largest number of computations that a human could possibly
perform. In other words, we humans are not only finite, we are very finite!
But computer scientists and mathematicians tend to ignore such human limitations
and pay attention only to the mathematical notion of finiteness. Even the mathematical
notion, which is quite a bit larger than the actual human notion (for more on this, see
[Knuth, 2001]), is still smaller than infinity, and so the computable numbers, as Turing
defines them, include quite a bit.
8.7 Section 9
8.7.1 Section 9, Paragraph 1
I want to skip now to Turings 9, The Extent of the Computable Numbers, because it
is this section that contains the most fascinating part of Turings analysis. Well return
to his 1 later. He begins as follows:
No attempt has yet been made [in Turings article] to show that the
computable numbers include all numbers which would naturally be
regarded as computable. (p. 249, my interpolation and italics.)
Here, Turing is comparing two notions of computability: the technical notion (signified
by the first occurrence of the word computablein scare quotes) and the informal
or natural notion. He is going to argue that the first includes the second. Presumably,
it is more obvious that the second (the natural notion) includes the first (the technical
notion), that is, that if a number is technically computable, then it is naturally
computable. The less obvious inclusion is the one that is more in need of support,
that if a number is naturally computable, then it is technically computable. But what
kind of argument would help convince us of this? Turing says:
All arguments which can be given are bound to be, fundamentally, appeals
to intuition, and for this reason rather unsatisfactory mathematically.
(p. 249.)
Why is this so? Because one of the two notionsthe natural oneis informal, and
so no formal, logical argument can be based on it. This is why Turings thesis (that is,
Churchs thesis, the Church-Turing thesis, or the computability thesis) is a thesis and
not a theoremit is a hypothesis and not something formally provable. Nonetheless,
Turing will give us appeals to intuition, that is, informal arguments, in fact, three
kinds:
The real question at issue is What are the possible processes which can
be carried out in computing a number? (p. 249.)
If Turing can answer this question, even informally, then we may be able to come up
with a formal notion that captures the informal one. That is the first step (and the only
one that we shall consider in this book).
8.7.2 Section 9, Subsection I

The first step is a direct appeal to intuition, and Turing notes that this argument
is only an elaboration of the ideas of 1 (p. 249) (which is why we have made this
digression to Turings 9 from his 1; when we return to his 1, we will see that it
summarizes his 9).
8.7.2.1 Section 9, Subsection I, Paragraph 1

The first part of the answer to the question, What are the possible processes which
can be carried out in computing a number?that is, the first intuition about natural
computationis this:
Computing is normally done by writing certain symbols on paper. (p. 249.)
So, we need to be able to write symbols on paper. Is this true? What kind of symbols?
And what kind of paper?
8.7. SECTION 9 297
8.7.2.1.1 Is it true? Is computing normally done by writing symbols on paper?

We who live in the 21st century might think that this is obviously false: Computers
dont have to write symbols on paper in order to their job. They do when we have the
computer print a document, but they dont when we are watching a YouTube video.
But remember that Turing is analyzing the natural notion of computing: the kind of
computing that humans do. And his model includes arithmetic computations. Those
typically are done by writing symbols on paper (or, perhaps, by imagining that we are
writing symbols on paper, as when we do a computation in our head).
8.7.2.1.2 What about the paper?
We may suppose this paper is divided into squares like a childs arithmetic
book. (p. 249.)
In other words, we can use graph paper! Presumably, we can put one symbol into each
square of the graph paper. So, for example, if were going to write down the symbols
for computing the sum of 43 and 87, we could write it like this:
-------
| |1| |
-------
| |4|3|
-------
|+|8|7|
-------
|1|3|0|
-------
We write 43 in two squares, then we write +87 in three squares beneath this, aligning
the ones and tens columns. To perform the computation, we compute the sum of 7 and
3, and write it as follows: The ones place of the sum (0) is written below 7 in the
ones column and the tens place of the sum (1) is carried to the square above the
tens place of 43. Then the sum of 1, 4, and 8 is computed and then written as follows:
The ones place of that sum, namely, 3 (which is the tens place of the sum of 43 and
87) is written below 8 in the tens column, and the tens place of that sumnamely,
1 (which is the hundreds place of the sum of 43 and 8namely, 1)is written in
the square to the left of that 3.
Turing continues:
In elementary arithmetic the two-dimensional character of the paper is

sometimes used. (p. 249.)
As we have just seen.
But such a use is always avoidable, and I think that it will be agreed that
the two-dimensional character of paper is no essential of computation.
(p. 249.)
In other words, we could have just as well (if not just as easily) written the computation
thus:
-----------------------
|1| |4|3|+|8|7|=|1|3|0|
-----------------------
Here, we begin by writing the problem 43+87 in five successive squares, followed,
perhaps, by an equals-sign. And we can write the answer in the squares following
the equals-sign, writing the carried 1 in an empty square somewhere else, clearly
separated (here, by a blank square) from the problem. So, the use of two-dimensional
graph paper has been avoided (at the cost of some extra bookkeeping). As a
consequence, Turing can say:
I assume then that the computation is carried out on one-dimensional

paper, i.e. on a tape divided into squares. (p. 249, my boldface.)
Here is the famous tape of what will become a Turing machine! (Note, though, that
Turing has not yet said anything about the length of the tape; at this point, it could be
finite.) We now have our paper on which we can write our symbols.
8.7.2.1.3 What about the symbols?
I shall also suppose that the number of symbols which may be printed
is finite. (p. 249, my boldface)
This is the first item that Turing has put a limitation on: There are only a finite number
of symbols.
(Actually, Turing is a bit ambiguous here: There might be an infinite number of
different kinds of symbols, but were only allowed to print a finite number of them.
Or there might only be a finite number of different kinds of symbolswith a further
vagueness about how many of them we can print: If the tape is finite, then we can
only print a finite number of the finite amount of symbols, but, if the tape is infinite,
we could print an infinite number of the finite amount of symbols. But it is clear from
what he says next that he means that there are only a finite number of different kinds
of symbols.)
Why finite? Because:
If we were to allow an infinity of symbols, then there would be symbols

differing to an arbitrarily small extent. (p. 249.)
There are two things to consider here: Why would this be the case? And why does
it matter? The answer to both of these questions is easy: If the human who is doing
the computation has to be able to identify and distinguish between an infinite number
of symbols, surely some of them may get confused, especially if they look a lot alike!
Would they have to look alike? A footnote at this point suggests why the answer is
yes:
8.7. SECTION 9 299
If we regard a symbol as literally printed on a square we may suppose that

the square is 0 x 1, 0 y 1. The symbol is defined as a set of points
in this square, viz. the set occupied by printers ink. (p. 249, footnote.)
That is, we may suppose that the square is 1 unit by 1 unit (say, 1 cm by 1 cm). Any
symbol has to be printed in this space. Imagine that each symbol consists of very tiny
dots of ink. To be able to print an infinite number of different kinds of symbols in
such a square, some of them are going to differ from others by just a single dot of ink,
and any two such symbols are going to differ to an arbitrarily small extent and, thus,
be impossible for a human to distinguish. So, the number of symbols which may be
printed must be finite in order for the human to be able to easily read them.
Is this really a limitation?
The effect of this restriction of the number of symbols is not very serious.
(p. 249.)
Why not? Because:
It is always possible to use sequences of symbols in the place of single

symbols. Thus an Arabic numeral such as 17 or 999999999999999 is
normally treated as a single symbol. Similarly in any European language
words are treated as single symbols. . . (pp. 249250.)
In other words, the familiar idea of treating a sequence of symbols (a string of

symbols, as mathematicians sometimes say) as if it were a single symbol allows us
to construct as many symbols as we want from a finite number of building blocks. That
is, the rules of place-value notation (for Arabic numerals) and of spelling (for words
in European languages) (that is, rules that tell us how to concatenate our symbols
to string them together) give us an arbitrarily large number (though still finite!) of
symbols.
What about non-European languages? Turing makes a (possibly politically
incorrect) joke:
. . . (Chinese, however, attempts to have an enumerable infinity of

symbols). (p. 250.)
Chinese writing is pictographic and thus would seem to allow for symbols that run
the risk of differing by an arbitrarily small extent, or, at least, that do not have to be
constructed from a finite set of elementary symbols. As Turing also notes, using a finite
number of basic symbols and rules for constructing complex symbols from them does
not necessarily avoid the problem of not being able to identify or differentiate them:
The differences from our point of view between the single and compound
symbols is that the compound symbols, if they are too lengthy, cannot be
observed at one glance. This is in accordance with experience. We cannot
tell at a glance whether 9999999999999999 and 999999999999999 are the
same. (p. 250.)
And probably you cant, either! So doesnt this mean that, even with a finite number
of symbols, were no better off than with an infinite number? Although Turing doesnt
say so, we can solve this problem using the techniques hes given us: Dont try to
write 15 or 16 occurrences of 9 inside one, tiny square: Write each 9 in a separate
square! (And then count them to decide which sequence of them contains 15 and
which contains 16, which is exactly how you can tell. . . whether 9999999999999999
and 999999999999999 are the same.)
The other thing that Turing leaves unspecified here is the minimum number of
elementary symbols we need. The answer, as we have seen, is: two (they could be
a blank and 1, or 0 and 1, or any other two symbols). Turing himself will use a
few more (just as we did in our addition example above, allowing for the 10 single-digit
numerals together with + and =).
8.7.2.1.4 States of mind. So, lets assume that, to compute, we only need a 1-
dimensional tape divided into squares and a finite number of symbols (minimally, two).
What else?
(*) The behaviour of the computer at any moment is determined by the

symbols which he is observing, and his state of mind at that
moment. (p. 250, my label and italics.)
I have always found this to be one of the most astounding and puzzling sentences!
computer? he? his? But it is only astounding or puzzling to those of us who live in
the late 20th/early 21st century, when computers are machines, not humans! Recall the
ad that we saw in 6.2 for a (human) computer from the 1892 New York Times. In 1936,
when Turing was writing this article, computers were still humans, not machines. So,
throughout this paper, whenever Turing uses the word computer, he means a human
whose job it is to compute. I strongly recommend replacing (in your minds ear, so to
speak) each occurrence of the word computer in this paper with the word clerk.9
So, the behavior of the clerk at any moment is determined by the symbols which he
[or she!] is observing. In other words, the clerk decides what to do next by looking at
the symbols, and which symbols the clerk looks at partially determines what the clerk
will do. Why do I say partially? Because the clerk also needs to know what to do
with them: If the clerk is looking at two numerals, should they be added? Subtracted?
Compared? The other information that the clerk needs is his or her state of mind.
What is that? Lets hold off on answering that question till we see what else Turing has
to say.
We may suppose that there is a bound B to the number of symbols or

squares which the computer [the clerk!] can observe at one moment.
If he[!] wishes to observe more, he must use successive observations.
(p. 250, my interpolation and boldface.)
This is the second kind of finiteness (so, we have a finite number of different kinds
of symbols and a finite number of them that can be observed at any given time.
9 Some writers, such as [Sieg, 1994], use the nonce word computor to mean a human who computes.
8.7. SECTION 9 301
This upper bound B can be quite small; in fact, it can equal 1 (and B = 1 in most
formal, mathematical presentations of Turing machines), but Turing is allowing for B
to be large enough so that the clerk can read a single word without having to spell
it out letter by letter, or a single numeral without having to count the number of its
digits (presumably, the length of 9999999999999999 exceeds any reasonable B for
humans). Successive observations will require the clerk to be able to move his or her
eyes one square at a time to the left or right.10
We will also suppose that the number of states of mind which need to
be taken into account is finite. (p. 250, my boldface.)
Here, we have a third kind of finiteness. But we still dont know exactly what a state
of mind is. Turing does tell us that:
If we admitted an infinity of states of mind, some of them will be
arbitrarily close and will be confused. (p. 250.)
just as is the case with the number of symbols. And he also tells us that the use of
more complicated states of mind can be avoided by writing more symbols on the tape
(p. 250), but why that is the case is not at all obvious at this point. (Keep in mind,
however, that we have jumped ahead from Turings 1, so perhaps something that he
said between then and now would have clarified this. Nevertheless, lets see what we
can figure out.)
8.7.2.1.5 Operations. So, a clerk who is going to compute needs only a (possibly
finite) tape divided into squares and a finite number of different kinds of symbols; the
clerk can look at only a bounded number of them at a time; and the clerk can be in
only a finite number of states of mind at a time. Moreover, what the clerk can do
(the clerks behavior) is determined by the observed symbols and his or her state of
mind. What kinds of behaviors can the clerk perform?
Let us imagine the operations performed by the computer [the clerk] to
be split up into simple operations which are so elementary that it is not
easy to imagine them further divided. (p. 250.)
These are going to be the basic operations, the ones that all other operations will be
constructed from. What could they be? This is an important question, because this is
going to be the heart of computation.
Every such operation consists of some change of the physical system
consisting of the computer [the clerk] and his[!] tape. (p. 250.)
So, what changes of the physical system can the clerk make? The only things that can
be changed are the clerks state of mind (i.e., the clerk can change him- or herself, so
to speak) and the tape, which would mean changing a symbol on the tape or changing
which symbol is being observed. What else could there be? Thats all we have to
manipulate: the clerk, the tape, and the symbols. And all weve been told so far is that
the clerk can write a symbol on the tape or observe one thats already written. Turing
makes this clear in the next sentence:
10 For more on this notion of bounds, see [Sieg, 2006, 16].
We know the state of the system if we know the sequence of symbols

on the tape, which of these are observed by the computer [by the clerk]
(possibly with a special order), and the state of mind of the computer [of
the clerk]. (p. 250.)
The system is the clerk, the tape, and the symbols. The only things we can know, or
need to know, are:
which symbols are on the tape,
where they are (their sequence),
which are being observed (and in which orderthe clerk might be looking from
left to right, from right to left, and so on), and
what the clerks (still mysterious) state of mind is.
Here is the first simple operation:
We may suppose that in a simple operation not more than one symbol is
altered. Any other changes can be split up into simple changes of this
kind. (p. 250, my boldface.)
Altering a single symbol in a single square is a simple operation. (And alterations
of sequences of symbols can be accomplished by altering the single symbols in the
sequence.) How do you alter a symbol? You replace it with another one; that is, you
write down a (possibly different) symbol. (And perhaps you are allowed to erase a
symbol, but that can be thought of as writing a special blank symbol, [.)
Which symbols can be altered? If the clerk is looking at the symbol in the first
square, can the clerk alter the symbol in the 15th square? Yes, but only by observing
the 15th square and then changing it:
The situation in regard to the squares whose symbols may be altered in this
way is the same as in regard to the observed squares. We may, therefore,
without loss of generality, assume that the squares whose symbols are
changed are always observed squares. (p. 250.)
But wait a minute! If the clerk has to be able to find the 15th square, isnt that a
kind of operation? Yes:
Besides these changes of symbols, the simple operations must include
changes of distribution of observed squares. The new observed squares
must be immediately recognisable by the computer [by the clerk]. (p. 250.)
And how does the clerk do that? Is finding the 15th square a simple operation?
Maybe. How about finding the 9999999999999999th square? No:
I think it is reasonable to suppose that they can only be squares whose
distance from the closest of the immediately previously observed squares
does not exceed a certain fixed amount. Let us say that each of the
new observed squares is within L squares of an immediately previously
observed square. (p. 250.)
8.7. SECTION 9 303
So here we have a fourth kind of boundedness or finiteness: The clerk can only look a
certain bounded distance away. How far can the distance be? Some plausible lengths
are the length of a typical word or small numeral (so L could equal B). The minimum
is, of course, 1 square (taking L = B = 1). So, another simple operation is looking
one square to the left or to the right (and, of course, the ability to repeat that operation,
so that the clerk can, eventually, find the 15th or the 9999999999999999th square).
What about a different kind of candidate for a simple operation: find a square
that contains the special symbol x:
In connection with immediate recognisability, it may be thought that
there are other kinds of square which are immediately recognisable.
In particular, squares marked by special symbols might be taken as
immediately recognisable. Now if these squares are marked only by single
symbols there can be only a finite number of them, and we should not upset
our theory by adjoining these marked squares to the observed squares.
(pp. 250252.)
So, Turing allows such an operation as being simple, because it doesnt violate the
finiteness limitations. But he doesnt have to allow them. How would the clerk be able
to find the only square that contains the special symbol x (assuming that there is one)?
By first observing the current square. If x isnt on that square, then observe the next
square to the left. If x isnt on that square, then observe the square to the right of the
first one (by observing the square two squares to the right of the current one). And so
on, moving back and forth, till a square with x is found. What if the clerk needs to find
a sequence of squares marked with a sequence of special symbols?
If, on the other hand, they [i.e., the squares marked by special symbols]
are marked by a sequence of symbols, we cannot regard the process of
recognition as a simple process. (p. 251.)
I wont follow Turings illustration of how this can be done. Suffice it to say that it is
similar to what I just sketched out as a way of avoiding having to include finding a
special square as a simple operation, and Turing admits as much:
If in spite of this it is still thought that there are other immediately
recognisable squares, it does not upset my contention so long as these
squares can be found by some process of which my type of machine is
capable. (p. 251.)
In other words, other apparently simple operations that can be analyzed into
some combination of the simplest operations of writing a symbol and observing are
acceptable. It is worth noting that this can be interpreted as a claim that subroutines
can be thought of as single operations.
Turing now summarizes his analysis of the minimum that a human computer (what
I have been calling a clerk) needs to be able to do in order to compute:
The simple operations must therefore include:
a Changes of the symbol on one of the observed squares.
b Changes of one of the squares observed to another square within L

squares of one of the previously observed squares.
It may be that some of these changes necessarily involve a change of state
of mind. The most general single operation must therefore be taken to be
one of the following:
A A possible change (a) of symbol together with a possible change of
state of mind.
B A possible change (b) of observed squares, together with a possible
change of state of mind. (p. 251.)
In other words, the two basic operations are to write a symbol on the tape (and to
change your state of mind) and to look somewhere else on the tape (and to change
your state of mind). Thats it: writing and looking! Well, and changing your state
of mind, which we havent yet clarified, but will next.
8.7.2.1.6 Conditions. How does the clerk know which of these two things (writing
or looking) to do? Turings next remark tells us:
The operation actually performed is determined, as has been suggested on
p. 250, by the state of mind of the computer [that is, of the clerk] and the
observed symbols. In particular, they determine the state of mind of the
computer [that is, of the clerk] after the operation is carried out. (p. 251,
my interpolation and italics.)
The passage on p. 250 that Turing is referring to is the one that I marked (*) and
called astounding, above; it says roughly the same thing as the present passage. So,
what Turing is saying here is that the clerk should
first consider his or her state of mind and where he or she is currently looking on
the paper,
then decide what to do next (either write something there or look somewhere
else), and, finally, change his or her state of mind.
Of course, after doing that, the clerk is in a (possibly) new state of mind and looking
at a (possibly) new location on the paper, which means that the clerk is ready to do the
next thing.
8.7.2.1.7 States of mind clarified. Now, think about a typical computer program,
especially an old-fashioned one, such as those written in (early versions of) Basic
or Fortran, where each line of the program has a line number and a statement to be
executed (a command). The computer (and here I mean the machine, not a clerk)
starts at the first line number, executes the command, and then (typically) moves to
the next line number; in atypical cases, the command might be a jump or go to
command, which causes the computer to move to a different line number. At whatever
line number the computer has moved to after executing the first command, it executes
8.7. SECTION 9 305
the command at that new line number. And so on. But, if you compare this description
with Turings, you will see that what corresponds to the line number of a program is
Turings notion of a state of mind! And what is the currently observed symbol? It is
the current input to the program!
So, lets paraphrase Turings description of the basic operation that a clerk performs
when computing. Well write the paraphrase in terms of a computer program that the
clerk is following:
The operation performed is determined by the current line number of the

program and the current input. The simple operations are: (a) print a
symbol and (b) move 1 square left or right on the tape (which is tantamount
to accepting new input), followed by changing to a new line number.11
We can also say this in a slightly different way:
If the current line number is N and the current input is I,

then print or move (or both) and go to line N 0 .
And a program for such a computer will consist of lines of code that look like this:
Line N: if input = I
then begin print (or move); go to Line N 0 end
else go to Line N 00
8.7.2.1.8 The Turing machine. I said above that passage (*) was astounding;
here is its sequel:
We may now construct a machine to do the work of this computer. (p. 251,
my italics.)
Reading this sentence out of context can make it sound very confusing; after all, isnt
a computer a machine? But, as we have seen, a computer (for Turing) is a human clerk
who computes. And what Turing is now saying is that the human can be replaced by a
machine, that is, by what we now call a computer (a mechanical device). This sentence
marks the end of Turings analysis of what a human computer does and the beginning
of his mathematical construction of a mechanical computer that can do what the human
does. His description of it here is very compact; it begins as follows:
11 Heres another way to think about this: Suppose that we have two tapes. (Incidentally, it can be
proved that any two-tape Turing machine is equivalent to a one-tape Turing machine; see [Dewdney, 1989,
Ch. 28].) Tape 1 will be the one we have been discussing so far, containing input (the symbols being
scanned) and output (the symbols being printed). Tape 2 will contain the computers program, with each
square representing a state of mind. The computer can be thought of as starting in a square on Tape 2,
executing the instruction in that square (by reading from, and writing to, a square on Tape 1 and then moving
to a(nother) square on Tape 1), and then moving to a(nother) square on Tape 2, and repeating this fetch-
execute loop. In reality, Turing machines only have one tape, and the instructions are not written anywaher;
rather, they are hardwired into the Turing machine. Any written verion of them is (merely) a description of
the Turing machines behavior (or of its wiring diagram). But, if we encode Tape 2 on a portion of Tape 1,
then we have a stored-programor universalcomputer. For a similar presentation of universal Turing
machines, see [Dewdney, 1989, Ch. 48].
To each state of mind of the computer [of the clerk!] corresponds an m-

configuration of the machine.
So, an m-configuration is something in the machine that corresponds to a line number of
a program.12 But, considering modern computers and programs, programs are separate
from the computers that run them, so what could Turing mean when he says that an
m-configuration belongs to a machine? He means that the machine is hardwired (as
we would now say) to execute exactly one program (exactly one algorithm). Being
hardwired, no separate program needs to be written out; it is already compiled into
the machine. A Turing machine can do one and only one thing; it can compute one
and only one function, using an algorithm that is hardwired into it.
Turing continues:
The machine scans B squares corresponding to the B squares observed by
the computer. (p. 251, my italics.)
A modern translation of this sentence would say: The computer scans B squares
corresponding to the B squares observed by the clerk. The clerk is limited to observing
a maximum of B squares on the tape, as we saw above (in an earlier quote from p. 250).
The machine analogue of that is to move, or scan, B squares to the left or right on the
tape. In modern mathematical treatments of Turing machines, B = 1.
Turing continues:
In any move the machine can change a symbol on a scanned square or can
change any one of the scanned squares to another square distant not more
than L squares from one of the other scanned squares. (pp. 251252.)
In other words, the machine (the Turing machine, or modern hardwired computer) can
pay attention to B squares at a time, and each line of its program allows it to print a
new symbol on any of those squares or move to any other square that is no more than L
squares away from any of the B squares. Again, modern treatments simplify this: The
machine is scanning a single square, and each line of its program allows it to print a
new symbol on that square or to move one square to its left or right (or both print and
move).
Which move should the machine make?
The move which is done, and the succeeding configuration, [that is,
the next m-configuration; that is, the next step in the algorithm], are
determined by the scanned symbol and the [current] m-configuration.
(p. 252.)
That is, the move that the machine should make, as well as the next m-configuation
(that is, the next step in the algorithm) are determined by the currently scanned symbol
and the current m-configuration. Or, put in terms of computer programs, the instruction
on the current line number together with the current input together determine what to
do now (print, move, or both) and what to do next (which instruction to carry out next).
12 Why m? It could stand for man, on the grounds that this is a machine analogue of a (hu)mans state
of mind; or it could stand for mental, on the grounds that it is an analogue of a state of mind. But I think
it most likely stands for machine, because it is a configuration, or state, of a machine. Of course, Turing
might have intended it to be ambiguous among all these options.
8.7.2.1.9 Turings thesis.

The machines just described do not differ very essentially from computing
machines as defined in 2, and corresponding to any machine of this type
a computing machine can be constructed to compute the same sequence,
that is to say the sequence computed by the computer. (p. 252.)
As for the first clause, please recall that we are in the middle of a very long digression
in which we have skipped ahead to Turings 9 from Turings 1; we have not yet read
Turings 2. When we do, we will see a more detailed version of the machines that
Turing has just described for us here in 9.
The next clause is a bit ambiguous. When Turing says any machine of this type,
is he referring to the machines of 9 or the machines of 2? It probably doesnt matter,
because he has said that the two kinds of machines do not differ very essentially from
each other. But I think that he is, in fact, referring to the machines of 9; computing
machines are the ones that are defined in 2.
The last phrase is of more significance: These (computing) machines (of Turings
2) compute the same sequence. . . computed by the clerk. In other words, whatever
a human clerk can do, these machines can also do. What a human clerk can do (i.e.,
which sequences, or functions, a human clerk can compute) is captured by the informal
notion of algorithm or computation. These machines are a formal counterpart, a
formal explication of that informal notion. So this last phrase is a statement of
Turings thesis (that is, Churchs thesis, or the Computation Thesis).
What about the other direction? Can a human clerk do everything that one of these
machines can do? Or are these machines in some way more powerful than humans? I
think the answer should be fairly obvious: Given the way the machines are constructed
on the basis of what it is that humans can do, surely a human could follow one of the
programs for these machines. So humans can do everything that one of the machines
can do, andby Turings thesisthese machines can do everything that humans can do
(well, everything that is computable in the informal sense). But these are contentious
matters, and we will return to them when we consider the controversies surrounding
hypercomputation (Ch. 11) and AI (Ch. 19).
For now, we have come to the end of our digression, and we now return to Turings
1, Computing Machines.
8.8 Section 1, Paragraph 2

This is the paragraph in which Turing gives a more detailed presentation of his abstract
computing machine, the outcome of his detailed analysis from 9 of human computing.
He begins as follows:
We may compare a man in the process of computing a real number
to a machine which is only capable of a finite number of conditions
q1 , q2 , . . . , qR which will be called m-configurations. (p. 231.)
Why may we do this? Turing will give his justification in his 9, which we have just
finished studying. By a man, Turing of course means a human, not merely a male
human. To compute a real number is to compute the output of a real-valued function.

And, as we have already seen, an m-configuration is a line of a computer program, that
is, a step in an algorithm. Here, Turing is saying that each such algorithm has a finite
number (namely, R) of steps, each labeled qi . Put otherwise, (human, or informal)
computation can be compared with (and, by Turings thesis, identified with) a finite
algorithm.
What else is needed?
The machine is supplied with a tape (the analogue of paper) running
through it, and divided into sections (called squares) each capable of
bearing a symbol. (p. 231.)
There are a couple of things to note here. First, from our study of Turings 9, we know
why this is the case and what, exactly, the tape, squares, and symbols are supposed to
be and why they are the way they are.
But, second, why does he put those three words in scare quotes? There are two
possible answers. I suspect that the real answer is that Turing hasnt, at this point in his
paper, explained in detail what they are; that comes later, in his 9.
But there is another possible reason, a mathematical reason: In Turings formal,
mathematical notion of a computing machine, the concepts of tape, squares of
a tape, and symbols are really undefined (or primitive) terms in exactly the same
way that point, line, and plane are undefined (or primitive) terms in Euclidean
plane geometry. As Hilbert famously observed, One must be able to say at all times
instead of points, lines, and planestables, chairs, and beer mugs.13 So, here, too, one
must be able to say at all timesinstead of tapes, squares, and symbolstables, chairs
(but Ill use place settings; it will make more sense, as you will see), and beer mugs. A
Turing machine, we might say, must have a table. Each table must have a sequence of
place settings associated with it (so we must be able to talk about the nth place setting
at a table). And each place setting can have a beer mug on it; there might be different
kinds of beer mugs, but they have to be able to be distinguished from each other, so that
we dont confuse them. In other words, it is the logical or mathematical structure of a
computing machine that matters, not what it is made of. So, a tape doesnt have to be
made of paper, a square doesnt have to be a regular quadrilateral that is physically
part of the tape (chairsor place settingsare presumably physically separate from
their associated table), and symbols only have to be capable of being borne by a
square (a numeral can be written on a square of the tape, a beer mug can be placed
at a place setting belonging to a table).
Turing continues:
At any moment there is just one square, say the r-th, bearing the symbol
S(r) which is in the machine. We may call this square the scanned
square. (p. 231.)
First, S is just the capital letter S in a font called German Fraktur, or black
letter. Its a bit hard to read, so I will replace it with S in what follows (even when
quoting Turing).
13 Cf. Hilberts Gesammelte Abhandlungen (Complete Works), vol. 3, p. 403, as cited in [Coffa, 1991,
135]; see also [Shapiro, 2009, 176].

Note, second, that this seems to be a slight simplification of his 9 analysis, with
B = 1. Second, being in the machine might be another undefined (or primitive)
term merely indicating a relationship between the machine and something else. But
what else? Turings punctuation allows for some ambiguity.14 It might be the symbol
(whatever it is, 0, 1, or a beer mug) that is in the machine, or it might be the scanned
square. I think that it is the latter, from remarks that he makes next.
The symbol on the scanned square may be called the scanned symbol.
The scanned symbol is the only one of which the machine is, so to speak,
directly aware. (p. 231.)
Here, Turings scare quotes around directly aware, together with the hedge so to
speak, clearly indicate that he is not intending to anthropomorphize his machine. His
machines are not really aware of anything; only humans can be really aware of
things. But the machine analogue of human awareness is: being a scanned symbol.
There is nothing anthropomorphic about that: Either a square is being scanned (perhaps
a light is shining on a particular place setting at the table) or it isnt, and either there
is a symbol on the scanned square (there is a beer mug at the lighted place setting), or
there isnt.
However, by altering its m-configuration the machine can effectively

remember some of the symbols which it has seen (scanned) previously.
(p. 231.)
What does this mean? Lets try to paraphrase it: By altering the line number of its
program, the computing machine can effectively. . . can effectively do what? It can
remember previously scanned symbols. This is to be contrasted with the currently
scanned symbol. How does the machine remember by altering a line number?
Well, how would it remember what symbol was on, say, the 3rd square if its now on
the 4th square? It would have to move left one square and scan the symbol thats there.
To do that, it would have to have an instruction to move left. And to do that, it would
need to go to that instruction, which is just another way of saying that it would have
to alter its m-configuration. (For another slow-reading analysis of this sentence, see
[Dresner, 2003, Dresner, 2012].)
The possible behaviour of the machine at any moment is determined by

the m-configuration qn and the scanned symbol S(r). (p. 231, my italics.)
It is only a possible behavior, because a given line of a program is only executed when
control has passed to that line. If it is not being executed at a given moment, then it
is only possible behavior, not actual behavior. The machines m-configuration is the
14 More precisely, his lack of punctuation: If which is preceded by a comma, then which is in the
machine is a non-restrictive relative clause that refers to the square. With no comma, the which clause
is a restrictive relative clause modifying symbol S(r). For more on relative clauses, see Which vs.
that , online at http://www.cse.buffalo.edu/rapaport/howtowrite.html#whichVthat
analogue of a line number of a program, and the scanned symbol is the analogue of the
external input to the machine.15
This pair qn , S(r) will be called the configuration: thus the configuration
determines the possible behaviour of the machine. (p. 231.)
Giving a single name (configuration) to the combination of the m-configuration and

the currently scanned symbol reinforces the idea that the m-configuration alone is an
analogue of a line number and that this combination is the condition (or antecedent)
of a condition-action (or a conditional) statement: Line qn begins, if the currently
scanned symbol is S(r), then. . . , or if the current instruction is the one on line qn
and if the currently scanned symbol is S(r), then. . . .
What follows the then? That is, what should the machine do if the condition is
satisfied?
In some of the configurations in which the scanned square is blank (i.e.

bears no symbol) the machine writes down a new symbol on the scanned
square: in other configurations it erases the scanned symbol. The machine
may also change the square which is being scanned, but only by shifting
it one place to right or left. In addition to any of these operations the
m-configuration may be changed. (p. 231, my boldface.)
So, we have 5 operations:
1. write a new symbol
2. erase the scanned symbol
3. shift 1 square left
4. shift 1 square right
5. change m-configuration.
The first 4 of these can be simplified to only 2 operations, each of which is slightly
more complex:
10 write a new symbol including [
20 shift (which is now an operation that takes an argument: left or right).
There are four things to note:
First, the symbols are left unspecified (which is why we can feel free to add a
blank symbol), though, as we have seen, they can be limited to just 0 and 1
(and maybe also [).
15 It is also possible, given what the reader knows at this stage of Turings paper, not yet having read his 9,
that an m-configuration is the entire internal state of the machine, perhaps encoding what could be called the
machines prior or background knowledgein contrast to external information from the outside world,
encoded in the scanned symbol.
Second, Turing has, again, simplified his 9 analysis, letting L = 1.
Third, change m-configuration is essentially a jump or go to instruction.

The whole point of structured programming, as we have seen, is that this can be
eliminatedso we really only need the first two of our slightly more complex
operations, as long as we require our programs to be structured.
Fourth, there is no halt command. (In 8.9.3, below, we will see why this is
not needed.)
Turing next clarifies what symbols are needed. Recall that the kind of computation
that Turing is interested in is the computation of the decimal of a real number.
Some of the symbols written down will form the sequence of figures which
is the decimal of the real number which is being computed. The others are
just rough notes to assist the memory. It will only be these rough notes
which will be liable to erasure. (pp. 231232.)
So, either we need symbols for the 10 Arabic numerals (if we write the real number in
decimal notation) or we only need symbols for the 2 binary numerals (if we write the
real number in binary notation). Any other symbols are merely used for bookkeeping,
and they (and only they) can be erased afterwards, leaving a clean tape with only the
answer on it.
There is one more thing to keep in mind: Every real number (in decimal notation)16
has an infinite sequence of digits to the right of the decimal point, even if it is an integer
or (a non-integer) rational number, which are typically written with either no digits or
a finite number of digits in the decimal expansion (1, 1.0, 2.5, etc.). If the number is
an integer, this is an infinite sequence of 0s; for example, 1 = 1.000000000000 . . .
(which I will abbreviate as 1.0). If the number is rational, this is an infinite sequence
of some repeating subsequence; for example:
1
= 0.500000000000 . . . = 0.50
2
1
= 0.333333333333 . . . = 0.3
3
1
= 0.142857142857 . . . = 0.142857
7
And if the number is irrational, this is an infinite, non-repeating sequence; for example:

2 = 1.41421356237309 . . .
= 3.1415926535 . . .
What this means is that one of Turings computing machines should never halt when
computing (i.e., writing out) the decimal of a real number. It should only halt if it is
writing down a finite sequence, and it can do this in two ways: It could write down the
16 Similar remarks can be made for binary notation.
finite sequence and then halt. Or it could write down the finite sequence and then go
into an infinite loop (either rewriting the last digit over and over in the same square, or
just looping in a do-nothing operation such as the empty program).
Finally,
It is my contention that these operations include all those which are used
in the computation of a number. (p. 232.)
This is another statement of Turings thesis: To compute, all you have to do is arrange
the operations of writing and shifting in a certain way. The way they are arranged
what is now called the control structure of a computer programis controlled
by the configuration and the change in m-configuration (or, in modern structured
programming, by Boehm and Jacopinis three control structures of sequence, selection,
and while-repetition). For Turing, it goes unsaid that all computation can be reduced
to the computation of a number; this is the insight we discussed in 7.6.1 that all
the information about any computable problem can be represented using only 0
and 1; hence, any informationincluding pictures and soundscan be represented
as a number. (But it is also important to realize that this kind of universal binary
representation of information doesnt have to be thought of as a number, because the
two symbols dont have to be 0 and 1!)
8.9 Section 2: Definitions

We are now ready to look at the section in which Turings computing machines are
defined, as we read in his 9 (see our 8.7.2.1.9, above).
8.9.1 Automatic Machines

Turing begins by giving us a sequence of definitions. The first is the most famous:
If at each stage the motion of a machine (in the sense of 1) is completely

determined by the configuration, we shall call the machine an automatic
machine (or a-machine). (p. 232.)
Clearly, such a machines motion (or behavior) is at least partly determined by its
configuration (that is, by its m-configuration, or line number, together with its currently
scanned symbol). Might it be determined by anything else? For all that Turing has
said so far, maybe such a machines human operator could help it along by moving
the tape for it, or by writing something on the tape. This definition rules that out by
limiting our consideration to such machines whose motion is completely determined
by the configuration. So, a human operator is not allowed to help it in any way: No
cheating allowed! Turing may have called such a machine an a-machine; we now call
themin his honorTuring machines.17 seems to have first been used
What about machines that get outside help?
17 Alonzo Church seems to have been the first person to use this term, in his review of Turings paper
[Church, 1937].
8.9. SECTION 2: DEFINITIONS 313
For some purposes we might use machines (choice machines or c-

machines) whose motion is only partially determined by the configuration
(hence the use of the word possible in 1). When such a machine reaches
one of these ambiguous configurations, it cannot go on until some arbitrary
choice has been made by an external operator. (p. 232.)
First, note that Turings explanation of the use of possible may be slightly different
from mine. But I think that they are consistent explanations. In the previous statements,
Turing used possible to limit the kind of operations that a Turing machine could
perform. Here, he is introducing a kind of machine that has another kind of possible
operation: writing, moving, or changing m-configuration not as the result of an explicit
instruction but as the result of a choice. . . made by an external operator. Note, by
the way, that this external operator doesnt have to be a human; it could be another
Turing machine! In any case, such c-machines were the topic of Turings doctoral
dissertation under Church, in which he called them oracle machines, or o-machines.
Such machines are now used to model interactive computing, and we will return to
them in Ch. 11 on hypercomputation.
8.9.2 Computing Machines

Turing gives us some more definitions:
If an a-machine prints two kinds of symbols, of which the first kind (called
figures) consists entirely of 0 and 1 (the others being called symbols of
the second kind), then the machine will be called a computing machine.
(p. 232.)
The principal definition here is that of computing machine, a special case of an a-
(or Turing) machine that outputs its results as a binary numeral (in accordance with
the first insight we discussed, in 7.6.1). Once again, here Turing is simplifying his 9
analysis of human computation, restricting the symbols to 0 and 1. Well, not quite,
because he also allows symbols of the second kind, used for bookkeeping purposes
or intermediate computations. Note, however, that any symbol of the second kind could
be replacedat the computational cost of more processingby sequences of 0s and
1s.
Turing continues:
If the machine is supplied with a blank tape and set in motion, starting from
the correct initial m-configuration, the subsequence of the symbols printed
by it which are of the first kind will be called the sequence computed by
the machine. (p. 232.)
Here, he seems to be allowing for some of the symbols of the second kind to remain on
the tape, so that only a subsequence of the printed output constitutes the result of the
computation. In other words, these secondary symbols need not be erased. One way to
think of this is to compare it to the way we write decimal numerals greater than 999,
namely, with the punctuation aid of the non-numerical symbols known as a comma
and a decimal point: 1,234,567.89
I almost wrote, to remain on the tape after the computation halts. But does it
halt? It cantbecause every real number has an infinite decimal part! The secondary
symbols could still be erased, during the computation; thats not of great significance
(obviously, its easier to not erase them and to just ignore them). The important point to
remember is that computations of decimal representations of real numbers never halt.
Well return to this in a moment.
One more small point that simplifies matters:
The real number whose expression as a binary decimal is obtained by

prefacing this sequence by a decimal point is called the number computed
by the machine. (p. 232.)
What about the part of the expression that is to the left of the decimal point? It looks
as if the only numbers that Turing is interested in computing are the reals between 0
and 1 (presumably including 0, but excluding 1). Does this matter? Not really; first,
all reals can be mapped to this interval, and, second, any other real can be computed
simply by computing its non-decimal part in the same way. Restricting our attention
to this subset of the reals simplifies the discussion without loss of generality.
A couple more definitions:
At any stage of the motion of the machine, the number of the scanned
square, the complete sequence of all symbols on the tape, and the m-
configuration will be said to describe the complete configuration at that
stage. The changes of the machine and tape between successive complete
configurations will be called the moves of the machine. (p. 232.)
Three points to note: First, at any stage of the motion of the machine, only a finite
number of symbols will have been printed, so it is perfectly legitimate to speak of
the complete sequence of all symbols on the tape even though every real number
has an infinite number of numerals after the decimal point. Second, the sequence
of all symbols on the tape probably includes all occurrences of [ that do not occur
after the last non-blank square (that is, that do occur before the last non-blank square);
otherwise, there would be no way to distinguish the sequence h0, 0, 1, [, 0i from the
sequence h[, 0, [, 0, [, 1, 0i.
Third, we now have three notions called configurations; lets summarize them for
convenience:
1. m-configuration = line number, qn , of a program for a Turing machine.
2. configuration = the pair: hqn , S(r)i,

where S(r) is the symbol on the currently scanned square, r.
3. complete configuration = the triple:

hr, the sequence of all symbols on the tape,18 qn i.
18 As described in the last paragraph.

8.9. SECTION 2: DEFINITIONS 315
8.9.3 Circular and Circle-Free Machines

We now come to what I have found to be one of the most puzzling sections of Turings
paper. It begins with the following definitions:
If a computing machine never writes down more than a finite number of

symbols of the first kind, it will be called circular. Otherwise it is said to
be circle-free. (p. 233.)
Lets take this slowly: A computing machine is a Turing machine that only prints a
binary representation of a real number together with a few symbols of the second kind.
If such a machine never writes down more than a finite number of 0s and 1s, then,
trivially, it has only written down a finite number of such symbols. That means that is
has halted! And, in that case, Turing wants to call it circular! But, to my ears, at least,
circular sounds like looping, which, in turn, sounds like it means not halting.
And, if it does write down more than a finite number of 0s and 1s, then, trivially,
it writes down an infinite number of them. That means that it does not halt! In that
case, Turing wants to call it circle-free! But that sounds like loop-free, which, in
turn, sounds like it means that it does halt.
Whats going on? Before looking ahead to see if, or how, Turing clarifies this,
heres one guess: The only way that a Turing machine can print a finite number of
figures (Turings name for 0 and 1) and still be circular (which I am interpreting
to mean loop) is for it to keep repeating printingthat is, to overprintsome or
all of them, that is, for it to circle back and print some of them over and over again.
(In this case, no halt instruction is needed!)
And the only way that a Turing machine can print an infinite number of figures
and also be circle-free is for it to continually print new figures to the right of
the previous one that it printed (and, thus, not circle back to a previous square,
overprinting it with the same symbol thats on it).
Is that what Turing has in mind? Lets see. The next paragraph says:
A machine will be circular if it reaches a configuration from which there

is no possible move or if it goes on moving, and possibly printing symbols
of the second kind, but cannot print any more symbols of the first kind.
The significance of the term circular will be explained in 8. (p. 233.)
The first sentence is rather long; lets take it phrase by phrase: A machine will be
circularthat is, will print out only a finite number of figuresif [Case 1] it reaches
a configuration from which there is no possible move. That is, it will be circular if
it reaches a line number qn and a currently scanned symbol S(r) from which there is
no possible move. How could that be? Easy: if theres no line of the program of the
form: Line qn : If currently scanned symbol = S(r) then. . . . In that case, the machine
stops,19 because theres no instruction telling it to do anything.20
19 At this point, I cannot resist recommending that you read E.M. Forsters wonderfully prescient, 1909(!)
short story, The Machine Stops, [Forster, 1909].

20 Another possibility is that line q says: If currently scanned symbol = S(r), then go to line q . In that
n n
case, the machine never stops, becuase it forever loops (circles?) back to the same line.
Thats even more paradoxical than my interpretation above; here, he is clearly

saying that a machine is circular if it halts! Of course, if you are the operator of a Turing
machine and you are only looking at the tape (and not at the machinery), would you be
able to tell the difference between a machine that was printing the same symbol over
and over again on the same square and a machine that was doing nothing?21 Probably
not. So, from an external, behavioral point of view, these would seem to amount to the
same thing.
But Turing goes on: A machine will also be circular if [Case 2] it goes on moving,
and possibly printing [only] symbols of the second kind but not printing any more
figures. Here, the crucial point is that the machine does not halt but goes on moving.
It might or might not print anything, but, if it does, it only prints secondary symbols.
So we have the following possibilities: a machine that keeps on moving, spewing out
square after square of blank tape; or a machine that keeps on moving, occasionally
printing a secondary symbol. In either case, it has only printed a finite number of
figures. Because it has, therefore, not printed an infinite decimal representation of a
real number, it has, for all practical purposes, haltedat least in the sense that it has
finished its task, though it has not succeeded in computing a real number.
Once again, a machine is circular if it halts (for all practical purposes; its still
working, but just not doing anything significant). This isnt what I had in mind in my
interpretation above. But it does seem to be quite clear, no matter how you interpret
what Turing says, that he means that a circular machine is one that does not compute
a real number, either by halting or by continuing on but doing nothing userful (not
computing a real number). Machines that do compute real numbers are circle-free,
but they must also never halt; they must loop forever, in modern terms, but continually
doing useful work (computing digits of the decimal expansion of a real number).22
In other words, a good Turing machine is a circle-free one that does not halt
and that continually computes a real number. This seems to be contrary to modern
terminology and the standard analysis of algorithms that we saw in 7.5. And how
does this fit in with Turings claim at the beginning of his paper that the computable
numbers may be described briefly as the real numbers whose expressions as a decimal
are calculable by finite means (my italics)? The only way that I can see to make these
two claims consistent is to interpret by finite means to refer to the number of steps
in an algorithm, or the amount of time needed to carry one step out, or the number
of operations needed to carry one step out (in case any of the steps are not just basic
operations). It cannot mean, as we have just seen, that the entire task can be completed
in a finite amount of time or that it would necessarily halt.
Finally, what about the allusion to Turings 8? That section, which we will not
investigate and which is uninformatively titled Application of the Diagonal Process,
is the section in which he proves that the Halting Problem is not computable (more
precisely, that a Godel-like number of a program for the Halting Problem is not a
computable number). And, pretty obviously, his proof is going to be a little bit
different from the one that we sketched in 7.8 because of the difference between
21 The machine described in the text and the machine described in the previous footnote have this property.
22 A machine that computes a real number in this sense was called circle-free; one that does not (because
it never prints more than a finite number of 0s and 1s) was called circular [Davis, 1995b, 141].
8.10. SECTION 3: EXAMPLES OF COMPUTING MACHINES 317
our modern idea that only Turing machines that halt are good and Turings idea
that only Turing machines that are circle-free are good. The interested reader
is referred to [Petzold, 2008, Ch. 10], who notes, by the way, that the concept of
halting was introduced into the modern literature by Martin Davis [Petzold, 2008,
179], despite the fact that Turings original machines never halt! [Petzold, 2008,
329]. [van Leeuwen and Wiedermann, 2013] analyzes these notions in modern terms.
8.9.4 Section 2,
Subsection Computable Sequences and Numbers
Here are Turings final definitions from this section:
A sequence is said to be computable if it can be computed by a circle-free

machine. (p. 233, my italics.)
Although this is presented as a definition of computable (actually, of computable

sequence), it can, in fact, be understood as another statement of Turings thesis. Being
computable by a circle-free machine is a very precise, mathematical concept. In this
definition, I think that Turing is best understood as suggesting that this precise concept
should replace the informal notion of being computable. Alternatively, Turing is
saying here that he will use the word computable in this very precise way.
A number is computable if it differs by an integer from the number

computed by a circle-free machine. (p. 233, my italics.)
Circle-free machines compute (by printing out) a sequence of figures (a sequence of

0s and 1s). Such a sequence can be considered to be a decimal (actually, a binary)
representation of a number between 0 and 1 (including 0, but not including 1). Here,
Turing is saying that any real number can be said to be computable if it has the same
decimal part (that is, the same part after the decimal point) of a number representable
as a computable sequence of figures. (So, for instance, = 3.1415926535 . . . differs
by the integer 3 from the number 0.1415926535. . . computable by a circle-free Turing
machine.)
8.10 Section 3: Examples of Computing Machines

We are now ready to look at some real Turing machines, more precisely, computing
machines, which, recall, are automatic a-machines that print only figures (0, 1)
and maybe symbols of the second kind. Hence, they compute real numbers. Turing
gives us two examples, which we will look at in detail.
8.10.1 Example I
A machine can be constructed to compute the sequence 010101. . . .
(p. 233.)
Actually, as we will see, it prints
0[1[0[1[0[1[ . . .
What real number is this? First, note that it is a rational number of the form 0.01.
Treated as being written in binary notation, it = 31 ; treated as being written in decimal
1
notation, it = 99 .
The machine is to have the four m-configurations b, c, f, e and is

capable of printing 0 and 1. (p. 233.)
The four line numbers are (in more legible italic font): b, c, f , e.
The behaviour of the machine is described in the following table in which

R means the machine moves so that it scans the square immediately
on the right of the one it was scanning previously. Similarly for L.
E means the scanned symbol is erased and P stands for prints.
(p. 233.)
This is clear enough. It is, however, interesting to note that it is the Turing machine
that moves, not the tape!
Before going on with this paragraph, lets look at the table. In later writings by
others, such tables are sometimes called machine tables; they are computer programs
for Turing machines, written in a Turing-machine programming language that Turing
is here giving us the syntax and semantics for.
However, it is important to keep in mind that the Turing machine does not consult
this table to decide what to do. We humans would consult it in order to simulate
the Turing machines behavior. But the Turing machine itself simply behaves in
accordance with that table, not by following it. The table should be thought of as a
mathematical-English description of the way that the Turing machine is hardwired
to behave. (Well revisit this idea in 10.5.1.)
Heres the table, written a bit more legibly than in Turings paper:
Configuration Behaviour
m-config. symbol operations final m-config.
b None P0, R c
c None R e
e None P1, R f
f None R b
This program consists of 4 lines. It is important to note that it is a set of lines, not a
sequence: The order in which the lines are written down in the table (or program) is
irrelevant; there will never be any ambiguity as to which line is to be executed. Perhaps
a better way of saying this is: There will never be any ambiguity as to which line is
causing the Turing machine to move.
Each line consists of two principal parts: a configuration and a behavior. Each
configuration, as you may recall, consists of two parts: an m-configuration (or line
number) and a symbol (namely, the currently scanned symbol). Each behavior consists
also of two parts: an operation (one or more of E, L, R, or P) and a final m-
configuration (that is, the next line number to be executed).
This table (and all succeeding tables of the same kind) is to be understood
to mean that for a configuration described in the first two columns the
operations in the third column are carried out successively, and the
machine then goes over into the m-configuration described in the last
column. (p. 233, my boldface.)
That is, each line of the program should be understood as follows: Under the
circumstances described by the configuration, do the operation and then go to the
instruction at the final m-configuration. Or, to use Turings other terminology: If
your current state of mind is the one listed in the current m-configuration, and if the
symbol on the current square being scanned is the one in the symbol column, then do
the operation and change your state of mind to the one in the final m-configuration
column.
A modern way of thinking about this is to consider it to be a production system.
Production systems are an architecture introduced by Emil Post in his analysis of
computation and used by many researchers in AI [Post, 1941]. A production system
consists of a set of condition-action pairs; if the condition is satisfied, then the
action is carried out. Thats exactly what we have in this Turing-machine table: The
configuration is the condition, and the behavior is the action.23
A further qualification:
When the second column [that is, the symbol column] is left blank, it is
understood that the behaviour of the third and fourth columns applies for
any symbol and for no symbol. (p. 233.)
That is the situation we have in this first example, where None is the entry in each row
of the symbol column. So the only condition determining the behavior of this Turing
machine is its current state of mind, that is, its current line number.
Finally, we need to know what the initial situation that this production system is
intended to apply to:
The machine starts in the m-configuration b with a blank tape. (p. 233.)
Perhaps b stands for begin, with subsequent states of mind (in alphabetical as
well as sequential order) being c, e, and f ( f for final? What happened to d?).
Lets trace this program. We start with a blank tape, which I will show as follows:
[[[[[[[[[[[ . . .
We are in state b.
Looking at the table, we see that if we are in state b, then (because any symbol that
might be on the tape is irrelevant), we should do the sequence of operations P0, R.
23 [Sieg, 2000, 7] notes that even Turing considered Turing machines as production systems.
Turing hasnt told us what P0 means, but, because P means print, its pretty
obvious that this means print 0 on the currently scanned square.
Note, too, that he hasnt told us which square is currently being scanned! It
probably doesnt matter,24 but its worth thinking about some of the options. One
possibility is that we are scanning the first, or leftmost, square; this is the most
likely option and the one that I will assume in what follows. But another possibility
is that we are scanning some other square somewhere in the middle of the tape.
That probably doesnt matter, because Turing only told us that it would compute the
sequence 010101. . . ; he didnt say where it would be computed! There is one further
possbility, not very different from the prvious one: The tape might not have a first
squareit might be infinite in both directions! And now we need to consider something
that Turing hasnt mentioned: How long is the tape? For all that he has told us, it
could be infinitely long. But many mathematicians and philosophers (not to mention
engineers!) are not overly fond of actual infinities. So, we can say, instead, that, at
any moment, the tape only has a finite number of squares, but there is no limit to the
number of extra squares that we are allowed to add on at one (or maybe both) ends.
(As my former teacher and colleague John Case used to put it, if we run out of squares,
we can always go to an office-supply store, buy some extra squares, and staple them
onto our tape!)
So, lets now show our intial tape as follows, where the currently scanned square is
underlined:
[ [ [ [ [ [ [ [ [ [ [ ...
Performing the two operations on line b converts our initial tape to this one:
0 [ [ [ [ [ [ [ [ [ [ ...
and puts us in state c. That is, we next execute the instruction on line c.
Looking at line c, we see that, no matter what symbol is on the current square (it is,
in fact, blank), we should simply move right one more square and change our mind to
e. So now our tape will look like this:
0 [ [ [ [ [ [ [ [ [ [ ...
Because we are now in state e, we look at line e of the program, which tells us that,
no matter what, if anything, is on the current square, print 1 there, move right again,
and go into state f . So, our tape becomes:
0 [ 1 [ [ [ [ [ [ [ [ ...
Now we are in state f , so, looking at line f , we see that we merely move right once
again, yielding:
0 [ 1 [ [ [ [ [ [ [ [ ...
24 It probably doesnt matter, because all squares on the tape are blank. If the tape is infinite (or endless)
in both directions, then each square is indistinguishable from any other square, at least until something is
printed on one square.
And we go back into state b. But that starts this cycle all over again; we are indeed in
an infinite loop! One more cycle through this turns our tape into:
0 [ 1 [ 0 [ 1 [ [ [ [ [ ...
Clearly, repeated cycles through this infinitely looping program will yield a tape
consisting entirely of the infinite sequence 010101. . . with blank squares separating
each square with a symbol printed on it:
0 [ 1 [ 0 [ 1 [ 0 [ 1 [ ...
Can this program be written differently?

If (contrary to the description in 1) we allow the letters L, R to appear
more than once in the operations column we can simplify the table
considerably. (p. 234.)
In the description in 1 (p. 231), Turing allowed the machine to change the square
which is being scanned, but only by shifting it one place to right or left (my italics).
Now, he is allowing the machine to move more than one place to the right or left; this
is accomplished by allowing a sequence of moves. Here is the modified program:
m-config. symbol operations final m-config.

None P0 b
b 0 R, R, P1 b
1 R, R, P0 b
Note that there is only one m-configuration, that is, only one line number; another
way to think about this is that this program has only one instruction. Turing would say
that this machine never changes its state of mind. But that one instruction is, of course,
more complex than the previous ones. This one is what would now be called a case
statement: In case there is no current symbol, print 0; in case the current symbol =
0, move right 2 squares and print 1; and in case the current symbol = 1, move right 2
squares and print 0.
I urge you to try to follow this version of the program, both for practice in reading
such programs and to convince yourself that it has the same behavior as the first one.
(Another interesting exercise is to write a program for a Turing machine that will print
the sequence 010101. . . without intervening blank squares.)
So, our machine has compute[d] the sequence 010101. . . . Or has it? It has
certainly written that sequence down. Is that the same thing as computing it?
And here is another question: Earlier, I said that 010101. . . was the binary
representation of 13 and the decimal representation of 99 1
. Have we just computed 13
1
in base 2? Or 99 in base 10?25 Even if you are inclined to answer yes to the question
in the previous paragraph, you might be more inclined to answer no to the question
in this one. Although Turing may have a convincing reason (in his 9) to say that
computing consists of nothing more than writing down symbols, surely there has to be
25 For an interesting discussion of this, see [Rescorla, 2013]. Well return to some of the issues discussed
in Rescorlas paper in Ch. 10.

more to it than that; surely, just writing down symbols is only part of computing. The
other parts have to do with which symbols get written down, in what order, and for
1
what reason. If I asked you to compute the decimal representation of 99 , how would
you know that you were supposed to write down 010101. . . ? Surely, that is the heart
of computation. Or is it?
At this point, however, we should give Turing the benefit of the doubt. After all, he
1
did not say that we were going to compute 99 , only that we were going to compute
010101. . . , and, after all, computing that sequence really just is writing it down;
its a trivial, or basic, or elementary, or primitive computation (choose your favorite
adjective). Moreover, arguably, Turing showed us this trivial example so that we could
clearly see the format of his Turing-machine programs before getting a more complex
example.
Before turning to such a more complex program, lets consider the syntax of these
programs a bit more. NEED TO EXPLAIN SYNTAX SOMEWHERE. Each line of
the program has the following, general form:
qB SI O M qE
where:
1. qB is an initial (or beginning) m-configuration (a line number)
2. SI is either an input that the machine reads (that is, the symbol on the currently
scanned square), or else it is empty
3. O is an operation (or a sequence of operations) to be performed (where the

operations are Px, E, L, R, and where x is any legally allowed symbol),26 and
4. qE is a final (or ending) m-configuration.
And the semantics (or interpretation) of this program line is:
if the Turing machine is in m-configuration qB

and if either the current input = SI or no input is specified
then begin
do the sequence of operations O;
{where each operation is either:

print x on the current square,
where x is a legally allowed symbol,
or erase the symbol that is on the current square,
where printing x overwrites whatever is currently printed
on the square,
and where erasing results in a blank square
(even if the square is already blank)}
26 In our first program, the only symbols were 0 and 1; we will see others in subsequent examples.
do the sequence of moves M;
{where each move is either:

move left one square,
or move right one square}
go to m-configuration qE
end
else
{if the machine is not in m-configuration qI

or the current symbol 6= SI and is not blank, then:}
do nothing (that is, halt)
8.10.2 Example II
We now come to a slightly more difficult example:
As a slightly more difficult example we can construct a machine to
compute the sequence 001011011101111011111. . . . (p. 234.)
First, note that the sequence to be computed consists of the subsequences
0, 1, 11, 111, 1111, 11111, . . .
That is, it is a sequence beginning with 0, followed by the numbers 1, 2, 3, 4, 5,. . .
written in base 1 (that is, as tally strokes)with each term separated by a 0.
But this seems very disappointing! It seems that this more difficult computation
is still just writing down some symbols without computing anything. Perhaps. But
note that what is being written down (or computed) here are the natural numbers.
This program will begin counting, starting with 0, then the successor of 0, the successor
of that, and so on. But, as we saw in 7.7.2, the successor function is one of the basic
recursive functions, that is, one of the basic computable functions.
Being able to (merely!) write down the successor of any number, being able to
(merely!) write down the predecessor of any non-0 number, and being able to find
a given term in a sequence of numbers are the only basic recursive (or computable)
functions. Turings slightly more difficult example will show us how to compute the
first of these. Devising a Turing machine program for computing the predecessor of the
natural number n should simply require us to take a numeral represented as a sequence
of n occurrences of 1 and erase the last one. Devising a Turing machine program for
computing the jth term in a sequence of k symbols should simply require us to move a
certain number of squares in some direction to find the term (or, say, the first square of
a sequence of squares that represents the term, if the term is complex enough to have
to be represented by a sequence of squares).
And any other other recursive function can be constructed from these basic
functions by generalized composition (sequencing), conditional definition (selection),
and while-recursion (repetition), which are just control structures for how to find a
path (so to speak) through a Turing-machine programthat is, ways to organize the
sequence of m-configurations that the Turing machine should go through.
So, it looks as if computation really is nothing more than writing things down,
moving around (on a tape), and doing so in an order that will produce a desired
result!
Lets now look at this slightly more difficult program:
The machine is to be capable of five m-configurations, viz., o, q, p,

f, b and of printing @, x, 0, 1. (p. 234.)
Ill simplify these m-configurations to the more legible o, q, p, f , b. The first

two symbols are going to be used only for bookkeeping purposes.27 So, once again,
Turing is really restricting himself to binary notation for the important information.
Continuing:
The first three symbols on the tape will be @@0; the other figures follow
on alternate squares. (p. 234, my italics.)
It may sound as if Turing is saying that the tape comes with some pre-printed
information. But, when we see the program, we will see that, in fact, the first instruction
has us print @@0 on the first three squares before beginning the real computation.
Had the tape come with pre-printed information, perhaps it could have been considered
as innate knowledge, though a less cognitive description could simply have been that
the manufacturer of the tape had simplified our life knowing that the first thing that
the program does to a completely blank tape is to print @@0 on the first three squares
before beginning the real computation. Because that only has to be done once, it
might have been simpler to consider it as pre-printed on the tape.
Note that Turing calls these symbols in the first clause, and then talks about
figures in the second clause. Figures, you may recall from 8.9.2, are the symbols
0 and 1. So, Turing seems to be saying that all subsequent occurrences of 0 and
1 will occur on alternate squares. What happens on the other squares? He tells us:
On the intermediate squares we never print anything but x. These letters

serve to keep the place for us and are erased when we have finished with
them. We also arrange that in the sequence of figures on alternate squares
there shall be no blanks. (p. 234.)
So, it sounds as if the final tape will begin with @@0; during the computation,
subsequent squares will have 0 or 1 interspersed with x; and at the end of the
computatation, those subsequent squares will only have 0 or 1, and no blanks. Of
course, at the end, we could go back and erase the initial occurrences of @, so that
there would only be figures and no other symbols.
27 The inverted e is called a schwa; it is used in phonetics to represent the sound uh, as in but.
Turing uses it merely as a bookkeeping symbol with no meaning.

Here is the program:

Configuraton Behaviour
m-config. symbol operations final
m-config.
b P@, R, P@, R, P0, R, R, P0, L, L o
1 R, Px, L, L, L o
o
0 q
Any (0 or 1) R, R q
q
None P1, L p
x E, R q
p @ R f
None L, L p
Any R, R f
f
None P0, L, L o
I think it will be helpful to restate this program in a more readable format:
b begin
print @@0 on the first 3 squares;
P0 on the 5th square;
move back to the 3rd square (which has 0 on it)
go to line o
end
o if current symbol = 1
then begin move right; Px; move back 3 squares; stay in o end
else if current symbol = 0
then go to q
q if current symbol = 0 or current symbol = 1
then begin move right 2 squares; stay in q end
else if current square is blank
then begin P1; move left; go to p end
p if current symbol = x
then begin erase the x; move right; go to q end
else if current symbol = @
then begin move right; go to f end
else if current square is blank
then begin move left 2 squares; stay in p end
f if current square is not blank
then begin move right 2 squares; stay in f end
else begin P0; move left 2 squares; go to o end
Note that no line of the program ends with the machine changing its state of mind
to m-configuration b. So that line of the program, which is the one that initializes the
tape with @@0 on the first 3 squares, is only executed once. Note also that whenever
an instruction ends with an command to go to that very same line, we are in a loop; a
structured version of the program would use a while. . . do control structure.
There are some odd things to consider in lines o, q, p: What happens if the machine
is in state o but the current symbol is not a figure? What happens in state q if the
current symbol is @ or x? And what happens in state p if the current symbol is a
figure? Turing doesnt specify what should happen in these cases. One possibility
is that he has already determined that none of these cases could occur. Still, modern
software engineering practice would recommend that an error message be printed out
in those cases. In general, in a computer program, when a situation occurs for which the
program does not specify what should happen, anything is legally allowed to happen,
and there is no way to predict what will happen; this is sometimes expressed in the
slogan, garbage in, garbage out.
Turing goes on to illustrate the working of this machine with a table. . . of the
first few complete configurations (p. 234.) Recall that a complete configuration
consists of information about which square is currently being scanned, the sequence
of all symbols on the tape, and the line number of the instruction currently being
executed. Rather than use Turings format, I will continue to use the format that I
used for Example I, adding the line number at the beginning, using underscoring to
indicate the currently scanned square, and assuming that any squares not shown are
blank; any blank square that is between two non-blank squares (if there are any) will
be indicated by our symbol for a blank that has been made visible: [. You are urged to
compare my trace of this program with Turings.
So, we begin with a blank tape. What is the machines initial state of mind, its
initial m-configuration? Turing has forgotten to tell us! But it is fairly obvious that b is
the initial m-configuration, and, presumably, we are scanning the leftmost square (or, if
the tape is infinite in both directions, then we are scanning any arbitrary square), and,
of course, all squares are blank:
b : [, [, . . .
The initial instruction tells us to print @, move right, print another @, move right again,
print 0, move right 2 more squares, print another 0, move 2 squares back to the left,
and go into state o. After doing this sequence of primitive operations, our complete
configuration looks like this:
o : @, @, 0, [, 0, [, . . .
(Just to help you in reading Turings paper, my notation for the initial situation
should be compared with his. Here is his:
:
b
He has a blank (literally), followed by a colon, with the m-configuration b underneath
the blank, marking the currently scanned square.
Instead, I have b: preceding a sequence of (visible) blanks, the first one of which
is marked as being the scanned square.
Turing then shows the second configuration:
@ @ 0 0 :
o
Turing has two occurrences of @ followed by two 0s that are separated by an

(invisible) blank, with the m-configuration o underneath the currently scanned square
(which contains the first 0), followed by a colon to mark the end of this complete
configuration.
I have o: preceding a sequence consisting of the two occurrences of @, followed
by a 0 that is marked as being the scanned square, followed by a (visible) blank,
followed by the second 0.)
We are now in m-configuration o, and the currently scanned square contains 0,
so the second case (that is, the bottom row) of this second instruction tells us merely
to go into state q. (The operations column is left empty, so there is no operation
to perform. It is worth noting that, although there does not always have to be an
operation to perform, there does always have to be a final state to go into, that is,
a next instruction to perform.) So, the tape looks exactly as it did before, except that
the machine is now in state q:
q : @, @, 0, [, 0, [, . . .
Because the machine is now in state q and still scanning a 0, the first case (that
is, the top row) of this third instruction tells us to move two squares to the right but to
stay in state q. So the tape now looks like this:
q : @, @, 0, [, 0, [, . . .
Because the machine is still in state q and still scanning a 0 (although the currently
scanned square is different), we perform the same (third) instruction, moving two more
squares to the right and staying in state q:
q : @, @, 0, [, 0, [, [, . . .
The machine is still in state q, but now there is no scanned symbol, so the second
case (bottom line) of the third instruction is executed, resulting in a 1 being printed
on the current square, and the machine moves left, going into state p.
Whenever the machine is in state p and scanning a blank (as it is now), the third
case (last line) of the fourth instruction is executed, so the machine moves two squares
to the left and stays in state p:
p : @, @, 0, [, 0, [, 1, . . .
Now the machine is in state p scanning a blank, so the same instruction is executed:
It moves two more squares to the left and continues in state p:
p : @, @, 0, [, 0, [, 1, . . .
But now it is the second case (middle line) of the fourth instruction that is executed,
so the machine moves right and goes into state f :
f : @, @, 0, [, 0, [, 1, . . .
When in state f scanning any symbol (but not a blank), the machine moves two
squares to the right, staying in f :
f : @, @, 0, [, 0, [, 1, . . .
Again, it moves two squares to the right, staying in f :
f : @, @, 0, [, 0, [, 1, . . .
And again:
f : @, @, 0, [, 0, [, 1, [, [, . . .
But now it executes the second case of the last instruction, printing 0, moving two
squares to the left, and returning to state o:
o : @, @, 0, [, 0, [, 1, [, 0, . . .
Now, for the first time, the machine executes the first case of the second instruction,
moving right, printing x, moving three squares to the left, but staying in o:
o : @, @, 0, [, 0, [, 1, x, 0, . . .
At this point, you will be forgiven if you have gotten lost in the woods, having
paid attention only to the individual trees and not seeing the bigger picture.28 Recall
that we are trying to count: to produce the sequence 0, 1, 11, 111,. . . with 0s between
each term:
0 0 1 0 11 0 111 0 . . .
We started with a blank tape:

[[[ . . .
and we now have a tape that looks like this:
@@0[0[1x0[ . . .
Clearly, we are going to have to continue tracing the program before we can see the
pattern that we are expecting; Turing, however, ends his tracing at this point. But we
shall continue; however, I will only show the complete configurations without spelling
out the instructions (doing that is left to the reader). Here goes, continuing from where
28 My apologies for the mixed metaphor.
we left off:
o: @ @ 0 [ 0 [ 1 x 0 ...
q: @ @ 0 [ 0 [ 1 x 0 ...
q: @ @ 0 [ 0 [ 1 x 0 ...
q: @ @ 0 [ 0 [ 1 x 0 ...
q: @ @ 0 [ 0 [ 1 x 0 [ [ ...
p: @ @ 0 [ 0 [ 1 x 0 [ 1 ...
p: @ @ 0 [ 0 [ 1 x 0 [ 1 ...
q: @ @ 0 [ 0 [ 1 [ 0 [ 1 ...
q: @ @ 0 [ 0 [ 1 [ 0 [ 1 ...
q: @ @ 0 [ 0 [ 1 [ 0 [ 1 [ [ [...
p: @ @ 0 [ 0 [ 1 [ 0 [ 1 [ 1 [...
Hopefully, now you can see the desired pattern beginning to emerge. The
occurrences of x get erased, and whats left is the desired sequence, but with blank
squares between each term and with two leading occurrences of @. You can see from
the program that there is no instruction that will erase those @s; the only instructions
that pay any attention to a @ are (1) the second case of m-configuration p, which only
tells the machine to move right and to go into state f , and (2) the first case of m-
configuration f , which, when scanning any symbol, simply moves two squares to the
right (but, in fact, that configuration will never occur!).
Turing goes on to make some remarks about various notation conventions that he
has adopted, but we will ignore these, because we are almost finished with our slow
reading. I do want to point out some other highlights, however.
8.11 Section 4: Abbreviated Tables

In this section, Turing introduces some concepts that are central to programming and
software engineering.
There are certain types of process used by nearly all machines, and these,
in some machines, are used in many connections. These processes include
copying down sequences of symbols, comparing sequences, erasing all
symbols of a given form, etc. (p. 235.)
In other words, certain sequences of instructions occur repeatedly in different programs

and can be thought of as being single processess: copying, comparing, erasing, etc.
Turing continues:
Where such processes are concerned we can abbreviate the tables for the
m-configurations considerably by the use of skeleton tables. (p. 235.)
The idea is that skeleton tables are descriptions of more complex sequences of
instructions that are given a single name. This is the idea behind subroutines (or
named procedures) and macros in modern computer programming. If you have a
sequence of instructions that accomplishes what might better be thought of as a single
task (for example, copying a sequence of symbols), and if you have to repeat this
sequence many times throughout the program, it is more convenient (for the human
writer or reader of the program!) to write this sequence down only once, give it a
name, and then refer to it by that name whenever it is needed.29 There is one small
complication: Each time that this named abbreviation is needed, it might require that
parts of it refer to squares or symbols on the tape that will vary depending on the current
configuration, so the one occurrence of this named sequence in the program might need
to have variables in it:
In skeleton tables there appear capital German letters and small Greek
letters. These are of the nature of variables. By replacing each capital
German letter throughout by an m-configuration and each small Greek
letter by a symbol, we obtain the table for an m-configuration.
(pp. 235236.)
Of course, whether one uses capital German letters, small Greek letters, or something
more legible or easier to type is an unimportant, implementation detail. The important
point is this:
The skeleton tables are to be regarded as nothing but abbreviations: they
are not essential. (p. 236.)
29 As Alfred North Whitehead wrote, Civilisation advances by extending the number of important
operations which we can perform without thinking about them. [Hayes, 2014b, 22].
8.12. SECTION 5: ENUMERATION OF COMPUTABLE SEQUENCES 331
8.12 Section 5:
Enumeration of Computable Sequences
Another highlight of Turings paper that is worth pointing out occurs in his 5: a way
to convert every program for a Turing machine into a number. Let me be a bit more
precise about this before seeing how Turing does it.
First, it is important to note that, for Turing, there really is no difference between
one of his a-machines (that is, a Turing machine) and the program for it. Turing
machines are hardwired to perform exactly one task, as specified in the program
(the table, or machine table) for it. So, converting a program to a number is the
same as converting a Turing machine to a number.
Second, converting to a numberthat is, assigning a number to an objectreally
means that you are counting. So, in this section, Turing shows that you can count
Turing machines by assigning a number to each one.
Third, if you can count Turing machines, then you can only have a countable
number of them. But there are an uncountable number of real numbers, so there will
be some real numbers that are not computable!
Here is how Turing counts Turing machines. First (using the lower-case Greek
letter gamma, ):
A computable sequence is determined by a description of a machine
which computes . Thus the sequence 001011011101111. . . is determined
by the table on p. 234, and, in fact, any computable sequence is capable of
being described in terms of such a table. (p. 239.)
A description of a machine is one of the tables such as those we have been looking
at; that is, it is a computer program for a Turing machine.
But, as we have also seen, it is possible to write these tables in various ways. So,
before we can count them, we need to make sure that we dont count any twice because
we have confused two different ways of writing the same table with being two different
tables. Consequently:
It will be useful to put these tables into a kind of standard form. (p. 239.)
The first step in doing this is to be consistent about the number of separate
operations that can appear in the operations column of one of these tables. Note
that in the two programs we have looked at, we have seen examples in which there
were as few as 0 operations and as many as 10 (not to mention the variations possible
with skeleton tables). So:
In the first place let us suppose that the table is given in the same form as
the first table, for example, I on p. 233. [See our 8.10.1, above.] That
is to say, that the entry in the operations column is always of one of the
forms E : E, R : E, L : Pa : Pa, R : Pa, L : R : L : or no entry at all. The table
can always be put into this form by introducing more m-configurations.
(p. 239, my interpolation.)
In other words, the operation in the operations column will be exactly one of:
erase
erase and then move right
erase and then move left
print symbol a
(where a is a variable ranging over all the possible symbols in

a given program)
print a and then move right

print a and then move left
move right
move left
do nothing
Introducing more m-configurations merely means that a single instruction such as:
b 0 P1, R, P0, L f
can be replaced by two instructions:
b 0 P1, R f1
f1 P0, L f
where f1 is a new m-configuration not appearing in the original program. Put

otherwise, a single instruction consisting of a sequence of operations can be replaced
by a sequence of instructions each consisting of a single operation. (For convenience,
presumably, Turing allows pairs of operations, where the first member of the pair is
either E or P and the second is either R or L. So a single instruction consisting of a
sequence of (pairs of) operations can be replaced by a sequence of instructions each
consisting of a single operation or a single such pair.)
Numbering begins as follows:
Now let us give numbers to the m-configurations, calling them q1 , . . . , qR

as in 1. The initial m-configuration is always to be called q1 . (p. 239.)
So, each m-configurations number is written as a subscript on the letter q.

The numbering continues:
We also give numbers to the symbols S1 , . . . , Sm and, in particular, blank =

S0 , 0 = S1 , 1 = S2 . (pp. 239240.)
So, each symbols number is written as a susbscript on the letter S.

Note, finally, that Turing singles out three symbols for special treatment, namely,
0, 1, and what I have been writing as [. (So, Turing is finally making the blank
visible.)
At this point, we have the beginnings of our standard forms, sometimes called
normal forms (which Turing labels N1 , N2 , N3 :
The lines of the table are now of [one of the following three] form[s]
(p. 240):
m-config. Symbol Operations Final

m-config.
qi Sj PSk , L qm (N1 )
qi Sj PSk , R qm (N2 )
qi Sj PSk qm (N3 )
So, we have three normal forms:
N1 m-configuration qi = if currently scanned symbol is S j ,

then begin print symbol Sk ; move left; go to qm end

then begin print symbol Sk ; move right; go to qm end

then begin print symbol Sk ; go to qm end
As Turing notes in the following passage (which I will not quote but merely
summarize), erasing (E) is now going to be interpreted as printing a blank (PS0 ),
and a line in which the currently scanned symbol is S j and the operation is merely to
move right or left is now going to be interpreted as overprinting the very same symbol
(PS j ) and then moving. So, all instructions require printing something: either a visible
symbol or a blank symbol, and then either moving or not moving. As Turing notes:
In this way we reduce each line of the table to a line of one of the forms
(N1 ), (N2 ), (N3 ). (p. 240.)
Turing simplifies even further, eliminating the print command and retaining only
the symbol to be printed. After all, if all commands involve printing something, you
dont need to write down P; you only need to write down what youre printing. So
each instruction can be simplified to a 5-tuple consisting of the initial m-configuration,
the currently scanned symbol (and there will always be one, even if the symbol is
blank, because the blank has been replaced by S0 ), the symbol to be printed (again,
there will always be one, even if its the blank), and the final m-configuration:
From each line of form (N1 ) let us form an expression qi S j Sk Lqm ; from
each line of form (N2 ) we form an expression qi S j Sk Rqm ; and from each
line of form (N3 ) we form an expression qi S j Sk Nqm . (p. 240.)
Presumably, N means something like no move. A slightly more general interpretation

is that, not only do we always print something (even if its a blank), but we also always
move somewhere, except that sometimes we move to our current location. This
standardization is consistent with our earlier observation (in out 7.6.2, above) that the
only two verbs that are needed are print(symbol) and move(location).
Next:
Let us write down all expressions so formed from the table for the machine
and separate them by semi-colons. In this way we obtain a complete
description of the machine. (p. 240.)
Turings point here is that the set of instructions can be replaced by a single string
of 5-tuples separated by semi-colons. There are two observations to make. First,
because the machine table is a set of instructions, there could (in principle) be several
different strings (that is, descriptions) for each such set, because strings are sequences
of symbols. Second, Turing has here introduced the now-standard notion of using a
semi-colon to separate lines of a program; however, this is not quite the same thing as
the convention of using a semi-colon to signal sequencing, because the instructions of a
Turing-machine program are not an ordered sequence of instructions (even if, whenever
they are written down, they have to be written down in some order).
So, Turing has developed a standard encoding of the lines of a program: an m-
configuration encoded as qi (forget about b, f , etc.), a pair of symbols encoded as
S j , Sk (the first being the scanned input, the second being the printed output; again,
forget about things like 0, 1, x, etc.), a symbol (either L, R, or N) encoding the
location to be moved to, and another m-configuration encoded as qm . Next, he gives an
encoding of these standardized codes:
In this description we shall replace qi by the letter D followed by the
letter A repeated i times, and S j by D followed by C repeated j
times. (p. 240.)
Before seeing why he does this, lets make sure we understand what he is doing. The
only allowable m-configuration symbols in an instruction are: q1 , . . . , ql , for some l
that is the number of the final instruction. What really matters is that each instruction
can be assumed to begin and end with an m-configuration symbol, and the only thing
that really matters is which one it is, which can be determined by the subscript on
q. In this new encoding, D simply marks the beginning of an item in the 5-tuple,
and the i occurrences of letter A encode the subscript. Similarly, the only allowable
symbols are: S1 , . . . , Sn , for some n that is the number of the last symbol in the alphabet
of symbols. What really matters is that, in each instruction, the second and third
items in the 5-tuple can be assumed to be symbols (including a visible blank!), and
the only thing that really matters is which ones they are, which can be determined by
the subscript on S. In our new encoding, D again marks the beginning the next item
in the 5-tuple, and the j occurrences of C encode the subscript.
Turing then explains that:
This new description of the machine may be called the standard
description (S.D). It is made up entirely from the letters A, C, D,
L, R, N, and from ;. (p. 240.)
So, for example, this 2-line program:
q3 S1 S4 Rq5
q5 S4 S0 Lq5
will be encoded by an S.D consisting of this 38-character string:
DAAADCDCCCCRDAAAAA; DAAAAADCCCCDLDAAAAA
The next step in numbering consists in replacing these symbols by numerals:
If finally we replace A by 1, C by 2, D by 3, L by 4, R
by 5, N by 6, ; by 7 we shall have a description of the machine
in the form of an arabic numeral. The integer represented by this numeral
may be called a description number (D.N) of the machine. (p. 240; Turing
did not capitalize arabic.)
Just as Godel numbering is a way to create a number corresponding to a string, so is

this another way to do that. The D.N of the machine in our previous example is this
numeral:
31113232222531111173111113222234311111
which, written in the usual notation with commas, is:
31,113,232,222,531,111,173,111,113,222,234,311,111
or, in words, 31 undecillion, 113 decillion, 232 nonillion, 222 octillion, 531 septillion,
111 sextillion, 173 quintillion, 111 quadrillion, 113 trillion, 222 billion, 234 million,
311 thousand, one hundred eleven. That is the Turing number of our 2-line program!
Turing observes that:
The D.N determine the S.D and the structure of the machine uniquely. The
machine whose D.N is n may be described as M (n). (pp. 240242.)
Clearly, given a D.N, it is trivial to decode it back into an S.D in only one way.
Equally clearly (and almost as trivially), the S.D can be decoded back into a program
in only one way. Hence, the structure of the machine encoded by the D.N is
determine[d]. . . uniquely by the D.N. However, because of the possibility of writing
a program for a machine in different ways (permuting the order of the instructions),
two different D.Ns might correspond to the same machine, so there will, in general be
distinct numbers n, m (that is, n 6= m) such that M (n) = M (m). That is, the Turing
machine whose number = n might be the same machine as the one whose number =
m; a given Turing machine might have two different numbers. Alternatively, we could
consider that we have here two different machines that have exactly the same input-
output behavior and that execute exactly the same algorithm. Even in that latter case,
where we have more machines than in the former case, the machines are enumerable;
that is, we can count them. Can we also count the sequences that they compute?
Yes; Turing explains why (with my comments interpolated in brackets):
To each computable sequence [that is, to each sequence that is printed to

the tape of a Turing machine] there corresponds at least one description
number [we have just seen why there might be more than one], while to
no description number does there correspond more than one computable
sequence [that is, each machine prints out exactly one sequence; there is
no way a given machine could print out two different sequences, because
the behavior of each machine is completely determined by its program,
and no program allows for any arbitrary, free, or random choices that
could vary what gets printed on the tape]. The computable sequences and
numbers [remember: every sequence corresponds to a unique number]30
are therefore enumerable [that is, countable]. (p. 241.)
Next, on p. 241, Turing shows how to compute the D.N of program I (the one that
printed the sequence 01). And he gives a D.N without telling the reader what program
corresponds to it.
Finally, he alludes to the Halting Problem:
A number which is a description number of a circle-free machine will be
called a satisfactory number. In 8 it is shown that there can be no general
process for determining whether a given number is satisfactory or not.
(p. 241.)
A satisfactory number is the number of a circle-free Turing machine, that is, a Turing
machine that never halts and that does compute the infinite decimal representation of a
real number. That is, a satisfactory number is the number of a Turing machine for a
computable number. So, in Turings 8, he is going to show that there is no general
processthat is, no Turing machine that can decide (by computing)whether a
given number is satisfactory, that is, whether a given number is the number of a
circle-free Turing machine. It is easy to determine if a given number is the number
of a Turing machine: Just decode it, and see if the result is a syntactically correct,
Turing-machine program. But, even if it is a syntactically correct, Turing-machine
program, there will be no way to decide (that is, to compute) whether it halts or not.
(Remember: For Turing, halting is bad, not halting is good; in modern presentations of
computing theory, halting is good, not halting is (generally considered to be)31 bad.)
8.13 Section 6: The Universal Computing Machine

Although this section is at least one of, if not the most important section of Turings
paper, we will only look at it briefly in this chapter. As before, you are encouraged to
consult [Petzold, 2008] for aid in reading it in detail.
Turing begins with this claim:
It is possible to invent a single machine which can be used to compute any
computable sequence. (p. 241.)
So, instead of needing as many Turing machines as there are computable numbers,
we only need one. Recall that our first great insight was that all information can be
represented using only 0 and 1 (7.6.1, above). That means that all information that
30 Although, because of a curiosity of decimal representation, some numbers correspond to more than one
sequence. The classic example is that 1 = 1.0 = 0.9.

31 But see Ch. 11!
8.13. SECTION 6: THE UNIVERSAL COMPUTING MACHINE 337
we would want to compute with can be represented by a sequence of 0s and 1s, that
is, as a computable number (in binary notation). So, Turings claim is that there is a
single machine that can be used to compute anything that is computable.
Most of you own one. Indeed, most of you own several, some of which are small
enough to be carried in your pocket! They are made by Apple, Dell, et al.; they
come in the form of laptop computers, smartphones, etc. They are general-purpose,
programmable computers.
If this machine U is supplied with a tape on the beginning of which is

written the S.D of some computing machine M , then U will compute the
same sequence as M . (pp. 241242.)
Your laptop or smartphone is one of these U s. A program or app that you download
to it is an S.D (written in a different programming language than Turings) of a Turing
machine that does only what that program or app does. The computer or smartphone
that runs that program or app, however, can also run other programs, in fact, many
of them. Thats what makes it universal:
But to do all the things a smartphone can do without buying one,. . . [a]
consumer would need to buy the following:
A cellphone. . . .
A mobile e-mail reader. . . .
A music player. . . .
A point-and-shoot camera. . . .
A camcorder. . . .
A GPS unit. . . .
A portable DVD player. . . .
A voice recorder. . . .
A watch. . . .
A calculator. . . .
Total cost: $1,999
In a smartphone, all those devices are reduced to software.
[Grobart, 2011]
How does Turings universal computer work? Pretty much the same way that a
modern computer works: A program (an app) is stored somewhere in the computers
memory; similarly, the S.D of a Turing machine is written at the beginning of the
universal machines tape. The operating system of the computer fetches (that is, reads)
an instruction and executes it (that is, simulates its behavior [Dewdney, 1989, 315]),
then repeats this fetch-execute cycle until there is no next instruction; similarly,
the single program for the universal machine fetches the first instruction on its tape,
executes it, then repeats this cycle until there is no next instruction on its tape. The
details of how it does that are fascinating, but beyond our present scope. (Again,
[Petzold, 2008], as well as [Dewdney, 1989, Chs. 1, 28, 48], cover this in detail.)
8.14 The Rest of Turings Paper

Sections 15 of Turings paper cover the nature of computation, defining it precisely,
and stating what is now called Turings (computability) thesis. Sections 6 and 7 of
Turings paper cover the universal machine. Section 8 covers the Halting Problem.
We have already examined Section 9 in detail; that was the section in which Turing
analyzed how humans compute and then designed a computer program that would do
the same thing. The branch of computer science that analyzes how humans perform a
task and then designs computer programs to do the same thing is AI; so, in Section 9,
Turing has developed the first AI program:
Turing machines appear [in Turings paper] as a result, a codification, of

his analysis of calculations by humans. [Gandy, 1988, 82]
Turing was not unaware of this aspect of his work:
One way of setting about our task of building a thinking machine

would be to take a man [sic] as a whole and to try to replace
all the parts of him by machinery. [Turing, 1948, 420] as cited in
[Proudfoot and Copeland, 2012, 2]
But thats almost exactly what Turings analysis of human computation in his 1936
paper does (at least in part): It takes a humans computational abilities and replaces
them by (abstract) machinery.
Section 10 shows how it can be that many numbers that one might think
are not computable are, in fact, computable. Section 11 proves that Hilberts
Entscheidungsproblem can have no solution (p. 259). And the Appendix proves that
Turings notion of computation is logically equivalent to Churchs.
Except for modern developments and some engineering-oriented aspects of
computer science, one could create an undergraduate degree program in computer
science based solely on this one paper that Turing wrote in 1936!

1. The childrens game of Candyland is like a TM: The (randomly shuffled) deck
of color cards is like the TM table, telling us what to do. At each point, we can
move a marker left or right. Some squares on the board, which constitutes the
tape, have other instructions on them, but those could have been encoded in the
cards. The game is completely deterministic, except that it doesnt necessarily
compute anything of interest because of the random arrangement of the cards.
Chess is also completely deterministic, but so large that we cant play it
deterministically.
2. On stored program computer:
(a) [Haigh, 2013] argues that the phrase stored-program computer is

ambiguous between several different readings and often conflated with the
notion of a universal TM. In my experience, the phrase refers to the idea

that a computers program can be stored in the computer itself (for example,
on a TMs tape) and be changed, either by storing a different program or
by modifying the program itself (perhaps while it is being executed, and
perhaps being (self-)modified by the program itself). However, when I
asked a colleague who first came up with the notion of stored program
(fully expecting him to say either Turing or von Neumann), he replied
quite reasonablyJacquard.32
On this understanding, the phrase becomes key to understanding the

difference between software and hardware (or programmed vs. hardwired
computer)see Ch. 13 for more on thisand it becomes a way of viewing
the nature of the universal TM. But [Haigh, 2013] gives reasons to believe
that there are so many other ways to understand the phrase that it would be
better to refrain from using it.
(b) [Randell, 1994, 12], also discussing the controversy over what stored
programming ismakes a statement that suggests that the main difference
between programmed vs. hardwired computers might lie in the fact that a
program held on some read-only medium, such as switches, punched cards,
or tape. . . was quite separate from the (writable) storage device used to hold
the information that was being manipulated by the machine. [S]toring the
program within the computer, in a memory that could be read at electronic
speeds during program execution had certain advantages first noted by
the ENIAC/EDVAC team (p. 13). Randell, however, thinks that the
analogy with UTMs is more central (p. 13), and that EDVAC does not
qualify as a stored-program computer because the representations [of
data and instructions] were quite distinct, and no means were provided for
converting data items into instructions or vice versa (p. 13)
(c) Computers, in any informal sense of the term (think laptop or even
mainframe computer), are programmable. Turing machines are not! But
universal TMs are!
The ability to store a program on a UTMs tape is what makes it

programmable (that is, it can be changed from simulating the behavior of
one TM to simulating the behavior of a different one).
But, logically, a program need not be stored physically in the computer:

It could control the computer via a wireless connection from a different
location. The ability to store a program in the computer along with the data
allows for the program to change itself.
(d) [A] Turing machine. . . was certainly thought of as being
programmed in a hardwired way. . . . It is reasonable to view the
universal Turing machine as being programmed by the description
32 Stuart C. Shapiro, personal communication, 7 November 2013.
of the machine it simulates; since this description is written on

the memory tape of the universal machine, the latter is an abstract
stored program computer. [Carpenter and Doran, 1977, 270]
First, note that TMs are hardwired. Second, note that what makes the UTM
stored-program is not so much that its program is not hardwired, but that
it is stored on its tape.
(e) See also [Robinson, 1994], cited in Further Sources.
(f) See also Vardi 1013 and Copeland 2013, cited in Ch. 6s Further Sources.
(g) And here is von Neumann on the concept:
If the device [the very high speed automatic digital computing
sytem (1.0, p. 1) is to be elastic, that is as nearly as possible
all purpose, then a distinction must be made between the specific
instructions given for and defining a particular problem, and the
general control organs which see to it that these instructions
no matter what they areare carried out. The former must
be stored in some way. . . the latter are represented by definite
operating parts of the device. By the central control we mean this
latter function only. . . . [von Neumann, 1945, 2.3, p. 2; italics in
original, my boldface].
The specific instructions seems clearly to refer to a specific Turing
machines program as encoded on the tape of a universal Turing machine.
The central control seems clearly to refer to the universal Turing
machines program. So, if this is what is meant by stored program, then
it pretty clearly refers to the way that a universal Turing machine works.
(h) The following comments might belong with the discussion of software and
hardware in Ch. 12 or of rule-following in Ch. 10:
There is a sense in which a stored-program computer does two things:

It executes a hard-wired fetch-execute cycle, and it executes whatever
software program is stored on its tape. Which is it really doing?
[Newell, 1980, 148] suggests that it is the former:
A machine is defined to be a system that has a specific determined
behavior as a function of its input. By definition, therefore, it
is not possible for a single machine to obtain even two different
behaviors, much less any behavior. The solution adopted is to
decompose the input into two parts (or aspects): one part (the
instruction) being takento determine which input-output function
is to be exhibited by the second part (the input-proper along with
the output.
I confess to not fully understanding this passage. But one way to
interpret it is to take the decomposition to refer to the hard-wired, fetch-
execute program of the univeral Turing machine, on the one hand, and
the software, stored program of the virtual machine, on the other hand.
The universal Turing machine thus (consciously) executes the software

program indirectly by (unconsciously) directly executing its hard-wired
program. It simulates a Turing machine that has that software program
hard-wired as its machine table [Newell, 1980, 150].
(i) According to [Daylight, 2013, p. XX, n. 3], the phrase stored program
was not commonly used in the early 1950s. Furthermore, according to
[Daylight, 2013, 3, p. VII], storing data and instructions together was
based on practical concerns, not theoretical reasoning of the sort that might
have been inspired by Turings notion of a universal Turing machine.
3. The ability to erase has a downside: It destroys information, making it difficult,

if not impossible, to reverse a computation. (See, for example, [Hayes, 2014b,
23].)
4. The idea that somehow printing out 010101. . . computes (say) 1/3 in base 2
is related to the idea that TM-computation is automatic or mechanical.
Consider any of the lengthy TM programs in his paper. Does a human following
them understand what s/he is doing? This gives rise to Searles Chinese Room
Argument. Dennett has things to say about this, too, in his Darwins Strange
Inversion paper.
So does Moravec:
Several times during both matches [with Deep Blue], Kasparov

reported signs of mind in the machine.
. . . In all other chess computers, he reports a mechanical
predictability. . . . In Deep Blue, to his consternation, he saw instead
an alien intelligence.
. . . [T]he evidence for an intelligent mind lies in the machines
performance, not its makeup.
Now, the team that built Deep Blue claim no intelligence
in it, only a large database of opening and end games, scoring
and deepening functions tunes with consulting grandmasters, and,
especially, raw speed that allows the machine to look ahead an
average of fourteen half-moves per turn. . . .
Engineers who know the mechanism of advanced robots most
intimately will be the last to admit they have real minds. From
the inside robots will indisputably be machines, acting according
to mechanical principles, however elaborately layered. Only on
the outside, where they can be appreciated as a whole, will the
impression of intelligence emerge. A human brain, too, does not
exhibit the intelligence under a neurobiologists microscope that it
does participating in a lively convesation. [Moravec, 1998, 10]
But this shows that there are two issues, both of which are consistent with
the strange inversion: First, Moravecs discussion, up to the last sentence,
is clearly about external behavior independent of internal mechanism. In this
sense, its consistent with the Turing test view of intelligence. Intelligence is
computable, even if it isnt computed. Interestingly, in Deep Blue, it is computed,
just not in the way that humans compute it or that other kinds of computers might
compute it. But the last sentence points to the second interpretation, which is
more consistent with the strange inversion, namely, that, even if the internal
mechanism is computing intelligent behavior in the way that humans do, looking
at it at that level wont make that intelligence manifest. Intelligent behavior at
the macro level can emerge from, or be implemented by, non-intelligent behavior
at the micro-level. This is Dennetts point about the ever-smaller homunculi who
bottom out in ones who can only say yes or no.
5. A TM is to a UTM as a music box is to a player piano: TMs and music boxes
can only play/execute the tune/program that is hardwired into them. UTMs
and player pianos can play/execute any tune/program that is encoded on its
tape/piano-roll. For discussion of this idea, see [Sloman, 2002].
Heres a related question: Why is a player piano not a computer?
[Kanat-Alexander, 2008]. Alternatively, when is a universal Turing machine a
player piano? The instructions on the piano roll cause certain keys to be
played; you can think of each key as a Turing-machine tape cell, with play
or dont play analogous to print-one or print-zero. One difference is that
a player piano would be a parallel machine, because you can play chords.
6. [Sieg, 2006, 13] points out that Turings analysis of (informal) computation
abstracts away from arithmetical computations (such as addition, multiplication,
etc.), in favor of purely symbolic processes (such as making a mark on paper,
moving to a different location on the paper, etc.). Sieg says, It is for this
very reason that Turing most appropriately brings in human computers in a
crucial way and exploits the limitations of their processing capacities, when
proceeding mechanically. Note, too, that this is a move (that abstracts) away
from semantically interpreted processes and towards purely (uninterpreted)
syntactic ones.
This is one of the reasons that people like Searle find it difficult to understand
how a purely syntactic device (a computer) can produce semantic results (can
do arithmetic, can understandor, at least, processnatural language, etc.). We
will return to this issue in Ch. 19. This paradox is discussed in [Dennett, 2009].
And it is related to issues about the goals of an algorithm (see 7.5) and the
relation between the internal workings of an algorithm and its relationships to
the external world (to be discussed in Chs. 10 and 17).
7. [Kleene, 1995, 19] observes that Turings emphasis on not allowing an infinity
of symbols that differ. . . to an arbitrarily small extent (8.7.2.1, above) marks
the distinction between digital computation rather than analog computation.
8. [T]he computer is [not] restricted to taking an ants eye view
of its work, squinting at the symbol on one square at a time.
. . . [T]he Turing-machine squares can corresond to whole sheets of
paper. If we employ sheets ruled into 20 columns and 30 lines, and
authorize 99 primary symbols, there are 100600 = 101200 possible

square conditions, and we are at the opposite extreme. The schoolboy
[sic] doing arithmetic on 8 12 by 12 sheets of ruled paper would never
need, and could never utilize, all this variety.
Another representation of a Turing machine tape is as a stack of
IBM cards, each card regarded as a single square for the machine.
[Kleene, 1995, 26]
9. A Turing machine is like an actual digital computing machine, except

that (1) it is error free (i.e., it always does what its table says it should
do), and (2) by its access to an unlimited tape it is unhampered by
any bound on the quantity of its storage of information or memory.
[Kleene, 1995, 27]
The second limitation of a real (physical) Turing machine is not a very serious
one, given (a) the option of always buying another square and (b) the fact that
no computation could require an actual infinity of squares (else it would not be
a finite computation). The more significant limitation is that some computations
might require more squares than there could be in the universe (as is the case
with NP computations such as chess).
The former limitation of real Turing machines does not obviate the need
for program verification. Even an ideal Turing machine could be poorly
programmed.
10. On Darwins and Turings strange inversion:
(a) Francis Crick, in his Danz lectures Of Molecules and Men,

discusses the problem of how life could have arisen:
[This] really is the major problem in biology. How
did this complexity arise?
The great news is that we know the answer to
this question, at least in outline. . . . The answer was
given over a hundred years ago by Charles Darwin. . . .
Natural selection. . . provides an automatic mechanism
by which a complex organism can survive and increase
in both number and complexity.
For us in Cognitive Science, the major problem is how it is
possible for mind to exist in this physical unverse. The great
news. . . is that we know, at least in outline, how this might be.
[Newell, 1980, 182]
According to Newell, the answer was given in 1936 by Alan Turing.
Computation provides an automatic mechanism by which a machine (living
or otherwise) can exhibit cognitive behavior.
(b) The inversion seems to concern the paradox that intelligent behavior
can emerge from unintelligent or mechanical behavior. Simon says
some things that suggest that the paradox originates in an equivocation on

mechanical: He says that both computers and brains are mechanisms:
If by a mechanism we mean a system whose behavior at a
point in time is determined by its current internal state combined
with the influences that simultaneously impinge upon it from
outside, then any system that can be studied by the methods of
science is a mechanism.
But the term mechanism is also used in a narrower
sense to refer to systems that have the relatively fixed, routine,
repetitive behavior of most of the machines we see around us.
[Simon, 1996a, 165]
Mechanisms in the latter sense do not exhibit self-generated spontaneity,
that is, behavior that is unpredicted, perhaps even by the behaving system
[Simon, 1996a, 165]. This kind of spontaneity is exhibited by intelligent
behavior. Turings inversion concerns the fact that a computer can be a
mechanism in the first sense without being one in the second sense. As
Simon says:
Clearly the computer occupies an ambiguous position here.
Its behavior is more complex, by orders of magnitude, than any
machine we have known; and not infrequently it surprises us,
even when it is executing a program that we wrote. Yet, as the
saying goes, it only does what you program it to do But truism
though that saying appears to be, it is misleading on two counts.
It is misleading, first, because it is often interpreted to mean: It
only does wht you believe you programmed it to do, which is
distinctly not the case.
More serious, it is misleading because it begs the question
of whether computers and people are different. They are different
(on this dimension) only if people behave differently from the way
they are programmed to behave. But if we include in program
the whole state of human memory, then to assert that people
dont do only what they are programmed to do is equivalent
to asserting that peoples brains are not mechanisms, hence not
explainable by the methods of science. [Simon, 1996a, 165]
Simons point is that people are no more spontaneous than computers and
that computers are no less mechanistic than people. (For similar comments,
see [Simon, 1996a, 1417].)
11. . . . Turing asked in the historical context in which he found himself
the pertinent question, namely, what are the possible processes
a human being can carry out (when computing a number or,
equivalently, determining algorithmically the value of a number
theoretic function)? [Sieg, 2000, 6; original italics, my boldface]
12. On TM as first AI program:
The intent [of a first computer science course should be] to reveal,
through. . . examples, how analysis of some intuitively performed
human tasks leads to mechanical algorithms accomplishable by a
machine. [Perlis, 1962, 189]
13. That an algorithm should demand no insight or ingenuity on the part of the
human being carrying it out [Copeland, 2000] has the implication that the
human is used as a biological computing machine [PolR, 2009].
First, although something has to carry the algorithm out, it need not be a human
being (though, of course, in Turings analysis, it is, or, at least, it is an idealized
human). So, more generally, the implication is that the interpreter/executor is the
computing machine. So far, so good.
Second, [PolR, 2009]s point is that the human must make no judgment call and
use no skill beyond his ability to apply the rules with pencel and paper. Clearly,
if the executor is not human, it wont be able to make such judgment calls or use
any such skill. Only if the executor is human does this become a worry. This
seems to be what lies behind Searles Chinese Room Argument, to be discussed
in Ch. 19. It is also what underlies Turings strange inversion, namely, that no
such judgment calls or skillsor even biologyare needed at all.
14. The halting theorem is often attributed to Turing in his

1936 paper. In fact, Turing did not discuss the halting
problem, which was introduced by Martin Davis in about 1952.
[Copeland and Proudfoot, 2010, 248, col. 2]
The authors dont elaborate on this. Perhaps it has something to do with the
difference between halting and circle-free?
15. More on Turings inversion (and magic):
A telegraph! Mme de Villefort repeated.

Yes, indeed, the very thing: a telegraph. Often, at the end of a
road, on a hilltop, in the sunshine, I have seen those folding black
arms extended like the legs of some giant beetle, and I promise you,
never have I contemplated them without emotion, thinking that those
bizarre signals so accurately travelling through the air, carrying the
unknown wishes of a man sitting behind one table to another man
sitting at the far end of the line behind another table, three hundred
leagues away, were written against the greyness of the clouds or the
blue of the sky by the sole will of that all-powerful master. Then I
have thought of. . . occult forces, and laughed. Never did I wish to
go over and examine these great insects with their white bellies and
slender black legs, because I was afraid that under their stone wings
I would find the human genie, cramped, pedantic, stuffed with arcane
science and sorcery. Then, one fine morning, I discovered that the
motor that drives every telegraph is a poor devil of a clerk who earns
twelve hundred francs a year and. . . spend sthe whole day staring at
the insect with the white belly and black legs that corresponds to his
own and is sited some four or five leagues away. At this, I became
curious to study this living chrysalis from close up and to watch the
dumbshow that it offers from the bottom of its shell to that other
chrysalis, by pulling bits of string one after the other.
[At the telegraph belonging to the Ministry of the Interior] I
should find people who would try to force me to understand things
of which I would prefer to remain ignorant. . . . . . . I want to keep my
illusions about insects. . . . [Dumas, 1844, Ch. 40, pp. 676677]
16. Related to the previous quote, and perhaps also to the Chinese Room Argument:
Did Monsieur come to look at the telgraph?

Yes, provided the rules do not forbid it.
Not at all, said [the telegraph operator]. There is not danger,
because no one knows or can know what we are saying.
Yes, indeed, said the count [of Monte Cristo]. I have even been
told that you repeat signals that you do not understand yourselves.
Indeed we do, Monsieur, and I much prefer that, the telegraph
man said, laughing.
Why do you prefer it?
Because in that way I have no responsibility. I am a machine and
nothing more. A long as I work, no one asks anything more from me.
[Dumas, 1844, Ch. 41, p. 681]
17. Need to say something about The Imitation Game movie, including a list
of the other films/plays about Turing. Also cite the websites that serve as
background/critique for The Imitation Game, including:
(a) The Imitation Game: The Philosophical Legacy of Alan Turing, The
Critique (31 December 2014)
http://www.thecritique.com/news/the-imitation-game-the-philosophical-
legacy-of-alan-turing/
Separately highlight some of the articles.
(b) Anderson, L.V. (2014), How Accurate Is The Imitation Game?, Slate
(3 December),
http://www.slate.com/blogs/browbeat/2014/12/03/the imitation game fact vs fiction how true the new movie is to alan turing
(c) Caryl, Christian (2015), Saving Alan Turing from His Friends, New York
Review of Books, (5 February): 1921,
http://www.nybooks.com/articles/archives/2015/feb/05/saving-alan-
turing-his-friends/
(accessed 13 February 2015).
(d) von Tunzelmann, Alex (2014), The Imitation Game: Inventing a New
Slander to Insult Alan Turing (20 November),
http://www.theguardian.com/film/2014/nov/20/the-imitation-game-
invents-new-slander-to-insult-alan-turing-reel-history
(e) Woit, Peter (2014), The Imitation Game,, Not Even Wrong,
http://www.math.columbia.edu/woit/wordpress/?p=7365
18. Here are some of the other films:
(a) Breaking the Code (1986),
http://www.imdb.com/title/tt0115749/,
http://en.wikipedia.org/wiki/Breaking the Code
(b) Enigma (2001),
http://en.wikipedia.org/wiki/Enigma (2001 film)
This fictionalized version omits Turing!
(c) Codebreaker: The Story of Alan Turing (2011),
http://en.wikipedia.org/wiki/Codebreaker (film)
Half dramatization, half documentary, with interviews of people who
knew Turing (including some who are fictionalized in The Imitation
Game).
19. A man provided with paper, pencil, and rubber [eraser], and subject to
strict discipline, is in effect a universal machine. [Turing, 1948, 416]
20. A Turing machine computing a partial function was called
circular. . . . For some inputs, circular machines did not generate
outputs of the desired type. [Piccinini, 2003, 42n18]
Check this against my interpretation of circular, above. Note that generating
output (for all inputs) makes a function total, but, presumably, generating output
that is not. . . of the desired type is the equivalent of generating garbage or
having unpredictable behavior, that is, of being undefined.
21. The shortest description of Turings accomplishment?:
[Turing showed that any computation can be described in terms of

a machine shifting among a finite numbe of states in response to a
sequence of symbols read and written one at a time on a potentially
infinite tape. [Mahoney and Haigh (ed.), 2011, 79]
Compare my theses and both [Thomason, 2003, 328] and [Lamport, 2011, 6]
cited in 9.4.

American Mathematical Society (2006), Special issue on Alan Turing, Notices of
the AMS 53(10) (November).
Table of contents and articles accessed 30 Mauy 2014 from:
http://www.ams.org/notices/200610/
Includes articles by Andrew Hodges, Solomon Feferman, S. Barry Cooper,
and Martin Davis, among others.
Bullynck, Maarten; Daylight, Edgar G.; & De Mol, Liesbeth (2015), Why Did
Computer Science Make a Hero out of Turing?, Communications of the ACM 58(3)
(March): 3739,
http://cacm.acm.org/magazines/2015/3/183592-why-did-computer-science-make-
a-hero-out-of-turing/fulltext
Carpenter, B.E.; & Doran, R.W. (1975?), The Other Turing Machine, The
Computer Journal 20(3): 269279.
Cooper, S. Barry (2012a), Incomputability after Alan Turing, Notices of the AMS
59(6): 776784.
Preprint accessed 26 February 2014 from:
http://arxiv.org/pdf/1304.6363.pdf
Is there a causal dimension that is not algorithmic? Does it matter if there
is? (p. 776)
Cooper, S. Barry (2012b), Turings Titanic Machine?, Communications of the
ACM 55(3) (March): 7483.
Accessed 26 February 2014 from:
http://cacm.acm.org/magazines/2012/3/146259-turings-titanic-
machine/fulltext
[E]xamine[s] challenges to. . . [the] continued primacy [of universal Turing
machines] as a model for computation in daily practice and in the wider
universe.
Cooper, S. Barry; & van Leeuwen, Jan (2013), Alan Turing: His Work and Impact
(Amsterdam: Elsevier).
A large, coffee table-sized anthology of Turings writings with
commentaries by, among many others, Rodney Brooks, Gregory Chaitin,
B. Jack Copeland, Martin Davis, Daniel Dennett, Luciano Floridi, Lance
Fortnow, Douglas Hofstadter, Wilfried Sieg, Aaron Sloman, Robert I. Soare,
and Stephen Wolfram.
Reviewed in:
Avigad, Jeremy (2014), Alan Turing: His Work and Impact, Notices of the
American Mathematical Society 61(8) (September): 886890,
http://www.ams.org/notices/201408/201408-full-issue.pdf (accessed
24 March 2015).
Copeland, B. Jack (1997), The Broad Conception of Computation, American
Behavioral Scientist 40(6) (May): 690716.
An essay on hypercomputation (or non-classical computation), but the

introductory section (pp. 690698) contains an enlightening discussion of the
scope and limitations of Turings accomplishments.
Copeland, B. Jack (ed.) (2004), The Essential Turing (Oxford: Oxford University
Press).
An anthology of Turings papers.
Critically reviewed in:
Hodges, Andrew (2006), Book Review of The Essential Turing, in Notices of
the AMS 53(10) (November): 11901199; accessed 13 June 2014 from:
http://www.ams.org/notices/200610/rev-hodges.pdf
Copeland, B. Jack; & Shagrir, Oron (2013), Turing versus Godel on Computability
and the Mind, in B. Jack Copeland, Carl J. Posy, & Oron Shagrir (eds.),
Computability: Turing, Godel, Church, and Beyond (Cambridge, MA: MIT Press):
133.
Explores both Godels interpretations of Turings work and the relation of the
human mind to Turing machines.
Curtis, M.W. (1965), A Turing Machine Simulator, Journal of the ACM 12(1)
(January): 113.
Contains a program written for the IBM 1620 designed for education purposes.
Davis, Martin (2004), The Myth of Hypercomputation, in C. Teuscher (ed.),
212.
Contains a good discussion of Turings role in the history of computation.
Daylight, Edgar G. (2014), A Turing Tale, Communications of the ACM 57(10)
(October): 3638,
http://cacm.acm.org/magazines/2014/10/178787-a-turing-tale/fulltext
(accessed 5 December 2014)
Assessing the accuracy of popular descriptions of Alan Turings influences
and legacy (Abstract, p. 36)
Dewdney, A.K.:
Ch. 1: Algorithms: Cooking Up Programs
Ch. 28: Turing Machines: The Simplest Computers
Ch. 48: Universal Turing Machines: Computers as Programs
from: Dewdney, A.K. (1989), The Turing Omnibus: 61 Excursions in Computer
Science (Rockville, MD: Computer Science Press).
Fitzsimmons, Emma G. (2013), Alan Turing, Enigma Code-Breaker and Computer
Pioneer, Wins Royal Pardon, New York Times (24 December).
http://www.nytimes.com/2013/12/24/world/europe/alan-turing-enigma-code-
breaker-and-computer-pioneer-wins-royal-pardon.html
Gandy, Robin (1980), Churchs Thesis and Principles for Mechanisms, in Jon
Barwise, H.J. Keisler, & K. Kunen (eds.), The Kleene Symposium (Amsterdam:
North-Holland): 123148.
Gandy, who was Turings only Ph.D. student, argues that Turings analysis of
computation by a human being does not apply directly to mechanical devices.
Sieg, Wilfried, & Byrnes, John (1999), An Abstract Model for Parallel
Computation: Gandys Thesis, The Monist 82(1) (January): 150164.
A follow-up article that simplifies and generalizes Gandys paper.
Israel, David (2002), Reflections on Godels and Gandys Reflections on
Turings Thesis, Minds and Machines 12(2) (May): 181201.
Shagrir, Oron (2002), Effective Computation by Humans and Machines,
Minds and Machines 12(2) (May): 221240.
Haigh, Thomas (2014), Actually, Turing Did Not Invent the Computer,
Discusses the role of the Turing machine in the history of computers,
minimizing its historical importance to the actual development of computing
machines.
Hayes, Brian (2007), Trains of Thought, American Scientist 95(2) (March-April):
108ff.
http://www.americanscientist.org/issues/pub/trains-of-thought
An implementation of a Turing machine using railroad trains. See also Stewart
1994, below.
Hodges, Andrew (2012), Alan Turing: The Enigma; Centenary Edition
(Princeton, NJ: Princeton University Press).
Reviews of first edition of Hodgess biography:
(a) Bernstein, Jeremy (1986), A Portrait of Alan Turing, New Yorker
(20 January): 78, 8187.
http://www.cdpa.co.uk/Newman/MHAN/view-
item.php?Box=5&Folder=6&Item=5&Page=1
(b) Hofstadter, Douglas B. (1983), Mind, Body and Machine, New York
Times Book Review (13 November).
(c) Toulmin, Stephen (1984), Fall of a Genius, New York Review of Books
(19 January): 34, 6.
http://www.nybooks.com/articles/archives/1984/jan/19/fall-of-a-
genius/
See also:
Stern, Howard; & Daston, Lorraine (1984), Turing & the System
(letter to the editor, with reply by Toulmin), New York Review of Books
(26 April): 5253; accessed 18 February 2014 from:
http://www.nybooks.com/articles/archives/1984/apr/26/turing-the-
system/
See also Hodgess Alan Turing: The Enigma website, accessed 18 February
2014 from:
http://www.turing.org.uk/index.html
Copeland, B. Jack; & Proudfoot, Diane (1996), On Alan Turings Anticipation of
Connectionism, Synthese 108: 361377.
. . . Turing was probably the first person to consider building computing
machines out of simple, neuron-like elements connected together into
networks in a largely random manner. . . . By the application of what
he described as appropriate interference, mimicking education. . . [such a]
machine can be trained to perform any task that a Turing machine can carry
out. . . . (p. 361.)
For more information on this, see:
http://tinyurl.com/CopelandProudfoot199633 (accessed 21 February 2014).
Feferman, Solomon (2006), Turings Thesis, Notices of the AMS 53(10): 1200
1206.
An overview of Turings work focusing on his PhD thesis (not the Church-
Turing thesis) that introduced oracle machines.
Kleene, Stephen C. (1988), Turings Analysis of Computability and Major
Applications of It, in Rolf Herken (ed.), Universal Turing Machine: A Half-
Century Survey; Second Edition (Vienna: Springer-Verlag, 1995): 1549.
Leavitt, David (2005), The Man Who Knew Too Much: Alan Turing and the
Invention of the Computer (W.W. Norton).
Another biography of Turing, reviewed by S. Barry Cooper in Notices of the
AMS 53(10) (2006): 12131217; accessed 13 June 2014 from:
http://www.ams.org/notices/200610/rev-cooper.pdf
Leiber, Justin (2006), Turings Golden: How Well Turings Work Stands Today,
Philosophical Psychology 19(1): 1346.
Michie, Donald (2008), Alan Turings Mind Machines, in Philip Husbands, Owen
Holland, & Michael Wheeler (eds.), The Mechanical Mind in History (Cambridge,
MA: MIT Press), Ch. 4, pp. 6174.
Muggleton, S. (1994), Logic and Learning: Turings Legacy, in K. Furukawa,
D. Michie, & S. Muggleton (eds.), Machine Intelligence 13: Machine Intelligence
and Inductive Learning (Oxford: Clarendon Press): 3756.
[I]n this paper it is shown that there is a direct evolution in Turings ideas
from his earlier investigations of computability to his later interests in machine
intelligence and machine learning.
Myhill, John (1952), Some Philosophical Implications of Mathematical Logic:
I. Three Classes of Ideas, Review of Metaphysics 6(2) (December): 165198.
Contains some observations on what it means for a problem to be solvable.
Myhill, John (1972), What Is a Real Number?, American Mathematical Monthly
79(7) (August-September): 748754.
33 http://www.alanturing.net/turing archive/pages/reference%20articles/connectionism/Turingsanticipation.html
Not directly related to Turing, but consider[s] the notion of real numbers from
a constructive point of view. The point of view requires that any real number
can be calculated (p. 748), that is, computed, which is (in part) what Turings
1936 paper is about.
Petzold, Charles (2008), The Annotated Turing: A Guided Tour through Alan
Turings Historic Paper on Computability and the Turing Machine (Indianpolis:
Wiley).
Davis, Martin (2008), Touring Turing (review of Petzold 2008), American
Scientist 96(6) (November-December): 520.
Also see websites below.
Piccinini, Gualtiero (2003), Alan Turing and the Mathematical Objection, Minds
and Machines 13: 2348,
http://www.umsl.edu/piccininig/Alan Turing and Mathematical Objection.pdf
Primarily about Turings views on AI (see Ch. 19), but also discusses his
theory of computation and the role of oracle machines.
Rapaport, William J. (1985), Turing Machines, Appendix B of Morton L.
Schagrin, Randall R. Dipert, & William J. Rapaport, Logic: A Computer Approach
(New York: McGraw-Hill): 327339.
Accessed 7 November 2012 from:
http://www.cse.buffalo.edu/rapaport/Papers/schagrinetal85-
TuringMachines.pdf
A chapter on Turing machines from an introductory logic textbook; see also
Tymoczko & Goodhart 1986 and Lodders website, below.
Rendell, Paul (2000), This Is a Turing Machine Implemented in Conways Game
of Life, accessed 21 February 2014 from:
http://rendell-attic.org/gol/tm.htm
See also:
Rendell, Paul (2001), A Turing Machine in Conways Game Life, accessed
21 February 2014 from:
https://www.ics.uci.edu/welling/teaching/271fall09/Turing-Machine-
Life.pdf
Rendell, Paul (2010), This Is a Universal Turing Machine (UTM) Implemented in
Conways Game of Life, accessed 21 February 2014 from:
http://rendell-attic.org/gol/utm/
Rey, Georges (2012), The Turing Thesis vs. the Turing Test, The Philosophers
Magazine 57 (2nd Quarter): 8489.
Robinson, J.A. (1994), Logic, Computers, Turing, and von Neumann, in
K. Furukawa, D. Michie, & S. Muggleton (eds.), Machine Intelligence 13: Machine
http://staffweb.worc.ac.uk/DrC/Courses%202013-
14/COMP3202/Reading%20Materials/robinson94.pdf
Reminiscences by one of the leading figures in computation theory.
Rothstein, Edward (2014), Turings Spirit Hovers at a Restored Estate, New York
Times (29 November): C1, C7;
http://www.nytimes.com/2014/11/29/arts/design/where-the-real-imitation-game-
happened.html
(accessed 25 December 2014)
A review of the Bletchley Park museum.
Shagrir, Oron (2006), Godel on Turing on Computability, in Adam Olszewski,
Jan Woenski, & Robert Janusz (eds.), Churchs Thesis after 70 Years (Ontos-
Verlag): 393419;
http://edelstein.huji.ac.il/staff/shagrir/papers/Goedel on Turing on Computability.pdf
Explores why Godel believed both that the correct definition of mechanical
computabilty was established beyond any doubt by Turing [Godel, 930s, 168]
and that this definition. . . rest[s] on the dubious assumption that there is a
finite number of states of mind [Shagrir, 2006a, 1].
Sieg, Wilfried (1994), Mechanical Procedures and Mathematical Experience, in
George Alexander (ed.), Mathematics and Mind (New York: Oxford University
Press): 71117.
3 contains a detailed analysis of Turings work.
Sieg, Wilfried (1997), Step by Recursive Step: Churchs Analysis of Effective
Calculability, Bulletin of Symbolic Logic 3(2) (June): 154180.
. . . Turings analysis is neither concerned with machine computations nor
with general human mental processes. Rather, it is human mechanical
computability that is being analyzed. . . (p. 171).
Smith, Alvy Ray (2014), His Just Deserts: A Review of Four Books, Notices of
the American Mathematical Society 61(8) (September): 891895,
http://www.ams.org/notices/201408/201408-full-issue.pdf (accessed 24 March
2015).
Reviews four biographies of Turing.
Soare, Robert Irving (2013), Turing and the Art of Classical Computability,
in S. Barry Cooper & Jan van Leeuwen (eds.), Alan Turing: His Work and
Impact (Elsevier): 6570.
Preprint accessed 30 April 2014 on complementary, illustrated website, The
Art of Classical Computability,
http://www.people.cs.uchicago.edu/soare/Art/
Stewart, Ian (1994), A Subway Named Turing, Scientific American (September):
104, 106107.
An implementation of a Turing machine using subway trains. See also Hayes
2007, above.
Tymoczko, Thomas; & Goodhart, Sarah (1986), From Logic to Computers: A
Suggestion for Logic Teachers, Teaching Philosophy 9(1) (March): 1533.
Advocates teaching Turing machines in introductory logic courses; see also

Rapaport 1985, above, and Lodders website, below.
Vardi, Moshe Y. (2013), Who Begat Computing?, Communications of the ACM
56(1) (January): 5; accessed 13 November 2013 from
http://cacm.acm.org/magazines/2013/1/158780-who-begat-computing/fulltext
For a reply, see:
Copeland, Jack (2013), What Apple and Microsoft Owe to Turing,
Huff[ington] Post Tech/The Blog (12 August), accessed 13 November 2013
from
http://www.huffingtonpost.com/jack-copeland/what-apple-and-microsoft-
b 374211 4.html
Vogel, Gretchen (2012), Turing Pattern Fingered for Digit Formation, Science
338 (14 December): 1406.
On Turings biological theory of morphogenesis.
Wang, Hao (1957), A Variant to Turings Theory of Computing Machines,
Journal of the ACM 4(1) (January): 6392.
[T]his paper. . . offer[s] a theory which is closely related to Turings but is
more economical in the basic operations.. . . [A] theoretically simple basic
machine can be. . . specified such that all partial recursive functions (and hence
all solvable computation problems) can be computed by it and that only four
basic types of instruction are employed for the programs: shift left one space,
shift right one space, mark a blank space, conditional transfer.. . . [E]rasing is
dispensable, one symbol for marking is sufficient, and one kind of transfer is
enough. The reduction is. . . similar to. . . the definability of conjunction and
implication in terms of negation and disjunction. . . . (p. 63.)
Weizenbaum, Joseph (1976), Computer Power and Human Reason (New York:
W.H. Freeman).
Ch. 2 (Where the Power of the Computer Comes From) contains a masterful
presentation of a Turing Machine implemented with pebbles and toilet paper!
This is also an excellent book on the role of computers in society, by the creator
of the Eliza AI program.
Whitemore, Hugh (1966), Breaking the Code (Samuel French Inc., 2010).
A play, later a film. For more information, see:
http://en.wikipedia.org/wiki/Breaking the Code (accessed 18 February 2014).
The film was accessed 18 February 2014 from:
http://www.youtube.com/watch?v=S23yie-779k
An excerpt of the play was published as:
Whitemore, Hugh (1988), The Enigma: Alan Turing Confronts a Question
of Right and Wrong, The Sciences 28(2) (March/April): 4041.
2. Websites:
(a) The Annotated Turing websites:
http://www.theannotatedturing.com/ (accessed 6 November 2012)
http://www.charlespetzold.com/AnnotatedTuring/ (accessed 6 November
2012)
http://tinyurl.com/AnnotatedTuring (accessed 19 January 2014)34

(b) Copeland, Jack (2012), Is Alan Turing Both Inventor of the Basic Ideas of the
Modern Computer and a Pioneer of Artificial Intelligence?, Big Questions Online
(12 November),
https://www.bigquestionsonline.com/content/alan-turing-both-inventor-basic-
ideas-modern-computer-and-pioneer-artificial-intelligence
(c) Copeland, B. Jack, AlanTuring.net Reference Articles on Turing, accessed
19 January 2014 from35
http://tinyurl.com/lkgqsch
Website that contains a variety of interesting papers on various aspects of
Turings work, most written by Copeland, a well-respected contemporary
philosopher, including:
Copeland, Jack (2000), The Church-Turing Thesis, accessed
http://tinyurl.com/mtch7f6
Copeland, Jack (2000), What Is a Turing Machine?, accessed
http://tinyurl.com/bddfe
(d) Daylight, Edgar G. (2013), Towards a Historical Notion of Turingthe Father of
Computer Science
http://www.dijkstrascry.com/sites/default/files/papers/Daylightpaper91.pdf
(accessed 7 April 2015).
Argues from historical evidence that the reasons that Turing is considered
the father of computer science have nothing to do with his involvement in
computer design or construction.
(e) Lammens, Joe (1990), Universal Program, a Common Lisp implementation of
Turings universal program as specified in [Davis and Weyuker, 1983], accessed
19 February 2014 from:
http://www.cse.buffalo.edu/rapaport/lammens.lisp
(f) Lodder, Jerry, Introducing Logic via Turing Machines, accessed 25 February
2014 from:
http://www.math.nmsu.edu/hist projects/j13.html
See also Rapaport 1985 and Tymoczko & Goodhart 1986, above.
See also other Turing-machine exercises at:
Bezhanishvili, Guram; Leung, Hing; Lodder, Jerry; Pengelley, David; &
Ranjan, Desgh, Teaching Discrete Mathematics via Primary Historical
Sources, accessed online, 25 February 2014, from:
http://www.math.nmsu.edu/hist projects/
and in PDF from:
http://www.math.nmsu.edu/hist projects/DMRG.turing.pdf
34 http://www.amazon.com/gp/reader/0470229055/ref=sib dp pt#reader-link
35 http://www.alanturing.net/turing archive/pages/Reference%20Articles/referencearticlesindex.html
36 http://www.alanturing.net/turing archive/pages/Reference%20Articles/The%20Turing-
Church%20Thesis.html
37 http://www.alanturing.net/turing archive/pages/Reference%20Articles/What%20is%20a%20Turing%20Machine.html
(g) Michaelson, Greg (2012), A Visit to the Turing Machine: A Short Story, CS4FN:
Computer Science for Fun 14 (April),
http://www.cs4fn.org/turing/avisittotheturingmachine.php
(h) Smith, Alvy Ray (2014), Turingtoys.com,
http://alvyray.com/CreativeCommons/TuringToysdotcom.htm
and A Business Card Universal Turing Machine (2 August 2014),
http://alvyray.com/CreativeCommons/BizCardUniversalTuringMachine v2.2.pdf
(both accessed 24 March 2015)
A universal Turing machine on a business card!
For more information, see:
Casselman, Bill (2014), About the Cover: Smart Card, Notices of the
American Mathematical Society 61(8) (September): 872,
http://www.ams.org/notices/201408/201408-full-issue.pdf (accessed
24 March 2015).
(i) Suber, Peter (1997), Turing Machines
http://www.earlham.edu/peters/courses/logsys/turing.htm
A 2-part handout;
click on second hand-out at the end of part I to get to part II:
http://www.earlham.edu/peters/courses/logsys/turing2.htm
Figure 8.3: YEAR

c by ??
Chapter 9
What Is a Computer?
A Philosophical Perspective
Version of 31 August 2015
What is computation? In virtue of what is something a computer? Why

do we say a slide rule is a computer but an egg beater is not? These
are. . . the philosophical questions of computer science, inasmuch as they
query foundational issues that are typicaly glossed over as researchers
get on with their projects. [Churchland and Sejnowski, 1992, 61, italics
added]
. . . everyone who taps at a keyboard, opening a spreadsheet or a word-

processing program, is working on an incarnation of a Turing machine. . .
(Time magazine, 29 March 1999, cited in [Davis, 2006a, 125])
359
360CHAPTER 9. WHAT IS A COMPUTER?A PHILOSOPHICAL PERSPECTIVE
Figure 9.1: 2012

c by ??

1. Required:
(a) Searle, John R. (1990), Is the Brain a Digital Computer?, Proceedings and
Addresses of the American Philosophical Association 64(3) (November): 2137.
i. online version (without the footnotes) accessed 7 November 2012 from:
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/searle.comp.html
ii. A slightly revised version is often cited instead of the 1990 article:
Searle, John (1992), The Rediscovery of the Mind (Cambridge, MA: MIT
Press), Ch. 9.
(b) Hayes, Patrick J. (1997), What Is a Computer? An Electronic Discussion, The
Monist 80(3).
Original emails accessed 7 November 2012 from:
http://www.philo.at/mii/wic.dir9601/maillist.html
(c) Lloyd, Seth; & Ng, Y. Jack (2004), Black Hole Computers, Scientific American
291(5) (November): 5261.
Argues that the universe is a computer.
(d) Weinberg, Steven (2002), Is the Universe a Computer?, New York Review of
Books 49(16) (24 October).
i. This is a review of Wolfram 2002 (see Strongly Recommended, below).
ii. There is a follow-up letter:
Shepard, Harvey; & Weinberg, Steven (2003, 16 January), Is the Universe a
Computer?, New York Review of Books.

(a) Samuel, Arthur L. (1953), Computing Bit by Bit, or Digital Computers Made
Easy, Proceedings of the IRE [Institute of Radio Engineers] 41(10) (October):
12231230.
An introduction to computers, aimed at (radio) engineers who might not be
familiar with them; written by an IBM researcher who later became famous
for his work on computer checkers-players.
(b) Shagrir, Oron (2006), Why We View the Brain as a Computer, Synthese 153(3)
(December): 393416.
1, The Problem of Physical Computation: What Does Distinguish
Computers from Other Physical Systems?, contains a good survey of various
theories of what a computer is.
(c) Piccinini, Gualtiero (2006), Computational Explanation in Neuroscience,
Synthese 153(3) (December): 343353.
14 are a good summary of issues related to the nature of computationalism,
observer-dependence (as opposed to what Searle calls intrinsic computation-
alism), and universal realizability (or pancomputationalism).
(d) Piccinini, Gualtiero (2007), Computational Modelling vs. Computational
Explanation: Is Everything a Turing Machine, and Does It Matter to the Philosophy
of Mind?, Australasian Journal of Philosophy 85(1): 93115
http://www.umsl.edu/piccininig/Is Everything a Turing machine.pdf
(a) 1 of:
Rapaport, William J. (1998), How Minds Can Be Computational Systems,
Journal of Experimental and Theoretical Artificial Intelligence 10: 403419.
i. 1 is relevant to Searle 1990; it presents my understanding of what it means to
say that cognition is computable (as opposed to computational).
ii. 23 criticize two other theories of computationalism; you can just skim
these for now.
(b) Wolfram, Stephen (2002), Introduction to A New Kind of Science (Wolfram
Media)
http://www.stephenwolfram.com/publications/recent/nksmain/
(c) Piccinini, Gualtiero (2010), The Mind as Neural Software? Understanding

Functionalism, Computationalism, and Computational Functionalism, Philosophy
and Phenomenological Research 81(2) (September); 269311
Preprint accessed online 29 January 2014 from:
http://www.umsl.edu/piccininig/Computational Functionalism.htm
(d) Powell, Corey S. (2006), Welcome to the Machine, New York Times Book Review
(2 April): 19.

http://www.nytimes.com/2006/04/02/books/review/02powell.html
A review of Seth 2006 (see Further Sources of Information, below)
(e) Schmidhuber, Jurgen (2006), The Computational Universe, American Scientist
94 (July-August): 364ff.
http://www.americanscientist.org/bookshelf/pub/the-computational-universe
Another review of Seth 2006 (see Further Sources of Information, below)
(f) Hayes, Brian (2007), Trains of Thought, American Scientist 95(2) (March-April):
108113.
Can a railroad layout be a computer?
(g) Rapaport, William J. (2007), Searle on Brains as Computers, American
Philosophical Association Newsletter on Philosophy and Computers 6(2) (Spring):
49.
i. Accessed 11 March 2014 from:
http://c.ymcdn.com/sites/www.apaonline.org/resource/collection/EADE8D52-
8D02-4136-9A2A-729368501E43/v06n2Computers.pdf
ii. A direct reply to Searle 1990.
Figure 9.2: 2013,

c Creators Syndicate
9.2 Introduction
We began our study of the philosophy of computer science by asking what computer
science is. We then asked what it studies: Does it study computers? Or computing? Or
maybe something else? In Chapter 6, we began our investigation into what a computer
is, from a historical perspective. And, in Chapters 7 and 8, we began our investigation
into what computing is.
In the present chapter, armed with the results of these investigations, we return to
our earlier question: If computer science is the study of computers, what is a computer?
At the very least, we might say that a computer is a machine designed (that is,
engineered) to compute, to do computations. But does it have to be a machine?
Does it have to be engineered? Is the brain a computer? If so, then it would seem
that computers could be biological entities (which, arguably, are not machines) that
evolved (which, arguably, means that they were not engineered).
And what kind of computations do computers perform? Are they restricted to
mathematical computations? Even if thats so, is that a restriction? The binary-
representation insight (7.6.1 suggests that any (computable) information can be
represented as a binary numeral; hence, any computation on such information could
be considered to be a mathematical computation.
And what about the difference between a hardwired Turing machine that can
only compute one thing and a universal machine that can compute anything that
is computable? Or the difference between a real, physical computer that can only
compute whatever is practically computable (that is, subject to reasonable space and
time constraints) and an abstract, universal Turing machine that is not thus constrained?
In other words, what is a computer?
If you ask a random person what a computer is, they will probably try to describe
their laptop. If you look up computer in a reference book, you will find things like
this (from the Oxford English Dictionary):
computer, n.
1. A person who makes calculations or computations; a calculator, a
reckoner; spec[ifically,] a person employed to make calculations in
an observatory, in surveying, etc. Now chiefly hist[orical]
[earliest citation dated 1613]
2. A device or machine for performing or facilitating calculation.
3. An electronic device (or system of devices) which is used to
store, manipulate, and communicate information, perform complex
calculations, or control or regulate other devices or machines, and
is capable of receiving information (data) and of processing it
in accordance with variable procedural instructions (programs or
software); esp[ecially] a small, self-contained one for individual use
in the home or workplace, used esp. for handling text, images, music,
and video, accessing and using the Internet, communicating with
other people (e.g. by means of email), and playing games.
(OED, http://www.oed.com/view/Entry/37975)
Or this (from the Encyclopedia of Computer Science):
A digital computer is a machine that will accept data and information

presented to it in a discrete form, carry out arithmetic and logical
operations on this data, and then supply the required results in an
acceptable form. [Morris and Reilly, 2000, 539]
One common thread in such definitions (ignoring the ones that are only of historical
interest) is that computers are:
1. devices or machines
2. that take input (data, information),
3. process it (manipulate it; or operate, calculate, or compute with it)
4. in accordance with instructions (a program),
5. and then output a result (presumably, more data or information, but also
including control of another device).
There are some other features that are usually associated with computers: The
kind that we are interested in must be, or typically are:
automatic
There is no human intervention (beyond, perhaps, writing the program). Of
course, the holy grail of programming is to have self-programmed computers,
possibly to include having the desire or intention to program themselves (as
in science fiction). Humans might also supply the input or read the output, but
that hardly qualifies as intervention. (We will explore interventionin the
guise of interactive computingin Ch. 11.)
general purpose
A computer must be capable of any processing that is algorithmic, by means
of a suitable program. This is the heart of Turings universal machine. Recall
that a Turing machine runs only one program. The universal Turing machine
is also a Turing machine, so it, too, also runs only one program, namely, the
fetch-execute cycle that simulates the running of another (that is, any other)
single-program Turing machine.
electronic
Modern computers are, as a matter of fact, electronic, but there is
work on quantum computers ([Grover, 1999, Aaronson, 2008, Aaronson, 2011,
Aaronson, 2014, Hayes, 2014b], and other references cited in 31), optical
computers CITE, biological computers [Adleman, 1998], etc. Being electronic
is not essential. The crucial property is, rather, to be constructed in such a way
as to allow for high processing speeds or other kinds of efficiencies.
digital
They should process information expressed in discrete, symbolic (typically,
alpha-numeric) form, but perhaps also including graphical form. The
contrast is typically with being analog, where information is represented by
means of continuous physical quantities (see [Jackson, 1960], [Pour-El, 1974],
[Holst, 2000], [Piccinini, 2008], [Piccinini, 2011] [McMillan, 2013]).
What about the calculations, the arithmetic and logical operations?

Presumably, these need to be algorithmic, though neither the OED nor the
Encyclopedia of Computer Science definitions say so. And it would seem that the
authors of those definitions have in mind calculations or operations such as addition,
subtraction, etc.; maybe solving differential equations; Boolean operations involving
conjunction, disjunction, etc; and so on. These require the data to be numeric (for
math) or propositional (or truth-functionalfor Boolean and logical operations), at
least in some ultimate sense: That is, any other data (pictorial, etc.) must be encoded
as numeric or propositional, or else would need to allow for other kinds of operations.
There certainly seem to be clear cases of digital computers: Macs, PCs, etc. There
certainly seem to be clear cases of analog computersslide rules, certain machines
at various universities (Indiana, Harvard)1 though these may be mostly of historical
interest, dont seem to be programmable (i.e., universal, in Turings sense), and seem
to be outside the historical development explored in Chapter 6.
And there seem to be clear cases of entities that are not computers: I would guess
that most people would not consider rocks, walls, ice cubes, or solid blocks of plastic
to be computers (note that I said most people!). And there are even clear cases of
1 For Indiana, see [Hedger, 1998]s report on the research of Jonathan W. Mills; for Harvard, see Harvard
Robotics Laboratory: Analog Computation, accessed 31 January 2014 from http://hrl.harvard.edu/analog/.

devices for which it might be said that it is not clear whether, or in what sense, they are
computers, such as Atanasoff and Berrys ABC.
So: What is a computer? What is the relation of a computer to a Turing machine and
to a univeral Turing machine? Is the (human) brain a computer? Is your smartphone
a computer? Could a rock or a wall be considered to be a computer? Might anything
be a computer? Might everythingsuch as the universe itselfbe a computer? Or are
some of these just badly formed questions?
9.3 John Searle: Anything Is a Computer

John Searles presidential address to the American Philosophical Association, Is the
Brain a Digital Computer? [Searle, 1990], covers a lot of ground and makes a lot of
points about the nature of computers, the nature of the brain, the nature of cognition,
and the relationships among them. In this section, we are going to focus only on what
Searle says about the nature of computers, with only a few side glances at the other
issues.2
Here is Searles argument relevant to our main question about what a computer is:
1. Computers are described in terms of 0s and 1s.
(This is what he says. Taken literally, he is sahying that computers are described
in terms of certain numbers. Instead, he might have said that computers are
described in terms of 0s and 1s. In other words, he might have said that
computers are described in terms of certain numerals. Keep this distinction in
mind as we discuss Searles argument!)3
2. Therefore, being a computer is a syntactic property.
(Syntax is the study of the properties of, and relations among, symbols or
uninterpreted marks on paper (or on some other medium); a rough synonym
is symbol manipulation.4
3. Therefore, being a computer is not an intrinsic property of physical objects.
4. Therefore, we can ascribe the property of being a computer to any object.
5. Therefore, everything is a computer.
Of course, this doesnt quite answer our question, What is a computer?. Rather,
the interpretation and truth value of these theses will depend on what Searle thinks a
computer is. Lets look at exactly what Searle says about these claims.
2 For more detailed critiques and other relevant commentary, see [Piccinini, 2006b, Piccinini, 2007b,
Piccinini, 2010a, Rapaport, 2007].
3 Recall the difference between numbers and numerals from 2.3.
4 We discuss the nature of syntax and its relationship to semantics in Ch. 22.
9.3. JOHN SEARLE: ANYTHING IS A COMPUTER 367
9.3.1 Computers Are Described in Terms of 0s and 1s

First, after briefly describing Turing machines as devices that can perform the actions
of printing 0 or 1 on a tape and of moving left or right on the tape, depending on
conditions specified in its program, he says this:
If you open up your home computer you are most unlikely to find any 0s
and 1s or even a tape. But this does not really matter for the definition. To
find out if an object is really a digital computer, it turns out that we do not
actually have to look for 0s and 1s, etc.; rather we just have to look for
something that we could treat as or count as or could be used to function
as 0s and 1s. (p. 25.)
So, according to Searle, a computer is a physical object that can be described as a

Turing machine. Recall from 8.8 that anything that satisfies the definition of a Turing
machine is a Turing machine, whether it has a paper tape divided into squares with the
symbols 0 or 1 printed on them or whether it is a table and chairs with beer mugs
on them. All we need is to be able to treat some part of the physical object as playing
the role of the Turing machines 0s and 1s. So far, so good.
Or is it? Is your home computer really a Turing machine? Or is it a device whose
behavior is merely logically equivalent to that of a Turing machine? That is, is it a
device that can compute all and only the functions that a Turing machine can compute,
even if it does so differently from the way that a Turing machine does? Recall that
there are lots of different mathematical models of computation: Turing machines and
recursive functions are two of them that we have looked at. Suppose someone builds a
computer that operates in terms of recursive functions instead of in terms of a Turing
machine. That is, it can compute successors, predecessors, and projection functions,
and it can combine these using generalized composition, conditional definition, and
while-recursion, instead of printing 0s and 1s, moving left and right, and combining
these using go to instructions (changing from one m-configuration to another). These
two computers (the Turing-machine computer and the recursive-function computer), as
well as your home computer (with a von Neumann architecture, whose method of
computation uses the primitive machine-language instructions and control structures
of, say, an Intel chip), are all logically equivalent to a Turing machine, but their
internal behaviors are radically different. Can we really describe the recursive-function
computer and your home computer in terms of a Turing-machines 0s and 1s? Or
are we limited to showing that anything that the recursive-function computer and your
home computer can compute can also be computed by a Turing machine (and vice
versa)?
Here is an analogy to help you see the issue: Consider translating between French
and English. To say It is snowing in French, you say: Il neige. The il means it,
and the neige means is snowing. This is very much like (perhaps it is exactly
like) describing the recursive-function machines behavior (analogous to the French
sentence) using 0s and 1s (analogous to the English sentence). But here is a different
example: In English, if someone says: Thank you, you might say, Youre welcome,
but, in French, if someone says Merci, you might reply: Je vous en prie. Have we
translated the English into French, in the way that we might translate a recursive-
function algorithm into a Turing machines 0s and 1s? Not really: Although merci
is used in much the same way in French that thank you is used in English, there is
no part of merci that means thank or you; and the literal translation of je vous
en prie is something like I pray that of you. So there is a way of communicating
the same information in both French and English, but the phrases used are not literally
inter-translatable.
So, something might be a computer without being described in terms of 0s and
1s, depending on exactly what you mean by described in terms of. But lets
grant Searle the benefit of the doubt (perhaps he should have said something like this:
computers are described in terms of the primitive elements of the mathematical model
of computation that they implement), and continue looking at his argument.
9.3.2 Being a Computer Is a Syntactic Property

Let us suppose, for the sake of the argument, that computers are described in terms
of 0s and 1s. Such a description is syntactic. This term (which pertains to
symbols, words, grammar, etc.) is usually contrasted with semantic (which pertains
to meaning), and Searle emphasizes that contrast early in his essay when he says that
syntax is not the same as, nor is it by itself sufficient for, semantics [Searle, 1990,
21]. But here Searle uses the term syntactic as a contrast to being physical. Just as
there are many ways to be computable (Turing machines, recursive functions, lambda-
calculus, etc.)all of which are equivalentso there are many ways to be a carburetor.
A carburetor. . . is a device that blends air and fuel for an internal combustion engine
(http://en.wikipedia.org/wiki/Carburetor), but it doesnt matter what it is made of , as
long as it can perform that blending function (purpose). [C]arburetors can be made
of brass or steel [Searle, 1990, 26]; they are multiply realizablethat is, you can
realize (or make) one in multiple (or different) ways. They are defined in terms
of the production of certain physical effects [Searle, 1990, 26].
But the class of computers is defined syntactically in terms of the
assignment of 0s and 1s. [Searle, 1990, 26, Searles italics, my boldface]
In other words, if something is defined in terms of symbols, like 0s and 1s, then it is
defined in terms of syntax, not in terms of what it is physically made of.
Hence, being a computer is a syntactic property, not a physical property. It is a
property that something has in virtue of. . . of what? There are two possibilities, given
what Searle has said. First, perhaps being a computer is a property that something has
in virtue of what it does, its function or purpose. Second, perhaps being a computer
is a property that something has in virtue of what someone says that it does, how it
is described. But what something actually does may be different from what someone
says that it does.
So, does Searle think that something is a computer in virtue of its function or in
virtue of its syntax? Suppose you find a black box (with a keyboard and a screen)
in the desert and that, by experimenting with it, you determine that it displays on its
screen the greatest common divisor (GCD) of two numbers that you type into it. (We
discussed a similar thought experiment in 3.8; see [Weizenbaum, 1976, Ch. 5] for
further discussion). It certainly seems to function as a computer (as a Turing machine
for computing GCDs). And you can probably describe it in terms of 0s and 1s,
so you can also say that it is a computer. It seems that if something functions as a
computer, you can describe it in terms of 0s and 1s.
What about the converse? Suppose that the black boxs behavior is inscrutable;
the symbols on the keys are unrecognizable, and the symbols displayed on the screen
dont seem to be related in any obvious way to the input symbols. But suppose that
someone manages to invent an interpretation of the symbols in terms of which the boxs
behavior can be described as computing GCDs. Is that really what it does? Might it not
have been created by some extraterrestrials solely for the purpose of entertaining their
young with displays of meaningless symbols, and that it is only by the most convoluted
(and maybe not always successful) interpretation that it can be described as computing
GCDs?
You might think that the boxs function is more important for determining what it
is. Searle thinks that our ability to describe it syntactically is more important! After all,
whether or not the box was intended by its creators to compute GCDs or to entertain
toddlers, if it can be accurately described as computing GCDs, then, in fact, it computes
GCDs (as well as, perhaps, entertaining toddlers with pretty pictures).
Again, lets grant this point to Searle. He then goes on to warn us:
But this has two consequences which might be disastrous:
1. The same principle that implies multiple realizability would seem to
imply universal realizability. If computation is defined in terms of the
assignment of syntax then everything would be a digital computer,
because any object whatever could have syntactical ascriptions made
to it. You could describe anything in terms of 0s and 1s.
2. Worse yet, syntax is not intrinsic to physics. The ascription of
syntactical properties is always relative to an agent or observer who
treats certain physical phenomena as syntactical. [Searle, 1990, 26]
Lets take these in reverse order.
9.3.3 Being a Computer Is Not an Intrinsic Property

of Physical Objects
According to Searle, being a computer is not an intrinsic property of physical objects,
because being a computer is a syntactic property, and syntax is not intrinsic to
physics. What does that quoted thesis mean, and why does Searle think that it is
true?
What is an intrinsic property? 5 Searle doesnt tell us, though he gives some
examples:
[G]reen leaves intrinsically perform photosynthesis[;]. . . hearts intrin-
sically pump blood. It is not a matter of us arbitrarily or conventionally
5 For more discussion on what intrinsic means, see [Skow, 2007] and [Bader, 2013] (the latter is a highly
technical essay, but it contains useful references to the literature on intrinsic properties). ALSO CITE
LEWIS ET AL.
assigning the word pump to hearts or photosynthesis to leaves. There

is an actual fact of the matter. [Searle, 1990, 26]
So, perhaps intrinsic properties are properties that something really has as opposed
to merely being said to have, much the way our black box in the previous section may
or may not really compute GCDs but can be said to compute them. But what does it
mean to really have a property?
Here are some possible meanings for intrinsic:
An object might have a property P intrinsically if it has P essentially rather

than accidentally. An accidental property is a property that something has such
that, if the object lacked that property, it would still be the same object. So, it is
merely an accidental property of me that I was wearing a tan shirt on the day that
I wrote this sentence. If I lacked that property, I would still be the same person.
An essential property is a property that something has such that, if the object
lacked that property, it would be a different object. So, it is an essential property
of me that I am a human being. If I lacked that property, I wouldnt even be a
person at all. (This is the plot of Franz Kafkas novel The Metamorphosis, in
which the protagonist awakes one day to find that he is no longer a human, but
a beetle.) The exact nature of the essential-accidental distinction, and its truth or
falsity, are matters of great dispute in philosophy. Here, I am merely suggesting
that perhaps this is what Searle means by intrinsic: Perhaps he is saying that
being a computer is not an essential property of an object, but only an accidental
property.
An object might have P intrinsically if it has P as a kind of first-order

property, not as a kind of second-order property. A second-order property
is a property that something has in virtue of having some other (or first-order)
property. Here, I am thinking of a property like an objects color. An apple has
the property of being red, not intrinsically, but in virtue of reflecting light with
wavelength approximately 650 nm. We perceive such reflected light in a certain
way, which we call red. But, conceivably, someone with a different neural
make-up (say, red-green color-blindness) might perceive it either as a shade of
gray that is indistinguishable from green or, in a science-fiction kind of case
called an inverted spectrum, as green. The point is that it is the measurable
wavelength of reflected light that might be an intrinsic property belonging to
the apple, whereas the perceived color is a property belonging to the perceiver
and only secondarily to the apple itself. So, perhaps Searle is saying that being
a computer is only a second-order property of an object.
Another way that an object might have P intrinsically (perhaps this is closer
to what Searle has in mind) is if having P is a natural kind. This is
another controversial notion in philosophy. The idea, roughly, is that a natural
kind is a property that something has as a part of nature and not as a
result of what an observer thinks.6 So, being a bear is a natural kind; there
6 Youll note that several of these concepts are closely related; they may even be indistinguishable. For
instance, perhaps essential properties are natural kinds or perhaps second-order properties are not natural
would be bears even if there were no people to see them or to call them
bears. This does not mean that it is easy to define such natural kinds. Is a
bear A heavily-built, thick-furred plantigrade quadruped, of the genus Ursus;
belonging to the Carnivora, but having teeth partly adapted to a vegetable diet
(http://www.oed.com/view/Entry/16537)that is, a heavily-built, thick-furred
mammal that walks on the soles of its four feet, eats meat, but can also eat
plants? What if a bear was born with only three feet, or never eats meat? Is
an animal a bear whether or not it satisfies such definitions, and whether or not
there were any humans to try to give such definitions? If so, then being a bear is
a natural kind and, perhaps, an intrinsic property. As we saw in Chapter 3.4,
Plato once said that it would be nice if we could carve nature into its joints,
that is, find the real, perhaps intrinsic, natural properties of things (Phaedrus
265de). But perhaps the best we can do is to carve joints into nature, that is, to
overlay categories onto nature so that we can get some kind of understanding
of it and control over it, even if those categories dont really exist in nature. Is
being a computer a natural kind? Well, its certainly not like being a bear! There
probably arent any computers in nature (unless the brain is a computer; and
see 9.7.2 on whether nature itself is a computer), but there may also not be any
prime numbers in nature, yet mathematical objects are something thought to exist
independently of human thought. If they do, then, because a Turing machine is a
mathematical object, it might exist independently of human thought and, hence,
being a computer might be able to be considered to be a natural mathematical
kind. So, perhaps Searle is saying that being a computer is not a natural kind in
one of these senses.
In fact, a computer is probably not a natural kind for a different reason: It

is an artifact, something created by humans. Again, the nature of artifacts is
controversial: Clearly, chairs, tables, houses, atomic bombs, and pencils are
artifacts; you dont find them in nature, and if humans had never evolved,
there probably wouldnt be any of these artifacts. But what about birds
nests, beehives, beaver dams, and other such things constructed by non-human
animals? What about socially constructed objects like money? One of the
crucial features of artifacts is that what they are is relative to what a person says
they are. You wont find a table occurring naturally in a forest, but if you find
a tree stump, you might use it as a table. So, something might be a computer,
Searle might say, only if a human uses it that way or can describe it as one.
In fact, Searle says this explicitly:
[W]e might discover in nature objects which had the same sort of
shape as chairs and which could therefore be used as chairs; but we
could not discover objects in nature which were functioning as chairs,
except relative to some agents who regarded them or used them as
chairs. [Searle, 1990, 28].
kinds. Investigating these relationships is one of the tasks of metaphysics. It is beyond the scope of the
philosophy of computer science.
Why is syntax not intrinsic to physics? Because syntax is not the name of
a physical feature, like mass or gravity.. . . [S]yntax is essentially an observer relative
notion [Searle, 1990, 27]. I think that what Searle is saying here is that we can analyze
physical objects in different ways, no one of which is privileged or more correct;
that is, we can carve nature into different joints, in different ways. On some such
carvings, we may count an object as a computer; on others, we wouldnt. By contrast,
an object has mass independently of how it is described: Having mass is not relative to
an observer. How its mass is measured is relative to an observer.
But couldnt being a computer be something like that? There may be lots of
different ways to measure mass, but an object always has a certain quantity of mass,
no matter whether you measure it in grams or in some other units. In the same way,
there may be lots of different ways to measure length, but an object always has a certain
length, whether you measure it in centimeters or in inches. Similarly, an object (natural
or artifactual) will have a certain structure, whether you describe it as a computer or as
something else. If that structure satisfies the definition of a Turing machine, then it is a
Turing machine, no matter how anyone describes it.
Searle anticipates this reply:
[S]omeone might claim that the notions of syntax and symbols are
just a manner of speaking and that what we are really interested in
is the existence of systems with discrete physical phenomena and state
transitions between them. On this view we dont really need 0s and 1s;
they are just a convenient shorthand. [Searle, 1990, 27]
Compare this to my example above: Someone might claim that specific units of
measurement are just a manner of speaking and that what we are really interested in is
the actual length of an object; on this view, we dont really need centimeters or inches;
they are just a convenient shorthand.
Searle replies:
But, I believe, this move is no help. A physical state of a system is

a computational state only relative to the assignment to that state
of some computational role, function, or interpretation. The same
problem arises without 0s and 1s because notions such as computation,
algorithm and program do not name intrinsic physical features of systems.
Computational states are not discovered within the physics, they are
assigned to the physics. [Searle, 1990, 27, my boldface, Searles italics.]
But this just repeats his earlier claim; it gives no new reason to believe it. He continues
to insist that being a computer is more like inches than like length.
So, we must ask again: Why is syntax not intrinsic to physics? Perhaps, if a
property is intrinsic to some object, then that object can only have the property in
one way. For instance, color is presumably not intrinsic to an object, because an object
might have different colors depending on the conditions under which it is perceived.
But the physical structure of an object that causes it to reflect a certain wavelength
of light is always the same; that physical structure is intrinsic. On this view, here
is a reason why syntax might not be intrinsic: The syntax of an object is, roughly,
its abstract structure (see 14.4 and 19.5.3.2). But an object might be able to be
understood in terms of several different abstract structures (and this might be the case
whether or not human observers assign those structures to the object). If an object has
no unique syntactic structure, then syntax is not intrinsic to it. But if an object has (or
can be assigned) a syntax of a certain kind, then it does have that syntax even if it also
has another one. And if, under one of those syntaxes, the object is a computer, then it
is a computer.
But that leads to Searles next point.
9.3.4 We Can Ascribe the Property of Being a Computer

to Any Object
There is some slippage in the move from syntax is not intrinsic to physics to we can
ascribe the property of being a computer to any object. Even if syntax is not intrinsic
to the physical structure of an object, perhaps because a given object might have several
different syntactic structures, why must it be the case that any object can be ascribed
the syntax of being a computer?
One reason might be this: Every object has (or can be ascribed) every syntax. That
seems to be a very strong claim. To refute it, however, all we would need to do is to find
an object O and a syntax S such that O lacks (or cannot be ascribed) S. One possible
place to look would be for an O whose size in some sense is smaller than the size
of some S. I will leave this as an exercise for the reader: If you can find such O and S,
then I think you can block Searles argument at this point.
Here is another reason why any object might be able to be ascribed the syntax of
being a computer: There is something special about the syntax of being a computer
that is, about the formal structure of Turing machinesthat does allow it to be ascribed
to (or found in) any object. This may be a bit more plausible than the previous reason.
After all, Turing machines are fairly simple. Again, to refute it, we would need to find
an object O such that O lacks (or cannot be ascribed) the syntax of a Turing machine.
Again, I will leave this as an exercise for the reader, but we will return to it later (in
Ch. 14). Searle thinks that we cannot find such an object.
9.3.5 Everything Is a Computer

We can ascribe the property of being a computer to any object if and only if everything
is a computer.
Thus for example the wall behind my back is right now implementing the
Wordstar program, because there is some pattern of molecule movements
which is isomorphic with the formal structure of Wordstar. [Searle, 1990,
27]
Searle does not offer a detailed argument for how this might be the case, but other
philosophers have done so, and we will explore how they think it can be done when we
look at the nature of implementation (Ch. 14). Lets assume, for the moment, that it
can be done.
In that case, things are not good, because this trivializes the notion of being a
computer. If everything has some property P, then P isnt a very interesting property;
P doesnt help us categorize the world, so it doesnt help us understand the world.
How might we respond to this situation? One way is to bite the bullet and accept
that, under some description, any object (even the wall behind me) can be considered
to be a computer. And not just some specific computer, such as a Turing machine that
executes the Wordstar program:
[I]f the wall is implementing Wordstar then if it is a big enough wall it
is implementing any program, including any program implemented in the
brain. [Searle, 1990, 27]
If a big enough wall implements any program, then it implements the universal Turing
machine!
But perhaps this is OK. After all, there is a difference between an intended
interpretation of something and what I will call a gerrymandered interpretation. For
instance, the intended interpretation of Peanos axioms for the natural numbers is the
sequence h1, 2, 3, . . .i. There are also many other natural interpretations, such as
hI, II, III,. . .i. But extremely contorted ones, such as a sequence of all objects in the
universe arranged alphabetically by their name or description in English, are hardly
good examples. Admittedly, they are examples of natural numbers, but not very
useful ones.
A better reply to Searle, however, is to say that hes wrong: Some things are not
computers. Despite what he said in the last passage quoted above, the wall behind me
is not a universal Turing machine; I really cannot use it to post to my Facebook account
or even to write a letter. It is an empirical question whether something actually behaves
as a computer. And the same goes for other syntactic structures. Consider the formal
definition of a mathematical group (a set of objects [like integers] that is closed under
an associative, binary operation [like addition] that has an identity element [like 0] and
is such that every element of the set has an inverse [in the case of integer n, its negative
n]). Not every set is a group. Similarly, there is no reason to believe that everything
is a Turing machine.
9.3.6 Summary
So, Searle believes that a computer is anything that is a Turing machine, and that
everything is a Turing machine. I think that he is wrong about the second claim. But
is it reasonable to answer the question, What is a computer?, by saying that it is a
Turing machine?
Lets take stock of where we are. Presumably, computers are things that compute.
Computing is the process that Turing machines give a precise description for. That
is, computing is what Turing machines do. And, what Turing machines do is to move
around in a discrete fashion and print discrete marks on discrete sections of the space in
which they move around. So, a computer is a devicepresumably, a physical device
that does that. Searle agrees that computing is what Turing machines do, and he seems
to agree that computers are devices that compute. He also believes that everything is a
computer; more precisely, he believes that everything can be described as a computer
9.4. PATRICK HAYES: COMPUTERS AS MAGIC PAPER 375
(because thats what it means to be a computer). And weve also seen reason to think
that he might be wrong about that last point.
In the next two sections, we look at two other views about what a computer is.
9.4 Patrick Hayes: Computers as Magic Paper

Lets keep straight about three intertwined issues that we have been looking at:
1. What is a computer?
2. Is everything a computer?
3. Is the brain a computer?
Our principal concern is with the first question. Once we have an answer to that, we
can try to answer the others. As weve just seen, Searle thinks that a computer is a
Turing machine, that everything is (or can be correctly described as) a computer, and,
therefore, that the brain is (trivially) a computer.
The AI researcher Patrick J. Hayes [Hayes, 1997] gives two definitions. Heres the
first one:
Definition H1
By computer I mean a machine which performs computations, or which
computes. [Hayes, 1997, 390, my boldface.]
A full understanding of this requires a definition of computation; this will be clarified

in his second definition. But there are a few points to note about this first one.
He prefaces it by saying:
First, I take it as simply obvious both that computers exist and that not
everything is a computer, so that, contra Searle, the concept of computer
is not vacuous. [Hayes, 1997, 390].
So, there are machines that compute (that is, there are things that are machines-that-
compute), and there are things that are not machines-that-compute. Note that the latter
can be true in two ways: There might be machines that dont compute, or there might
be things that do compute but that arent machines. Searle disputes the first possibility,
because he thinks that everything (including, therefore, any machine) computes. But
contrary to what Hayes says, Searle would probably agree with the second possibility,
because, after all, he thinks that everything (including, therefore, anything that is not
a machine) computes! Searles example of the wall that implements (or that can be
interpreted as implementing) Wordstar would be such a non-machine that computes.
So, for Hayess notion to contradict Searle, it must be that Hayes believes that there
are machines that do not compute. Perhaps, that wall is one of them, or perhaps a
dishwasher is a machine that doesnt compute anything.7
7 A dishwasher might, however, be described by a (non-computable) function that takes dirty dishes as
input and that returns clean ones as output.

Are these two obvious points to be understood as axiomsor criteria of adequacy

for any definitionthat Hayes thinks need no argument? Or are they intended to
be theorems that follow from his first definition? If its the former, then there is no
interesting debate between Searle and Hayes; one simply denies what the other argues
for. If its the latter, then Hayes needs to provide arguments or examples to support his
position.
A second thing to note about Hayess definition is that he says that computers
perform computations, not can perform computations. Strictly speaking, your
laptop when it is turned off is not a computer by this definition, because it is not
performing any computation. And, as Hayes observes,
On this understanding, a Turing machine is not a computer, but a
mathematical abstraction of a certain kind of computer. [Hayes, 1997,
390]
What about Searles wall that implements Wordstar? There are two ways to think about
how the wall might implement Wordstar. First, it might do so statically, simply in virtue
of there being a way to map every part of the Wordstar program to some aspect of the
molecular or subatomic structure of the wall. In that case, Hayes could well argue
that the wall is not a Wordstar computer, because it is not computing (even if it might
be able to). But the wall might implement Wordstar dynamically; in fact, that is why
Searle thinks that the wall implements Wordstar:
[B]ecause there is some pattern of molecule movements which is
isomorphic with the formal structure of Wordstar. [Searle, 1990, 27, my
italics]
But a pattern of movements suggests that Searle thinks that the wall is computing, so it
is a computer!
Definition H2 is a bit more precise, and it is, presumably, his official one:
Definition H2
[F]ocus on the memory. A computers memory contains patterns. . . which are
stable but labile [i.e., changeable], and it has the rather special property that
changes to the patterns are under the control of other patterns: that is, some of
them describe changes to be made to others; and when they do, the memory
changes those patterns in the way described by the first ones.. . . A computer is
a machine which is so constructed that patterns can be put in it, and when
they are, the changes they describe will in fact occur to them. If it were paper,
it would be magic paper on which writing might spontaneously change, or new
writing appear. [Hayes, 1997, 393, my boldface]
There is a subtle difference between Hayess two definitions, which highlights an

ambiguity in Searles presentation. Recall the distinction between a Turing machine
and a universal Turing machine. A Turing machine computes one function. A universal
Turing machine can compute any computable function as long as a Turing machine-
program for that function is encoded on the universal Turing machines tape. In short, a
Turing machine is hardwired computer; a universal Turing machine is a programmable
9.4. PATRICK HAYES: COMPUTERS AS MAGIC PAPER 377
computer. Definition H1 seems to include physical Turing machines (but, as he noted,

not abstract ones), because, after all, they compute (at least, when they are turned on
and running). Definition H2 seems to exclude them, because the second definition
requires patterns that describe changes to other patterns. That first kind of pattern is a
program; the second kind is the data that the program operates on. So, Definition H2
is for a universal Turing machine.
Is Searles wall a Turing machine or a universal Turing machine? On Searles view,
Wordstar is a Turing machine, so the wall must be a Turing machine, too. So, the wall is
not a computer on Definition H2. Could a wall (or a rock, or some other suitably large
or complex physical object other than something like a PC or a Mac) be a universal
Turing machine? My guess is that Searle would say yes, but it is hard to see how one
would actually go about programming it.
The magic paper aspect of Definition H2 focuses, as he notes, on the memory,
that is, on the tape. It is as if you were looking at a universal Turing machine, but
all you saw was the tape, not the read-write head or its states (m-configurations) or its
mechanism. If you watch the universal Turing machine compute, you would see the
patterns (the 0s and 1s) on the tape magically change.
The idea of a computer as magic paper may seem a bit fantastic. But there are
more down-to-earth ways of thinking about this. The philosopher Richmond Thomason
has said that
. . . all that a program can do between receiving an input and producing an
output is to change variable assignments. . . [Thomason, 2003, 328].8
If programs tell a computer how to change the assignments of values to variables, then
a computer is a (physical) device that changes the contents of register cells (the register
cells that are the physical implementations of the variables in the program). This is
really just another version of Turings machines, if you consider the tape squares to be
the register cells.
Similarly, Stuart C. Shapiro points out that a computer is a device consisting
of a vast number of connected switches. . . . [T]he switch settings both determine
the operation of the device and can be changed by the operation of the device
[Shapiro, 2001, 3].
What is a switch? Here is a nice description from a 1953 article introducing
computers to engineers who might never have encountered them before:
To bring the discussion down to earth let us consider the ordinary electric
light switch in your home. This is by definition a switch. It enables one to
direct electric current to a lighting fixture at will Usually there is a detent
mechanism which enables the switch to remember what it is supposed
to be doing so that once you turn the lights on they will remain on. It
therefore has a memory. It is also a binary, or perhaps we should say
a bistable device. By way of contrast, the ordinary telegraphers key is
a switch without memory since the key will remain down only as long
as it is depressed by the operators hand. But the light switch and the
8 A similar point is made in [Lamport, 2011, 6]: an execution of an algoritihm is a sequence of states,
where a state is an assignment of values to variables.

telegraph key are binary devices, that is, they have but two operating states.
[Samuel, 1953, 1225]
So, a switch is a physical implementation of a Turing-machines tape cell, which can

also be in two states (that is, have one of two symbols printed on it) and also has
a memory (that is, once a symbol is printed on a cell, it remains there until it is
changed).
Hayess magic-paper patterns are just Shapiros switch-settings or Thomasons
variable assignments. (For more on computers as switch-setting devices, see the
discussions of how train switches can implement computations in [Stewart, 1994] and
[Hayes, 2007b]; both of these are also examples of Turing machines implemented in
very different media than silicon (namely, trains)!)
Does this definition satisfy Hayess two criteria? Surely, such machines exist. I
am writing this book on one of them right now. And surely not everything is such a
machine: At least on the face of it, the stapler on my desk is not such magic paper.
Searle, I would imagine, would say that we might see it as such magic paper if we
looked at it closely enough and in just the right way. And so the difference between
Searle and Hayes seems to be in how one is supposed to look at candidates for being
a computer: Do we look at them as we normally do? In that case, not everything is a
computer. Or do we look at them closely in a certain way? In that case, perhaps we
could see that everything could be considered to be a computer. Isnt that a rather odd
way of thinking about things?
Is the brain a computer in the sense of magic paper? If Hayess patterns are
understood as patterns of neuron firings, then, because surely some patterns of neuron
firings cause changes in other such patterns, I think Hayes would consider the brain to
be a computer.
9.5 Gualtiero Piccinini:

Computers as Digital String Manipulators
In a series of three papers, the philosopher Gualtiero Piccinini has offered an analysis
of what a computer is that is more precise than Hayess and less universal than
Searles [Piccinini, 2007b], [Piccinini, 2007c], [Piccinini, 2008]. It is more precise
than Hayess, because it talks about how the magic paper performs its tricks. And
it is less universal than Searles, because Piccinini doesnt think that everything is a
computer.
Unfortunately, there are two slightly different definitions to be found in Piccininis
papers:
Definition P1
The mathematical theory of how to generate output strings from input strings
in accordance with general rules that apply to all input strings and depend
on the inputs (and sometimes internal states) for their application is called
computability theory. Within computability theory, the activity of manipulating
strings of digits in this way is called computation. Any system that
9.5. GUALTIERO PICCININI:COMPUTERS AS DIGITAL STRING MANIPULATORS379
performs this kind of activity is a computing system properly so called.

[Piccinini, 2007b, 108, my emphasis]
Definition P2
[A]ny system whose correct mechanistic explanation ascribes to it the
function of generating output strings from input strings (and possibly
internal states), in accordance with a general rule that applies to all strings
and depends on the input strings (and possibly internal states) for its
application, is a computing mechanism. The mechanisms ability to perform
computations is explained mechanistically in terms of its components, their
functions, and their organization. [Piccinini, 2007c, 516, my emphasis]
These are almost the same, but there is a subtle difference between them.
9.5.1 Definition P1
Lets begin with Definition P1. It implies that a computer is any system (presumably,
a physical device, because only something physical can actively perform an action)
that manipulates strings of digits, that is, that generate[s] output strings from input
strings in accordance with general rules that apply to all input strings and [that]
depend on the inputs (and sometimes internal states) for their application. What kind
of general rule? In [Piccinini, 2008, 37], he uses the term algorithm instead of
general rule. This is consistent with the view that a computer is a Turing machine,
and explicates Hayess magic trick as being an algorithm.
The crucial point, according to Piccinini, is that the inputs and outputs must be
strings of digits. This is the significant difference between (digital) computers and
analog computers: The former manipulate strings of digits; the latter manipulate
real variables.
For Piccinini,
A digit is a particular [that is, a particular object or component of a device]

or a discrete state of a particular, discrete in the sense that it belongs to
one (and only one) of a finite number of types.. . . A string of digits is
a concatenation of digits, namely, a structure that is individuated by the
types of digits that compose it, their number, and their ordering (i.e., which
digit token is first, which is its successor, and so on) [Piccinini, 2007b,
107].9
In [Piccinini, 2007c, 510], he observes that a digit is analogous to a letter of an alphabet,

so they are like Turings symbols that can be printed on a Turing machines tape.
And
real variables are physical magnitudes that (i) vary over time, (ii) (are
assumed to) take a continuous range of values within certain bounds,
and (iii) (are assumed to) vary continuously over time. Examples of real
9 In the other two papers in his trilogy, Piccinini gives slightly different characterizations of what a digit
is, but these need not concern us here; see [Piccinini, 2007c, 510], [Piccinini, 2008, 34].
variables include the rate of rotation of a mechanical shaft and the voltage
level in an electrical wire [Piccinini, 2008, 48].
So far, so good. Neither Searle nor Hayes should be upset with this characterization.
9.5.2 Definition P2
But Piccininis second definition adds a curious phrase. This definition implies that
a computer is any system whose correct mechanistic explanation ascribes to it the
function of manipulating digit strings according to algorithms. What is the import of
that extra phrase?
It certainly sounds as if this is a weaker definition. In fact, it sounds a bit Searlean,
because it sounds as if it is not the case that a computer is an algorithmic, digit-
string manipulator, but rather that it is anything that can be so described by some kind
of mechanistic explanation. And that sounds as if being a computer is something
external and not intrinsic. So lets consider what Piccinini has in mind here. He
says:
Roughly, a mechanistic explanation involves a partition of a mechanism

into parts, an assignment of functions and organization to those parts, and
a statement that a mechanisms capacities are due to the way the parts and
their functions are organized. [Piccinini, 2007c, 502]
As explained in 14.4 and 19.5.3.2, syntax is the study of the properties of a collection
of objects and the relations among them. If a mechanism is considered as a collection
of its parts, then Piccininis notion of a mechanistic explanation sounds a lot like a
description of the mechanisms syntax. But syntax, you will recall, is what Searle says
is not intrinsic to a system (or a mechanism).
So how is Piccinini going to avoid a Searlean slippery slope and deny that
everything is a computer? One way he tries to do this is by suggesting that even if
a system can be analyzed syntactically in different ways, only one of those ways will
help us understand the systems behavior:
Mechanistic descriptions are sometimes said to be perspectival, in the

sense that the same component or activity may be seen as part of different
mechanisms depending on which phenomenon is being explained. . . . For
instance, the heart may be said to be for pumping blood as part of an
explanation of blood circulation, or it may be said to be for generating
rhythmic noises as part of an explanation of physicians who diagnose
patients by listening to their hearts. This kind of perspectivalism does
not trivialize mechanistic descriptions. Once we fix the phenomenon
to be explained, the question of what explains the phenomenon has an
objective answer. This applies to computations as well as other capacities
of mechanisms. A heart makes the same noises regardless of whether a
physician is interested in hearing it or anyone is interested in explaining
medical diagnosis. [Piccinini, 2007c, 516]
9.6. THE STORY SO FAR 381
Lets try to apply this to Searles Wordstar wall: From one perspective, the wall
is just a wall; from another, according to Searle, it can be taken as an implementation
of Wordstar. Compare this to Piccininis claim that, from one perspective, a heart is a
pump, and, from another, it is a noisemaker. If youre a doctor interested in hearing the
hearts noises, youll consider the heart as a noisemaker. If youre a doctor interested in
making a medical diagnosis, youll consider it as a pump. Similarly, if youre a house
painter, say, youll consider the wall as a flat surface to be colored, but if youre Searle,
youll try to consider it as a computer program. (Although I dont think youll be very
successful in using it to write a term paper!)
9.6 The Story So Far

So, what is a computer? It would seem that almost all proposed definitions agree on at
least the following:
Computers are physical devices; they interact with other physical devices in the
world.
They manipulate (physical) symbols (strings of digits), converting some into

others according to algorithms.
They are physical implementations of Turing machines in the sense that their
input-output behavior is logically equivalent to that of a Turing machine (even
though the details of their processing might not be). A slight modification of
this might be necessary to avoid the possibility that a physical device might be
considered to be a computer even if it doesnt compute: We probably want to
rule out real magic, for instance. So perhaps we need to say that a computer
must be a physical implementation of something that is logically equivalent to a
Turing machine (or a lambda-calculus machine, or a recursive-function machine,
etc.).
9.7 What Else Might Be a Computer?

Does such a definition include too much? Lets assume for a moment that something
like Piccininis reply to Searle carries the day, so that it makes sense to say that not
everything is a computer. Still, might there be some things that intuitively arent
computers but that turn out to be computers on even our narrow characterization?
This is always a possibility. Any time that you try to make an informal concept
precise, you run the risk of including some things under the precise concept that didnt
(seem to) fall under the informal concept. You also run the risk of excluding some
things that did. One way to react to this situation is to reject the formalization, or
else to refine it so as to exclude what shouldnt have been included and/or to include
what was erroneously excluded. But another reaction is to bite the bullet and agree
to the new inclusions and exclusions: For instance, you might even come to see that
something that you didnt think was a computer really was one.
9.7.1 Is a Brain a Computer?

Many people claim that the (human) brain is a computer. Searle thinks it is,
but only because he thinks that everything is a computer. But perhaps there is a
more interesting way in which the brain is a computer. Certainly, contemporary
computational cognitive science uses computers as at least a metaphor for the brain.
Before computers came along, there were many other physical metaphors for the
brain: The brain was considered to be like a telephone system or like a plumbing
system (see, e.g., [Squires, 1970, Sternberg, 1990, Gigerenzer and Goldstein, 1996,
Angier, 2010], and The Mind is a Metaphor website, accessed 31 January
2014 from http://mind.textdriven.com/?q=mind [concerning that website, see
[Guernsey, 2009]]).10
In fact, computationalism is sometimes taken to be the view that the brain
is a computer or that the brain computes or that brain states and processes are
computational states and processes:11
The basic idea of the computer model of the mind is that the mind
is the program and the brain the hardware of a computational system
[Searle, 1990, 21].
The core idea of cognitive science is that our brains are a kind of
computer. . . . Psychologists try to find out exactly what kinds of
programs our brains use, and how our brains implement those programs.
[Gopnik, 2009a, 43].
Computationalism. . . is the view that the functional organization of the
brain (or any other functionally equivalent system) is computational, or
that neural states are computational states [Piccinini, 2010a, 271] (see also
[Piccinini, 2010a, 277278]
But if one of the essential features of a computer is that it carries out computable
processes by computing rather than (say) by magic or by some other biological but
non-computable technique, then its at least logically possible that the brain is not
a computer even if brain processes are computable. How can this be? A process
is computable if and only if there are one or more algorithms that specify how that
process can be carried out. But it is logically possible for a process to be computable
in this sense without actually being computed. Hayess magic paper is a logically, if
not physically, possible example. Another example might be the brain itself. Piccinini
has argued [Piccinini, 2005], [Piccinini, 2007a] that neuron firings (more specifically,
spike trainsthat is, sequences of action potentialin groups of neurons) are not
representable as digit strings. But, because Piccinini believes that a device is not a
computer unless it manipulates digit strings, and, because it is generally considered
that human cognition is implemented by neuron firings, it follows that the brains
cognitive functioningeven if computableis not accomplished by computationis
not accomplished by computation.
10 There are also analogies between telephone systems and computers themselves [Lewis, 1953].
11 Sometimes it is taken to be the view that the mindis a computer; see the quotations in [Rapaport, 2012b,
2].
9.7. WHAT ELSE MIGHT BE A COMPUTER? 383
Yet, if cognitive functions are computable (as contemporary cognitive science

suggests [Edelman, 2008]), then there would still be algorithms that compute
cognition, even if the brain doesnt do it that way. Well return to this theme in
Chapter 19.
9.7.2 Is the Universe a Computer?
Figure 9.3: Abstruse Goose, http://abstrusegoose.com/219
Might the universe itself be a computer?12 Consider Keplers laws of planetary

motion. Are they just a computable theory that describes the behavior of the solar
system? If so, then a computer that calculates with them might be said to simulate the
solar system in the same way that any kind of program might be said to simulate a
physical (or biological, or economic) process, or in the same way that an AI program
might be said to simulate a cognitive process. (Well return to this idea in Chapter 15.)
Or does the solar system itself compute Keplers laws? If so, then the solar system
would seem to be a (special purpose) computer (that is, a kind of Turing machine).
12 In
addition to the cartoon in Fig. 9.3, see also Is the Universe a Computer?, http://abstrusegoose.com/115
(best viewed online!).
The second possibility does not necessarily follow from the first. As we just saw in
the case of the brain, there might be a computational theory of some phenomenon,
but the phenomenon itself need not behave computationally. Here are some other
examples: Someone might come up with a computational theory of the behavior of
the stock market, yet the actual stock markets behavior is determined by the individual
decisions made by individual investors. Calculations done by slide rules are done by
analog means, yet the calculations themselves are clearly computable.
Indeed, computational algorithms are so powerful that they can

simulate virtually any phenomena, without proving anything about
the computational nature of the actual mechanisms underlying these
phenomena. Computational algorithms generate a perfect description of
the rotation of the planets around the sun, although the solar system does
not compute in any way. In order to be considered as providing a model of
the mechanisms actually involved, and not only a simulation of the end-
product of mechanisms acting at a different level, computational models
have to perform better than alternative, non computational explanations.
[Perruchet and Vinter, 2002, 1.3.4]
Another argument along these lines has been put forth by Stephen Wolfram,
developer of the Mathematica computer program, in [Wolfram, 2002] CITE OTHER
DOCS FROM whatisacomputer.html; CITE ZUSE, FREDKIN. Wolfram argues as
follows:
1. Nature is discrete.
2. Therefore, possibly it is a cellular automaton.
3. There are cellular automata that are equivalent to a Turing machine.
4. Therefore, possibly the universe is a computer.
There are a number of problems with this argument. First, why should we believe
that nature (that is, the universe) is discrete? Presumably, because quantum mechanics
says that it is. But some distinguished physicists deny this [Weinberg, 2002]. So,
at best, for those of us who are not physicists able to take a stand on this issue,
Wolframs conclusion has to be conditional: If the universe is discrete, then possibly it
is a computer.
So, lets suppose (for the sake of the argument) that nature is discrete. Might it
be a cellular automaton? The easiest way to think of a cellular automaton is as a
two-dimensional Turing-machine tape. (For more information on cellular automata,
see [Burks, 1970]. Burks was one of the people involved in the construction of ENIAC
and EDVAC.) But, of course, even if a discrete universe might be a cellular automaton,
it need not be. If it isnt, the argument stops here. But, if it is, then the third premise
is reasonable. The conclusion appears to follow validly from the premises, of which
premise 2 is the one most in need of justification. But even if all of the premises and
(hence) the conclusion are true, it is not clear what philosophical consequences we are
supposed to draw from this.
9.8. EXERCISES FOR THE READER 385
9.8 Exercises for the Reader

(The material in this section was developed by Albert Goldfain.)
The following arguments are interesting to think about in relation to the question
whether everything a computer. Try to evaluate them.
Argument 1 P1 A Turing machine is a model of computation based on what a single

human (i.e., a clerk) does.
P2 Finite automata and push-down automata are mathematical models of
computation that recognize regular languages and context-free languages,
respectively.
P3 Recognizing strings in a languages is also something individual humans do.
C1 Thus Turing machines, finite automata, and push-down automata are all
models of computation based on the abilities of individuals.
Argument 2
John Conways Game of Life is a cellular-automaton model (albeit a very
simplistic one) of a society:13
P1 The Game of Life can be implemented in Java.
P2 Any Java program is reducible to a Turing-machine program.
C1 Thus, the Game of Life is Turing-machine-computable
Argument 3 P1 The Game of Life can be thought of as a model of computation.
P2 The Game of Life is a model of the abilities of a society.
P3 The abilities of a society exceed those of an individual.
C1 The abilities of a model of computation based on a society will exceed the
abilities of a model based on the abilities of an individual.
C2 It is not the case that every Turing-machine program could be translated to a
Game-of-Life computation.
Some of these arguments may have missing premises! To determine whether

the Game of Life might be a model of computation, do a Google search on:
game of life turing machine
Given an integer input (remember: everything can be encoded as an integer), how
could this integer be represented as live cells on an initial grid? How might stable
structures (remember: a 2 2 grid has 3 neighbors each) be used as memory? How
would an output be represented?
Can Turing-machine programs be reduced to Game-of-Life computations?
13 See http://en.wikipedia.org/wiki/Conway Game of Life for the rules of this game.


1. Comments on pencil-computer cartoons:
A pencil is really only an output device (it prints and erases). To turn it into a
full-fledged computer (or, at least, a physical Turing machine), you need to add
eyes (for input), hands (for moving left and right), and a mind (for states of
mind).
2. THERE ARE MANY OTHER COMMENTS TO BE MADE ABOUT PAT
HAYES: ON VIRTUAL MACHINES, ON O-MACHINES. DO LATER?
3. For 9.6:
SEE EXTRA MATERIAL IN LECTURE NOTES AT PP 89/77/ AND SHEETS
1-2
IS THERE ANYTHING MORE TO SAY ABOUT PICCININI?
HOW DOES PICCININI (OR SEARLE OR HAYES) HANDLE

INTERACTIVE COMPUTING?
DISCUSSION WITH STU:

COMPUTERS ARE PHYSICAL OBJECTS THAT INTERACT WITH
THE WORLD.
IF Turing machineS DONT DO THAT, SO MUCH THE WORSE FOR
THEM AS A MODEL OF WHAT A COMPUTER IS.
COMPUTERS ARENT Turing machineS; THEY CAN BE MODELED
BY Turing machineS.
COMPUTERS TAKE INPUT FROM THE WORLD. THEY MIGHT BE
MODELED AS HAVING THAT INPUT PRESTORED, BUT THATS
AN EXTENSIONAL, NOT AN INTENSIONAL MODEL.
4. All modern general-purpose digital computers
are physical embodiments of the same logical abstraction[:] Turings
universal machine. [Robinson, 1994, 45]
5. Lloyd:
[Lloyd and Ng, 2004] argues as follows:

(a) Nature is discrete. (This is the central maxim of quantum mechanics
(p. 54).)
(b) In particular, elementary particles have a spin axis that can be in one of
two directions.
(c) they encode a bit.
(d) elementary particles store bits of information.

(e) Interactions between particles can flip the spin axis;
this transforms the stored datathat is, these interactions are operations on
the data.
(f) (because any physical system stores and processes information) all
physical systems are computers.
(g) In particular, a rock is a computer.
(h) Also, the entire universe is a computer.
Premise 5a matches Wolframs fundamental premise and would seem to be
a necessity for anything to be considered a digital computer. The next four
premises also underlie quantum computing.
But the most serious problem with Lloyds argument as presented here is premise
5f. Is the processing sufficient to be considered to be Turing-machine-equivalent
computation? Perhaps; after all, it seems that all that is happening is that
cells change from 0s to 1s and vice versa. But thats not all thats involved in
computing (or is it?). What about the control structuresthe grammarof the
computation?
And although Lloyd wants to conclude that everything in the universe (including
the universe itself!) is a computer, note that this is not exactly the same as Searles
version of that claim. For Searle, everything can be interpreted as any computer
program. For Lloyd, anything is a computer, although they may not accept
input or give output in a form that is meaningful to humans (p. 55). So, for
Lloyd, its not a matter of intepretation. Moreover, analyzing the universe in
terms of bits and bytes does not replace analyzing it in conventional terms such
as force and energy (p. 54). Its not clear what the import of that is: Does he
mean that the computer analysis is irrelevant? Probably not: it does uncover
new and surprising facts (p. 54), though he is vague (in this general-audience
magazine article) on what those are. Does he mean that there are different
ways to understand a given object? That is true, but unsurprising (animals, for
instance, can be understood as physical objects satisfying the laws of quantum
mechanics as well as being understood as biological objects; and Dennetts
intentional-stance theory also spells out different ways in which things can be
understood). Does he mean that force and energy can, or should, be understood
in terms of the underlying computational nature of physical objects? He doesnt
say.
But he does end with a speculation on what it is that the universe is computing,
namely, itself! Or, as he puts it, computation is existence (p. 61). As mystical
as this sounds, does it mean anything different from the claim that the solar
system computes Keplers Law?
And heres an interesting puzzle for Lloyds view, relating it to issues concerning
whether a computer must halt (see the earlier discussion in WHERE?):
[Assuming the universe is computing its own evolution. . . , does it
have a finite lifetime or not? If it is infinite, then its self-computation
wont get done; it never produces an answer. . . . Hence, it does not

qualify as a computation. [Borbely, 2005, 15]
Related to Lloyd is Bostroms work on whether we are living in a computer

simulation. If Lloyd is right, then the universe is a computer, and we are
data structures in its program, brought to life as it were by its execution. If
Bostrom is right, then we are data structures in someone elses program. If
theists (computational theists?) are right, then we are data structures in Gods
program. Are there significant differences? See [Bostrom, 2003].
6. [A] computer. . . can be looked at from two different angles, which

Professor Hartree [[Hartree, 1949, 56]] has called the anatomical
and the physiological, that is, of what is it made? and how does
it tick? [Samuel, 1953, 1223]
Samuel then goes on to describe the anatomy in terms of things like magnetic
cores and vacuum tubes. Clearly, the anatomy has changed since then, so
defining computer in those terms doesnt seem to be the right way to go: Its too
changeable. Whats needed is a physiological, or functional, definition. Thus, a
computer should be defined as a physical device (where it doesnt matter what
the physical device is made of) that does such and such.
But is it even correct to limit a computer to a physical device? We want to
distinguish a real computer from a mathematical abstraction such as a Turing
machine. But what about a virtual computer implemented in some software,
such as a program that simulates a computer of a certain type but that runs on
a (physical) computer of a very different type? Note that, ultimately, there is a
physical substrate, however. (For more on this idea, see [Pylyshyn, 1992].)
7. [Samuel, 1953, 1223]s functional definition: A computer is an information or

data processing device which accepts data in one form and delivers it in an altered
form. This seems to be veryeven toohigh level a description. It omits any
mention of computation, of algorithms. It does mention that the delivered data
must have been processed from the accepted data by the device; so its
not just a function that relates the two forms of dataits more of a function
machine. But theres no specification of the kind of processing that it does.
Partly because of this, and on purpose, it also doesnt distinguish between
analog and digital computers. He resolves this by adding the modifier digital,
commenting that Any operation which can be reduced to arithmetic or to simple
logic can be handled by such a machine. There does not seem to be any
theoretical limit to the types of problems whch can be handle in this way
[Samuel, 1953, 1224]a nod, perhaps, to our binary-representation insight
(7.6.1. Still, this doesnt limit the processing to algorithmic processing.
It does, however, allow the brain to be considered as a computer: when the
human operator performs a reasonably complicated numerical calculation he
[sic]14 is forcing his brain to act as a digital computer [Samuel, 1953, 1224].
14 The use of the male gender here is balanced by Samuels earlier statement that computers have
A bit later (p. 1225), he does say that the processing must must be governed by
rules; this gets closer to the notion of an algorithm, though he (so far) puts no
constraints on the rules.
It is only after he discusses the control unit of the computer and its programming
(pp. 1226ff) that he talks about the kinds of control structures (loops, etc.) that
are involved with algorithms.
So, perhaps we could put all of this together and say that, for Samuel, a (digital)
computer is a physical device that algorithmically processes digital data
Further on, he adds the need for input and output devices (p. 1226). Are these
really needed? Are they part of the abstract, mathematical model of a computer,
namely, a Turing machine? Your first reaction might be to say that the tape
serves as both input and output device. But the tape is an integral part of the
Turing machine; it is really more like the set of internal switches of a physical
computer, whereas physical computers normally have input and output devices
(think of keyboards and monitors) as separate, additional components: Think of
a computer like the Mac Mini, which is sold without a keyboard and a monitor.
(This is surely related to issues that we have, or will, discuss elswhere, including
Knuths observations about the necessity of output but the non-necessity of input
(see our discussion of this in 7.5) and our discussion of the relation of computers
to the real world, which we begin in Ch. 10.)
8. This may not be relevant to the present book, but, in line with an observation
I made in [Rapaport, 2012b] about the output of an algorithm being an
interpretation of its input, [Samuel, 1953, 1225] says:
we might call the computer a mathematical translator in the same

sense that a literary translator takes information in one language
without adding or subtracting any basic information and renders this
information intelligible to someone not understanding the original
language.
9. On analog vs. digital computers:

Cite [Samuel, 1953, 1224, The Analogue Machine] and [Rubinoff, 1953].
And Shagir 1999.
And [Copeland, 1997, Nonclassical Analog Computing Machines, pp. 699
704].
10. Another argument that brains are computers:
(a) (Physical) computers are physical symbol sytems, in the sense of

[Newell and Simon, 1976].
(b) Humans are physical symbol systems [Newell, 1980, 174]
advantages in terms of the reductions in clerical manpower and woman power [Samuel, 1953, 1223, my
italics].
(c) there exists a physical architecture [a neural organization] that

supports that symbol system. [Newell, 1980, 174]
Now, this doesnt quite do it. First, we need the converse of the
second premise: All physical symbol systems are computers. (Is that in
[Newell and Simon, 1976] or [Newell, 1980]? Maybe!) Lets grant that. What
about the conclusion? Piccinini (CITE) argues that that neural architecture
doesnt compute. So, must a computer compute? Or could a computer be any
device that is extensionally (that is, input-output) equivalent to a computer-that-
computes?
11. There are many ways to describe a computing machine; they range
from Alan Turings classic account of a device for reading from and
writing to an infinite tape to the blueprints for the mechanism of
a contemporary supercomputer. Most such accounts are formally
equivalent to one another in the sense that what they describe can
solve the same classes of problems. [Hayes, 1990, 8]
Surely, a Turing machine is more than merely a input-output device: There has
to be some computing that goes on in between!
Here I want to take a highly abstract view of what constitutes a

computer. It is to be a device built out of objects drawn from
three realms: data, instructions and addresses. . . . Computing
requires interactions among the three kinds of objects. The most
obvious interactions are the ones in which instructions act on data.
[Hayes, 1990, 8]
And, to be a computer at least in the sense of something equivalent to a Turing

machine, the nature of the instructions, and the way that they act on data, must
satisfy the constraints that Turing imposed (or ones logically equivalent to them).
Another essential category of interactions reverses the flow of

information: data act on instructions. Such an event is generally
called a conditional branch. . . . conditional branching provides a
feedback channel, carrying information from the realm of data back
to the realm of instructions. Without such decision-making power
a computer would be crippled. It would always execute a fixed
sequence of instructions, oblivious to the data it was acting on.
[Hayes, 1990, 89]
This is an interesting insight! If the data are considered to be external to the

computer, then it is a way in which the external world acts on, or influences, the
computer. Note that the source of the data is not mentioned. If the data is already
written on the tape, that is, if the data is part of the pre-determined input, then
we have a Turing machine. But if the data could come from the external world in
the middle of a computation, then we would seem to have an interactive model
of computation, which, it is claimed, is hypercomputation. Note, however, that
what the computer (or hypercomputer?) actually does is no different in either

case: It prints and moves (it sets switches).
Hayess third componentaddressesadds something that classical Turing
machines lack, because the cells of a Turing machine have no addresses. Does
this add any kind of computing power? Probably not: Just replace every tape
cell by two cells, one of which contains an address for the other one. There are
details to be worked out, but it would seem that addresses only make certain
computations more efficient; they wouldnt allow non-computable functions to
be computed.
12. We count something as a computer because, and only when, its
inputs and outputs can usefully and systematically be interpreted
as representing the ordered pairs of some function that interests us.
. . . This means that delimiting the class of computers is not a sheerly
empirical matter, and hence that computer is not a natural kind.
. . . Similarly, we suggest, there is no intrinsic property necessary and
sufficient for all computers, just the interest-relative property that
someone sees value in interpreting a systems states as representing
states of some other system, and the properties of the system support
such an interpretation. . . . [I]n this very wide sense, even a sieve
or a threshing machine [or an eggbeater!?] could be considered a
computer. . . . [Churchland and Sejnowski, 1992, 6566]
This is essentially Searles point, only with a positive spin put on it. Note that
their definition in the first sentence has an objective component (the inputs and
outputs must be computationally related; note, too, that no specification is placed
on whether the mechanism by which the inputs are transformed into the outputs
is a computational one) and a subjective component (if the function computed
by the alleged computer is of no human interest, then it is not a computer!).
Thus, this is a bit different from Searle: Where Searle says that the wall behind
me is (or can be interpreted as) a word processor, Churchland & Sejnowski say
that the wall behind me is computing something, but we dont care what it is, so
we dont bother considering it to be a computer.
Both Searle as well as Churchland & Sejnowski consider the brain to be a
computer. On their definition of computer, its not how the outputs are produced
by the inputs that matters, only that the outputs can be computed by the inputs.
So we seem to have two definitions of computer, a weak and a strong one:
A strong computer is a physical device that causally transforms inputs into
outputs using Turing-machine-equivalent mechanisms.
A weak computer is a physical device that causally transforms inputs into
computable outputs by some mechanism that is not Turing-machine-equivalent.
Perhaps there is a weaker notion too: A look-up table which doesnt even have a
mechanism (they allude to this in their 2 (pp. 6974).
And there is yet another notion: Analog computers, such as slide rules (which
they discuss on p. 70), use non-computational mechanisms.
13. On Searles argument:
[A]n objection to Turings analysis. . . is that although Turings

account may be necessary it is not sufficient. If it is taken to be
sufficient then too many entities turn out to be computers. The
objection carries an embarrassing implication for computational
theories of mind: such theories are devoid of empirical content. If
virtually anything meets the requirements for being a computational
system then wherein lies the explanatory force of the claim that the
brain is such a system? [Copeland, 1996, 1, p. 335].
So, x is a computer iff x is a (physical) model of a Turing machine. To say that

this account is necessary means that, if x is a computer, then it is a model of
a Turing machine. That seems innocuous. To say that it is a sufficient account
is to say that, if x is a model of a Turing machine, then it is a computer. This
is allegedly problematic, because, allegedly, anything can be gerrymandered to
make it a model of a Turing machine; hence, anything is a computer (including,
for uninteresting reasons, the brain).
Interestingly, Copeland defends the sufficiency thesis (in this paper) but rejects
the necessity thesis (elsewhere, on hypercomputation grounds).
Copeland puts a constraint on algorithms that sounds right but that others (see our
discussion in 7.5) leave implicit: Algorithms are specific to an architecture
that specifies what the primitive operations are:
Thus an algorithm at least one of whose instructions calls explicitly

for a multiplication cannot be specific to an architecture that has
addition but not multiplication available as a primitive. A program
calling for multiplications can run on such an architecture only
because the compiler (or equivalent) replaces each multiplication
instruction in the program by a series of addition instructions.
[Copeland, 1996, 2, p. 337]
Copeland then suggests a definition of what it means for an entity e to compute

a function f : e computes f iff there is a labelling scheme L and an
algorithm+architecture A15 for computing f such that he, Li is a model of A,
where a labeling scheme is, roughly, a mapping from symbols to parts of e.
He then considers whether the solar system computes. . . Keplers law or
whether the universe is a computer dedicated to computing its own behaviour,
citing Fodor LOT p.74. (See 9.7.2, above.) His answer is that it doesnt, for
one reason because Keplers law isnt an algorithm and for another because any
algorithm that computed Keplers law would require a specific architecture that
doesnt match that of the solar system.
Copeland rejects this definition. His rejection begins with what he calls Searles
Theorem:
15 Copeland calls it SPEC.
For any entity x (with a sufficiently large number of discriminable

parts) and for any architecture-algorithm specification y there exists a
labelling scheme L such that hx, Li is a model of y. [Copeland, 1996,
2, p. 339]
In 4, he proves this theorem. The proof is beyond our scope, so let us accept
it for the sake of the argument. The problem with the Theorem, according to
Copeland, is that the proof only works for a nonstandard interpretation of x,
where by a nonstandard interpretation Copeland means an interpretation that
makes all statements come out true but is not about the entities that the standard
interpretation is about, something like saying that if Moscow is non-standardly
interpreted to mean the number 16, London is non-standardly interpreted to
mean the number 1, and is north of is non-standardly interpreted to mean is
greater than, then Moscow is north of London is true, but is not about Moscow
or London.
So, he refines Searles Theorem to this: e computes f iff there is an L and an A
such that he, Li is an honest model of A, where a model is honest iff it is not
non-standard. [Copeland, 1996, 5, p. 348]
But, he argues, from this it does not follow that, for any x, x is a computer. In
particular, and contra Searle, it does not follow that the brain is a computer. Nor
does it follow that the brain is not a computer. All we can ask is whether there
is a way to label the brain so that we can say honestly that it computes, and
that is an empirical question, not a trivial one. [Copeland, 1996, 8, p. 357]
14. On the brain as computer:

Searle is not the only one wondering if the brain is a computer.
[Churchland and Sejnowski, 1992] and [Ballard, 1997] have both written books
about whether, and in what ways, the brain might be a computer. As
[Ballard, 1997, 12] puts it, The key question. . . is, Is computation sufficient to
model the brain?. One reason this is an interesting question is that researchers
in vision have wondered how. . . an incomplete description, encoded within
neural states, [could] be sufficient to direct the survival and successful adaptive
behavior of a living system [Richards, 1988, as cited in Ballard 1997, p. 2]. If a
computational model of this ability is sufficient, then it might also be sufficient
to model the brain. And this might be the case even if, as, for example, Piccinini
CITE argues, the brain itself is not a computer, that is, does not behave in a
computational fashion. A model of a phenomenon does not need to be identical
in all respects to the phenomenon that it models, as long as it serves the purposes
of the modeling. But Ballard also makes the stronger claim when he says, a
few pages later, If the brain is performing computation, it should obey the laws
of computational theory [Ballard, 1997, 6, my italics]. But whether the brain
performs computations is a different question from whether its performance can
be modeled or described in computational terms. So, the brain doesnt have
to be a computer in order for its behavior to be describable computationally.
As [Churchland and Sejnowski, 1992] note, whether the brain is a computer
whether, that is, the brains functioning satisfies one of the (logically equivalent)
characterizations of computingis an empirical issue.
What about the wall behind me? Presumably, it, too, doesnt have to be a
computer in order for its (molecular or subatomic) behavior to be describable
computationally. Or is Searle making a stronger claim, namely, that, not only is
its behavior describable computationally, but it is a computation?
[Ballard, 1997, 11] has an interesting variation on that stronger claim:
Something as ordinary as a table might be thought of as running an

algorithm that adjusts its atoms continaually, governed by an energy
function. Whatever its variables are, just denote them collectively
by x. Then you can think of the table as solving the problem of
adjusting its atoms so as to minimize energy, that is, minx E(x). Is
this computation?
Note that this is different from Searles claim that the table (or a wall) might be
computing a word processor. It seems closer to the idea that the solar system
might be computing Keplers law. (See discussion on this elsewhere.)
15. Another claim in the vicinity of Searles and Ballards concerns DNA computing:
Computer. The word conjures up images of keyboards and monitors.

. . . But must it be this way? The computer that you are using to
read these words [that is, your brain!comment by WJR] bears
little resemblance to a PC. Perhaps our view of computation is too
limited. What if computers were ubiquitous and could be found in
many forms? Could a liquid computer exist in which interacting
molecules perform computations? The answer is yes. This is the
story of the DNA computer. [Adleman, 1998, 54, my italics]
Of course, Adelman is not making the Searlean claim that everything is a

computer and that, therefore, the interacting molecules of (any) liquid peform
computations. Nor is he making the Ballardian claim that DNA computes in
the way that a table computes. (Others have, however, made such a claim, on
the grounds that strands of DNA are similar to Turing-machine tapes with a
four symbols instead of two and with the processes of DNA transcription and
recombination as being computable processes. CITE) Rather, Adlemans claim
is that one can use DNA to solve mathematical problems. However, contrary
to what the editors of Scientific American wrote in their subtitle to Adlemans
article, it is unlikely that is redefining what is meant by computation .
After all, the advent of transistors did not change Turings mathematical
characterization of computing any more than the use of vacuum tubes did. At
most, DNA computers might change what the lay public means by computer.
But, as we have seen/will see elsewhere (WHERE?), that has already happened,
with the meaning changing from humans to computing machines.
16. On computers as switch-changing devices:

Advocates of both DNA and quantum computing focus on the idea that
computers are switch-changers. First, from the quantum perspective:
Boiled down to its essentials, any computer must meet two

requirements: it must be able to store information as strings of 1s and
0s, ro bits, and it must have a way of altering the bits in accordance
with instructions. [Grover, 1999, 25]
Actually, it need not store just 1s and 0s, and, of course, it doesnt store numbers:
It stores numerals (1s, not 1s). Furthermore, the instructions must satisfy the
constraints on algorithms.
From the DNA perspective:
To build a computer, only two things are really necessarya method

of storing information and a few simple operations for acting on
that information. . . . The remarkable thing is that just about any
method of storing information and any set of operations to act on
that information are good enough. [Adleman, 1998, 37]
Adleman is more correct about the storage method than Grover, but equally
vague about the operations, if not worse: Surely, not any operations will do.
(Maybe just about is the hedge that saves him; nonetheless, it would be nice to
be a bit more accurate.)
17. Can we define a computer simply as any device that carries out an algorithm?
[Davis, 2000] suggests (but does not explicitly endorse) this. I suppose that
it depends on what carries out means: Surely it has to include as part of its
meaning that the internal mechanism of the device must operate in accordance
withmust behave exactly likeone of the logically equivalent mathematical
models of computation. Surely, any computer does that. Is anything that does
that a computer? Is anything that can be gerrymanderedly described as doing
that a computer?
18. Re Thomason:
A computer is a set of registers with contents (or: switches with settings; if they
are binary switches, each is either on or else off), some of which are interpreted
(by whom? by what?) as data, some as program, and the rest as irrelevant (and
some as output?). Computation changes the contents (the settings).
19. von Neumanns definition of computer:

In his First Draft Report on the EDVAC, whichalong with Turings 1936
papermay be taken as one of the founding documents of computer science,
[von Neumann, 1945, 1.0, p. 1] gives the following definition:
An automatic computing system is a (usually highly composite)

device, which can carry out instructions to perform calculations of
a considerable order of complexity. . . . . . . The instructions. . . must
be given to the device in absolutely exhaustive detail. They include

all numerical information which is required to solve the problem
under consideration. . . . . . . All these precedures require the use of
some code to express. . . the problem. . . , as well as the necessary
numerical material. . . . . . . [T]he device. . . must be able to carry them
out completely and without any need for further intelligent human
intervention. At the end of the required operations the device must
record the rsults again in one of the forms referred to above.
Other comments (in this section of [von Neumann, 1945], as well as later, in
5.0 (pp. 6ff)) indicate that the code should be binary, hence that the computer
is a digital device (1.0, p. 1). This definition hues closely to being a
physical implementation of a Turing machine, with clear allusions to the required
algorithmic nature of the instructions, and with a requirement that there be both
input and output (recall our discussion of the necessity of thisor lack thereof
in 7.5.
20. Naur 2007 (see Further Sources, below) argues that the nervous system. . . has
no similarity whatever to a computer. His argument is difficult to follow, but
that is another story. Even if true (and others, like Piccinini, think that it is),
it raises another question: Does the nervous system compute? There are several
different versions of this kind of question: We might say that a device x computes
iff:
(a) x behaves like a Turing machine(or anything that is logically equivalent to

a Turing machine, such as a register machine), or else
(b) x has the same input-output behavior as a Turing machine (even if it is
magical), xor
(c) x has the same input-output behavior as a Turing machine and x does
something (even if what it does is not computation)
21. More on solar system computing Keplers law:

[Shagrir, 2006b, 394] raises an interesting question:
Computer models are often used to study, simulate, and predict the
behavior of dynamical systems. In most cases, we do not view the
modeled system, e.g., the solar system, as a computer. When studying
the brain, however, our approach is different. In this case, in addition
to using a computer model to simulate the system under investigation,
i.e., the brain, we also take the modeled system itself to compute,
viewing its dynamical processes as computing processes. Why is
this so? What is it about the brain that makes us consider it, along
with desktops and pocket calculators, a species of computer, when
it doesnt even occur to us to accord that status to solar systems,
stomachs and washing machines?
Of course, it has occurred to several people, as we have seen.

Shagrirs bigger point is that, even if anything can be described (by us) as being
a computereven if being a computer is a matter of perspective (p. 399)
discovering how this can be for a given device (such as the brain) can give
us useful information about the device in terms of what it represents and how
it operates on those representations. And, I suppose, in other cases, this
information is not deemed (at least at the current time) to be useful (as in the
case of Searles wall).
A further point of Shagrirs is that what makes a computational perspective on a
device interesting has to do with semantics, not (just) syntax: assigning content
and information to brain/mental states is often prior to taking the computational
outlook (p. 402). So, semantics can guide us to look for an interpretation of a
device and its activities that will turn out to be a computational interpretation:
[T]he problems the brain computes are certain regularities or

functions that are specified in terms of the representational
content of the arguments (inputs) and values (outputs). I
call these problems semantic tasks. . . . It is also consistent
with the computing technology tradition that calculators compute
mathematical functions. . . chess machines compute the next move on
the chess-board. . . . [Shagrir, 2006b, 403]
Recall Fodors story of the chess computer that can be interpreted as computing
a Civil War battle. There are two ways to describe a given devices (a given
computers) behavior: purely syntactically, or semantically, that is, solely in
terms of the devices own, internal behavior (syntactically), but also in terms
of its relations to other things (semantically). In some cases (Searles wall),
such a semantic interpretation is at best a curiosity. In others (the brain, the
chess computer), it is highly useful and insightful. In still others (the Civil
War interpretation of the chess computer), it may be serendipitously useful or
insightful (Wow, my chess computer does double duty; what a bargain!, or:
How interesting: This chess match has the same structure as this Civil War
battle! (Grant not only metaphorically, but literally, checkmated Lee!)).
We are now in a better position to say why we apply the

computational approach to some systems, e.g., brains, but not others,
e.g., planetary systems, stomachs and washing machines. . . . We
apply the computational approach when our goal is to explain a
semantic pattern manifested by the system in question. . . . We do
not view planetary systems, stomachs and washing machines as
computers because the tasks they perform. . . are not defined in terms
of representational content. Again, we could view them as computing
semantic tasks. . . . But we dont. . . . [Shagrir, 2006b, 405]
That is, we apply the computational approach to explain intentional-stance

descriptions of a system. But isnt explaining washing-machine behavior in
terms of clothes-washing semantic? (Even functional: input dirty blue shirt;
output clean blue shirt.)16 Well, yes, but we have not found that to be a useful or
insightful perspective.
What makes a computational analysis more useful than some other analysis (say,
a purely physical or mechanical analysis)? Here is Shagrirs answer:
A computational theory in neuroscience aims to explain how

the brain extracts information content G from another, F. We
know from electrophysiological experiments that G is encoded in the
electrical signals of a neural state Nn , and F in those of N0 . Let
us also assume that we can track chemical and electrical processes
that mediate between N0 and Nn . What we still want to know is
why the information content of the electrical signals of Nn is G,
e.g. target location in headcentered coordinates, and not, say, the
targets color. The question arises since the encoded information
G are not directly related to the target, but is mediated through a
representational state that encodes F. The question, in other words,
is why a chemical and electrical process that starts in a brain state N0
that encodes F, and terminates in another brain state, Nn , yields the
information content G. After all, this causal process N0 Nn is not
sensitive to what is being represented, but to chemical and electrical
properties in the brain.
A computational theory explains why it is G that is extracted from
F by pointing to correspondence relations between mathematical
or formal properties of the 27 representing stateswhat we call
computational structureand mathematical or formal properties of
the represented states. The idea is that by showing that N0 encodes
F, and that the mathematical relations between N0 and Nn correspond
to those of the worldly represented states that are encoded as F and
G, we have shown that Nn encodes G.
Note the similarity of this to classic philosophical arguments about dualism (our
mental thoughts have no causal influence, because they are all implemented
in physical mechanisms). But this is exactly right: It is the idea behind
functionalism in the philosophy of mind: It is the functional, or computational
or what Shagrir is calling the semanticlevel of analysis that tells us what it is
that the physical level is doing.
22. Albert Goldfain (personal communication, 15 January 2008) suggests calling

Searles position pancomputationalism. Others have had the same, or similar,
ideas; see, e.g.,
http://philosophyofbrains.com/2007/06/26/pancomputationalism-origins-of-a-
neologism.aspx
23. Thus every organized body of a living thing is a kind of divine

machine or natural automaton. It infinitely surpasses any artificial
16 Missing socks might be an exception!
automaton, because a man-made machine isnt a machine in every

one of its parts. For example, a cognitive on a brass wheel has parts
or fragments which to us are no longer anything artificial, and bear
no signs of their relation to the intended use of the wheel, signs that
would mark them out as parts of a machine. But Natures machines
living bodies, that isare machines even in their smallest parts,
right down to infinity. That is what makes the difference between
nature and artifice, that is, between divine artifice and our artifice.
[Leibniz, 1714, 64]
This bears interesting comparison to Leibnizs famous mill passage, worth
discussing in Ch. 19. Presumably, the brain, as a natural (or divine) artifact
is infinitely complex, whereas a computer, as a human artifact, is not. Note
that a human-artifactual computer must bottom out in something that is not a
computer, just as recursive functions must bottom out in primitive operations. It
is unknown whether natural artifacts such as the brain bottom out: The brain is
made of neurons, etc., which are made of molecules, which are made of atoms,
which are made of subatomic particles, which are made of quarks, etc. We dont
know if thats the bottom level or if there is a bottom level. In any case, if such a
bottom level is a computer, as Lloyd argues, then the brain (as well as everything
else) is made of computers all the way down. Otherwise, it, too, bottoms out in
something non-computational.17
24. On the brain as a computer:
If the brain is a machine that processes information, then its cognitive

activity can be interpreted as a set of information processing states
linking stimulus to response (i.e. as a mechanism or an algorithm).
[Schyns et al., 2008, 20]
But thats not sufficient to be an algorithm. There are two issues: Does the
brain do anything that we can study? Does it compute? The first question
was answered negatively by the behaviorists. The original cognitivists (such as
[Miller et al., 1960]) answered positively and argued that the mechanism was
computational. But, as Piccinini CITE has argued (and as I pointed out in
9.7.1), there might be an internal mechanism that is not computational. So, what
properly follows from the antededent of the above quotation is this: . . . then the
brain is a computer iff the processing isnot merely computablebut actual
computation. Merely begin able to process information is not sufficient for being
an algorithm.
To be fair, Schyns et al. argue that the information processing that the brain
does is computational, not merely computable. For instance, they claim
that it is possible to translate [a] brain microstate. . . into [an] information
processing macrostate [Schyns et al., 2008, 22; see also pp. 2324], and that
17 Thoughts inspired by discussion with Albert Goldfain, 15 January 2008.
such macrostates are computational states. However, I will leave the dispute
between Piccinini and them for you to consider on your own.
25. [Newell, 1980, 174] also holds that the brain is a computer:
There must exist a neural organization that is an architecturei.e.,
that supports a symbol structure.
But this does not necessarily follow from a thesis that cognition is computable
(which Newell certainly holds), becauseas we have seen WHERE?it might
be possible that the neural organization of the brain is not computational even
though its behavior is computable.
26. On the GCD discussion in 9.3.2:
See [Goodman, 1987, 484].
27. On intrinsic properties:
[Rescorla, 2014a] uses slightly different terms for similar ideas:
Inherited meanings areise when a systems semantic properties are
assigned to it by external observers, either through explicit stipulation
or through tacit convention. Nothing about the system helps generate
its own semantics. Indigenous meanings arise when a system
helps generate its own semantics (perhaps with ample help from
its evolutionary, desgin, or causal history, along with other factors).
The system helps confer content upon itself, through its internal
operations or its interactions with the external world. Its semantics
does not simply result from external assignment. (p. 180)
28. [Aaronson, 2011] claims that quantum computing has overthrown views
like those of [Wolfram, 2002] that the universe itself is basically a giant
computer. . . by showing that if [it is, then] its a vastly more powerful kind of
computer than any yet constructed by humankind.
29. On magic paper, see Ch. 14, Notes for Next Draft No. 8.
30. Is the brain a computer in this sense? Arguably. For a start, the
brain can be programmed to implement various computations by
the laborious means of conscious serial rule-following; but this is a
fairly incidental ability. On a different level, it might be argued that
learning provides a certain kind of programmability and parameter-
setting, but this is a sufficiently indirect kind of parameter-setting that
it might be argued that it does not qualify. In any case, the question
is quite unimportant for our purposes. What counts is that the brain
implements various complex computations, not that it is a computer.
[Chalmers, 2011, 2.2, esp. p. 336]
There are two interesting points made here. The first is that the brain can simulate
a Turing machine. The second is the last sentence: What really matters is that
the brain can compute, not that it is a computer. To say that it is a computer
raises the question of what kind of computer it is: A Turing machine? A register
machine? Something sui generis? And these questions seem to be of less interest
than the fact that it can compute.
31. On the Searle-Putnam argument that anything implements every computation:
. . . there is nothing approaching a formal demonstration that

the structure-from-motion computation or vector subtraction. . .
[that is, alleged semantic computations] do not have unintended
implementations. However, there is no reason to think that the
possibility of such implementations renders the explanations that
computational theories do provide trivial or otiose. To see why not
we must again focus on actual practice.
The epistemological context in which computational models
are developed is instructive. We dont start with a computational
description and wonder whether a given physical system implements
it. Rather, we start with a physical system such as ourselves. More
precisely, we start with the observation that a given physical system
has some cognitive competencethe system can add, or it can
understand and produce speech, or it can see the three-dimensional
layout of its immediate environment. Thus, the explanandum of the
theory is a manifest cognitive capacity. The computational theorists
job is to explain how the physical system accomplishes the cognitive
task. There is, of course, no evidence that Putnams rock or Searles
wall has any cognitive capacities. With the target capacity in view,
the theorist hypothesizes that the system computes some well-defined
function (in the mathematical sense), and spells out how computing
this function would explain the systems observed success at the
cognitive task.
[Egan, 2012, 45]
There cant be any formal demonstration of the non-existence of unintended

implementations, because any abstraction can have multiple implementations
(and some, such as the natural numbers as specified by Peanos axioms, can
have an infinite number of implementations). What is important, however, is
Egans point that the existence of unintended implementations (such as, perhaps,
Searles wall that implements Wordstar or a computational theory of cognition)
does not make Wordstar or a computational theory of cognition trivial. The
second paragraph spells out the point that I make elsewhere (WHERE?) that
we begin with a physical entity (a physical implementation), and subsequently
derive an Abstraction from (that is, a theory of) it. From that Abstraction,
other implementations can be constructed or found. If some of those other
implementations are odd, that does not make the Abstraction of less importance.
We can either bite the bullet and accept the unintended implementation, or try to
refine the Abstraction. (The latter task, however, is unlikely to succeed, because
the refined Abstraction will have its own unintended implementations.) As Egan
points out, if the Abstraction succeeds in explaining the original implementation,

thats all that we need.
I would, however, disagree with one possible interpretation of the claim that
There is, of course, no evidence that Putnams rock or Searles wall has any
cognitive capacities. I agree with this statement. But it should not be interpreted
to mean that no intended or even unintended implementation has cognitive
capacities.
And the reason for not drawing that stronger conclusion can be found in this
passage:
In order for the system to be used to compute the addition function

these causal relations have to hold at a certain level of grain, a level
that is determined by the discriminative abilities of the user. That
is why, whatever the status of Putnams argument, no money is to
be made trying to sell a rock as a calculator. Even if (per mirabile)
there happens to be a set of state-types at the quantum-mechanical
level whose causal relations do mirror the formal structure of the
addition function, microphysical changes at the quantum level are not
discriminable by human users, hence human users could not use such
a system to add. (God, in a playful mood, could use the rock to add.)
[Egan, 2012, 46]
In other words, even if Searles wall implements Wordstar, we wouldnt be able

to use it as such.
32. [Chalmers, 2012, 215216] makes much the same point:
On my account, a pool table will certainly implement various a-

computations and perform various c-computations [that is, concrete
computational processes]. It will probably not implement interesting
computations such as algorithms for vector addition, but it will at
least implement a few multi-state automata and the like. These
computations will not be of much explanatory use in understanding
the activity of playing pool, in part because so much of interest in
pool are not organizationally invariant and therefore involve more
than computational structure.
33. On the Searle-Putnam Puzzle:
Why are we are so worried about liberalism? . . . One concern,

emphasized by Searle [Searle, 1990], is that liberalism makes the
claim that the brain computes vacuous, in the sense that its truth-
value is not a matter of facts about the brain (computation is not
discovered in the physics) ibid., p. [35]), but a matter of our whims
(computation is not an intrinsic feature of the world. It is assigned
relative to observers, Searle 1992 [ADD TO BIB!], p. 212).
One can note that there must be something wrong with this
objection. For there is a trivial sense in which everything represents,
and yet we accept the claim that brains (or minds) represent very
seriously. We do not tend to complain that it is vacuous. I think that
the same is true for computation: everything is a computer in some
trivial sense, and yet it does not follow that the claim that the brain
computes is senseless. The computational properties of the brain are
discovered, and not assigned relative to an observer. [Shagrir, 2010,
277, col. 1, my italics]
We can now turn to the second objection, of failing to draw the

line between computing and non-computing systems. The reply
is this: my account implies that every system computes in the
sense that it could be a computer, but it does not imply that it
computes in the sense that it is one. According to my account,
an AM-computer is either (i) a natural computer, in the sense
that the simulation processes are defined over a natural system of
representations, or (ii) a conventional computer, in the sense that
its simulation processes are defined over a conventional system
of representations. Arguably, brains belong to the first category.
Desktops, defined over representational systems of type I, and robots,
defined over representational systems of type II, are confined to the
second. Stomachs, tornadoes and washing machines are neither.
Their processes are defined over neither natural nor conventional
systems of representations. True, my account implies, as it should,
that stomachs could be (conventional) computers: we could interpret
its states as representing the states of some automaton and simulating
its behavior. But we wont, and so stomachs dont compute.
[Shagrir, 2010, 277, col. 2, my boldface and underlined emphases,
original italics]
I think this approach (independent of Shagrirs notion of an analog-model

computer) is the best way to deal with Searle and Putnam.
34. On the brain as computer:
Should we think of this relation betwen inputs and outputs as a kind of

effective computational procedure? If we know that that the system
is in state A and receives input B will we be able to tell what the
next set of representations will be like? The answer at this point is
simply that we do not know whether there is a procedure like this or
not, nor much about how it would operate if there were. What we do
know, however, is that there are consistent causal relations between
input and representations, both in science and childhood, and my
empirical claim is that those relations are similar. The best current
explanation of how a brain could instantiate this kind of system of
rules and representations is that it is a kind of computer.
An abductive argument that (i.e., a scientific hypothesis that) the brain is a

computer. Note that this is a much more reasonable argument than Searles or
than trying to model the brain as, say, a Turing machine.
35. Is the human race itself a computer? Is a tree a computer? According to Cesar
Higalgo, matter can can compute:
A tree. . . is a computer that knows in which direction to grow its
roots and leaves. Trees know when to turn genes on and off to fight
parasites, when to sprout or shed their leaves and how to harvest
carbon from the air via photosynthesis. As a computer, a tree begets
order in the macrostructure of its branches and the microstructures of
its cells. We often fail to acknowledge trees as computers, but the
fact is that trees contribute to the growth of information in our planet
because they compute. [Hidalgo, 2015, 75].
Also see the interview with Hidalgo in [ONeill, 2015]. But what is his argument
here? He doesnt seem to have a definition of computer. Except for the
last three words of the above quotation, one might think that his definition
would be something like: A computer is an information-processing machine.
Then his argument might go as follows: Trees are information-processing
machines (because they contribute to the growth of information); hence, they
are computers. But those last three words suggest that his argument goes the
other way: that trees are computers; hence, they contribute to the growth of
information.
36. Possible end-of-chapter discussion question:
Which physical processes are computing processes? Are all physical processes
computations?18 Of course, if a physical process is a computation, then,
presumably, the physical object carrying out that process is a computer, so this
question amounts to sahying that all physical objects that carry out processes are
computers.
One argument, adapted from [Fekete and Edelman, 2011], is this:

(a) A process is a computation iff it operates on representations.
(b) All physical processes can represent.
(c) All physical processes are computations.
Is this argument sound? Keep in mind that, even if all physical processes can
represent, it does not follow that they all do represent. (Or does that suggest that
computing is in the eye of the beholder. If a rock heating up in the sun is not
taken as a representer, then it is not computing, but if I use how hot it is to do
something else, then the hot rock is representing and so computing.19 Another
consideration is this: Computation is done over uninterpreted marks. Whether
18 Thanks to Russ Abbott and Eric Dietrich for suggesting these.
19 Dietrich, personal communication, 28 June 2015.
those marks represent anything is a separate matter. I might choose to interpret

them as representing something; or the computational system itself might choose
to (self-?)interpret them as representing something.20
Does this argument adequately represent Fekete & Edelmans actual argument?
(See 7.1046.)
And here is a related question:21 Is a (physical) implementation of a computation

itself a computation? (See our discussion of implementation in Ch. 14.)
37. Modern porgrammers would say. . . [that Jacquard] loom programs
are not computer programs: looms could not compute mathematical
functions. [Denning and Martell, 2015, 83]
Even if we accept the definition of a computer (program) as one that computes
mathematical functions, it does not follow that Jacquard looms could not be
computers, because it would depend on whether or not bits could be implemented
as patterns in looms. There seems no reason why they couldnt be. Looms might
not have been computers, but that doesnt mean that they couldnt have been.
38. Historians might also wonder if mathematicians who devised
algorithms were programmers. . . . Modern programmers would again
say no because these algorithms were not encoded for a particular
machine. [Denning and Martell, 2015, 83]
But they were! They were encoded for humans (human computers)! (Curiously,
on the very next page, Denning & Martell say this:
The women who calculated ballistic tables for the Army during World
War II were also programmers, although their programs were not
instructions for a machine but for themselves to operate mechanical
calculators. In effect, they were human processing units.
But why should this be treated as a kind of joke or metaphor? These women
were the computers!
20 See Paul Schweizers presentation at IACAP-2015.

21 Also due to Dietrich.

(a) Brunjes, Shannon (1977), What Is a Computer?, Journal of Medical Systems 1(1):
7985.
A computer is an information-processing machine, varying in size from the
very large system to the very small microprocessor, operating by means of
stored programs which can be modified, and using various auxiliary devices
to communicate with the user. (p. 85)
(b) Chaitin, Gregory (2006), How Real Are Real Numbers?, International
Journal of Bifurcation and Chaos
i. Accessed 6 November 2012 from:
2006 version: http://www.cs.auckland.ac.nz/chaitin/olympia.pdf
2009 version: http://www.umcs.maine.edu/chaitin/wlu.html
ii. On the halting problem, the non-existence of real numbers(!), and the idea
that everything is software, God is a computer programmer,. . . and the world
is. . . a giant computer(!!).
(c) Chalmers, David (1993), A Computational Foundation for the Study of
Cognition, Journal of Cognitive Science (South Korea) 12(4) (October-December
2011): 323357.
http://j-cs.org/gnuboard/bbs/board.php?bo table= vol012i4&wr id=2
ii. Originally written in 1993; slightly revised version published 2011.
iii. See esp. the section What about computers? (pp. 335336).
iv. The 2011 version was accompanied by commentaries (including [Egan, 2012],
[Rescorla, 2012b], [Scheutz, 2012], and [Shagrir, 2012]) and a reply by
Chalmers [Chalmers, 2012].
(d) Davis, Martin (2004), The Myth of Hypercomputation, in C. Teuscher (ed.),
212.
Contains interesting comments on the nature of of computation and
computers.
(e) Floridi, Luciano (1999), The Digital Workshop, Ch. 2 of Philosophy and
Computing: An Introduction (London: Routledge).
(f) Giere, Ronald N. (2003), The Role of Computation in Scientific Cognition,
Journal of Experimental and Theoretical Artificial Intelligence 15: 195202.
(g) Gruber, John (2007), Apples Computer, Incorporated, Macworld (April): 112.
Never mind the name changethe Apple TV and iPhone are computers to
the core.
(h) Harnish, Robert M. (2002), Coda: Computation for Cognitive Science, or What IS
a Computer, Anyway?, Minds, Brains, Computers: An Historical Introduction to
the Foundations of Cognitive Science (Malden, MA: Blackwell Publishers): 394
412.
(i) Kanat-Alexander, Max (2008), What Is a Computer?, Code Simplicity

(10 October)
http://www.codesimplicity.com/post/what-is-a-computer/
ii. A computer is Any piece of matter which can carry out symbolic instructions
and compare data in assistance of a human goal.
(j) Kroes, Peter (2010), Engineering and the Dual Nature of Technical Artefacts,
properties
(k) Lloyd, Seth (2000), Ultimate Physical Limits to Computation, Nature 406
(31 August): 10471054.
i. Pre-prints accessed 4 June 2014 from:
http://arxiv.org/pdf/quant-ph/9908043.pdf?origin=publicationDetail
and
http://cds.cern.ch/record/396654/files/9908043.pdf
ii. Investigates quantitative bounds to the computational power of an ultimate
laptop with a mass of one kilogram confined to a volume of one litre.
(l) Lloyd, Seth (2002), Computational Capacity of the Universe, Physical Review
Letters 88(23) (June): 237901-1237901-4.
All physical systems register and process information. The laws of physics
determine the amount of information that a physical system can register
(number of bits) and the number of elementary logic operations that a system
can perform (number of ops). The Universe is a physical system. The amount
of information that the Universe can register and the number of elementary
operations that it can have performed over its history are calculated. The
Universe can have performed 10120 ops on 1090 bits (10120 bits including
gravitational degrees of freedom). (from the abstract)
(m) Lloyd, Seth (2006), Programming the Universe: A Quantum Computer Scientist
Takes on the Cosmos (New York: Alfred A. Knopf).
See Powell 2006, Schmidhuber 2006 in Strongly Recommended readings,
above.
(n) Marcus, Gary (2015), Fact It, Your Brain Is a Computer, New York Times, Sunday
Review (28 June): SR12,
http://www.nytimes.com/2015/06/28/opinion/sunday/face-it-your-brain-is-a-
computer.html
For most neuroscientists, this is just a bad metaphor. But its still the most
useful analogy that we have.. . . The sooner we can figure out what kind of
comptuer the brain is, the better.
For a reply, see:
Linker, Damon (2015), No, Your Brain Isnt a Computer, The Week (1 July),
http://theweek.com/articles/563975/no-brain-isnt-computer
(o) Marshall, Dan (2014), The Varieties of Intrinsicality, Philosophy and

Phenomenological Research, forthcoming; preprint accessed 31 December 2014
from:
https://www.academia.edu/7096450/The Varieties of Intrinsicality
An analysis of different theories of what intrinsic means, hence useful for
critiquing Searle 1990.
(p) Naur, Peter (2007), Computing Versus Human Thinking, Communications of the
ACM 50(1) (January): 8594.
http://www.computingscience.nl/docs/vakken/exp/Articles/NaurThinking.pdf
[T]he nervous system. . . has no similarity whatever to a computer. (p. 85)
(q) Samuel, Arthur L. (1953), Computing Bit by Bit, or Digital Computers Made
Easy, Proceedings of the Institute of Radio Engineers 41(10) (October): 1223
1230.
A description of how computers work, written, for radio engineers who (in
1953) may never have seen one(!), by the creator of one of the earliest AI
programs: Samuels checkers player.
Comes close to defining a computer as an information or data processing
device which accepts data in one form and delivers it in an altered form
(p. 1223).
(r) Schulman, Ari N. (2009), Why Minds Are Not Like Computers, The New Atlantis
(Winter): 4668.
(s) Shagrir, Oron (1999), What Is Computer Science About?, The Monist 82(1):
131149.
A short, but wide-ranging, paper on the nature of computers; also discusses
hypercomputation, analog computation, and computation as not being purely
syntactic
(t) Shannon, Claude E. (1953), Computers and Automata, Proceedings of the
Institute of Radio Engineers 41(10) (October): 12341241.
Discusses the use of computers for non-numerical computing.
(u) Shapiro, Ehud; & Benenson, Yaakov (2006), Bringing DNA Computers to Life,
Scientific American 294(5) (May): 4451.
accessed 24 March 2014 from:
http://www.wisdom.weizmann.ac.il/udi/papers/ShapiroBenensonMay06.pdf
Can DNA be a computer? See also [Adleman, 1998, Qian and Winfree, 2011].
(v) On a related question: If the brain is a computer, is it a Turing machine?:
Sackur, Jerome; & Dehaene, Stanislas (2009), The Cognitive Architecture
for Chaining of Two Mental Operations, Cognition 111: 187211.
Zylberberg, Ariel; Dehaene, Stanislas; Roelfsema, Pieter R.; & Sigman,
Mariano (2011), The Human Turing Machine: A Neural Framework for
Mental Programs, Trends in Cognitive Sciences 15(7) (July): 293300.
2. Websites:
(a) Anderson, David L. (2006), The Nature of Computers, accessed 7 November

2012 from:
http://www.mind.ilstu.edu/curriculum/modOverview.php?modGUI=196
From The Mind Project, http://www.mind.ilstu.edu/ (accessed 7 November
2012).
(b) Chirimuuta, Mazviita; Boone, Trey; & DeMedonsa, Michal (2014), Is Your Brain
a Computer?, Instant HPS (19 September), accessed 23 September 2014 from:
https://www.youtube.com/watch?v=-8q UXpHsaY
(c) Kuczynski, John-Michael (2014), Is the Human Brain a Computer? Alan Turing
on Mind & Computers,
http://www.thecritique.com/articles/is-the-human-brain-a-computer-alan-turing-
on-mind-computers/
Part of a series of online articles providing background for the movie The
Imitation Game. In order to answer his title question, he reviews Turing
1936. Among the interesting, if controversial, points that he makes are that
a recursive function is one that is defined for each of its own outputs and
that the recursive definitions of addition, multiplication, and exponentiation
are the reason that arithmetic requires no thought at all (reminiscent of
[Dennett, 2009] and [Dennett, 2013]s notion of Turings inversion).
(d) Schmidhuber, Jurgen, Zuses Thesis: The Universe is a Computer, accessed
7 November 2012 from:
http://www.idsia.ch/juergen/digitalphysics.html
Zuse was one of the inventors of modern computers; see:
http://ei.cs.vt.edu/history/Zuse.html (accessed 7 November 2012).
Chapter 10
The Church-Turing
Computability Thesis:
I. What Is a Procedure?
Version of 23 December 2014
The Church-Turing Thesis[?]: A problem is computable just in case it

wants to be solved.
anonymous undergraduate student in the authors course, CSE 111,
Great Ideas in Computer Science (2000)1
I believe that history will record that around the mid twentieth century
many classical problems of philosophy and psychology were transformed
by a new notion of process: that of a symbolic or computational process.
[Pylyshyn, 1992, 4]
1 http://www.cse.buffalo.edu/rapaport/111F04.html
411
412 CHAPTER 10. WHAT IS A PROCEDURE?

1. Required:
Cleland, Carol E. (1993), Is the Church-Turing Thesis True?, Minds and

Machines 3(3) (August): 283312.
Preston, Beth (2013), The Centralized Control Model, Ch. 1 of Preston, Beth
(2013), A Philosophy of Material Culture: Action, Function, Mind (New York:
Routledge): 1643.
p. 17, first full paragraph
Introduces three theories of production (producing objects by following
procedures): Aristotles, Marxs, and Diperts.
Aristotle, from p. 17 to p. 18 (end of top paragraph)
Marx, p. 22 (from first full paragraph) to p. 23 (end of first full paragraph)
Dipert, p. 29, first paragraph
The Centralized Control Model, pp. 3032
Control & Improvisation, pp. 3943.
Skim the rest of Ch. 1.
Farkas, David K. (1999), The Logical and Rhetorical Construction of Procedural

Discourse, Technical Communication 46(1) (February): 4254.
http://tinyurl.com/Farkas19992
An interesting look at how to write procedures (i.e., instructions),
from the point of view of a technical writer.
Horsten, Leon; & Roelants, Herman (1995), The Church-Turing Thesis and
Effective Mundane Procedures, Minds and Machines 5(1): 18.
A reply to Cleland 1993.
Cleland, Carol E. (1995), Effective Procedures and Computable Functions, Minds
and Machines 5(1): 923.
A reply to Horsten & Roelants 1995.
Cleland, Carol E. (2001), Recipes, Algorithms, and Programs, Minds and
Machines 11(2) (May): 219237.
Cleland, Carol E. (2002), On Effective Procedures, Minds and Machines 12(2)
(May): 159179.
4. Recommended:
Replies to Cleland:
Israel, David (2002), Reflections on Godels and Gandys Reflections on
Turings Thesis, Minds and Machines 12(2) (May): 181201.
2 http://www.hcde.washington.edu/sites/default/files/people/docs/proceduraldiscourse.pdf
Seligman, Jeremy (2002), The Scope of Turings Analysis of Effective

Procedures, Minds and Machines 12(2) (May): 203220.
Cleland, Carol E. (guest ed.) (2002), Special Issue on Effective Procedures, Minds
and Machines 12(2) (May).
Contains Israel 2002, Seligman 2002, cited above.
Full table of contents accessed 4 February 2014 from:
http://link.springer.com/journal/11023/12/2
10.2 Introduction
Algorithmsincluding procedures and recipescan fail for many reasons; here are
three of them:
1. They can omit crucial steps:
Figure 10.1: 2011,

c Creators Syndicate
2. They can fail to be specific enough (or make too many assumptions):
Figure 10.2: ??,

c Tribune Col. Syndicate
3. They can be highly context dependent or ambiguous (see the first instruction
below):
The general theme of the next few chapters is this: finding faults with each part of
the informal definition of algorithm:
Does it have to be a finite procedure?

Does it have to be effective?
Does it have to halt?

Does it have to solve the problem? (Counterexample: heuristics)
Does it have to be ambiguous/precisely described? (Counterexample: recipes)
Let us take stock of where we are. We proposed two possible answers to the
question of what computer science is. The first was that it is the science (or the study,
or the engineering) of computers, and we have just been considering what a computer
is, examining the history and philosophy of computers. The second possible answer
was that perhaps computer science is the science (or the study) of computing, that is,
of algorithms, and we then investigated what algorithms are, their history, and their
analysis in terms of Turing machines.
But some philosophers have questioned the Turing machine analysis of algorithms,
suggesting that a more general notion, sometimes labeled procedure, is needed. In
this chapter, we will consider some issues related to the environment in which a
computation takes place. We will look at other issuesrelated to hypercomputation
(see below)in Ch. 11.
Recall from 7.6.4 that Churchs thesis is, roughly, the claim that the informal
notion of algorithm or effective computation is equivalent to (or is completely
captured by, or can be completely analyzed as) the lambda-calculus.3 Turings thesis
is, roughly, the claim that the informal notion of algorithm or computability
is equivalent to/completely captured by, or completely analyzed as, the notion of a
Turing machine. Turing proved that the lambda calculus was logically equivalent
3 For an elemetary introduction to the lambda-calculus, see [PolR, 2009, Alonzo Churchs Lambda-
Calculus].
to the Turing machine; that is, he proved that any function that was computable by
Churchs lambda calculus was also computable by a Turing machine (more precisely,
that any lambda-computation could be compiled into a Turing machine)4 and
vice versathat any function that was computable by a Turing machine was also
computable by the lambda-calculus (so that the lambda-calculus and Turing machines
were inter-compilable). Consequently, their theses are often combined under the name
the Church-Turing thesis. [Soare, 2009, 12] has argued that it should be called
simply the Computability Thesis, on the grounds thatgiven the equivalence of all
mathematical models of computation (Churchs, Turings, Godels, etc.)there are
really many such theses. I will call it the Church-Turing Computability Thesis.
In this chapter, we will look at one kind of objection to the thesis, namely, that there
is a more general notionthe notion of a procedure. The objection takes the form
of claiming that there are informally computable tasks that are procedures but that are
not computable by Turing machines.
In the next chapter, we will look at a related, but slightly different kind of objection,
namely, that some non-Turing-machine computable functionssuch as the Halting
Problemcan be computed in a formal, mathematical sense of computation that
goes beyond Turing-machine computation. This is called hypercomputation.
10.3 Preliminary Thought Experiment

Lets begin with a thought experiment, that is, a mental puzzle designed to give you
some intuitions about the arguments to come. Consider this quotation from Herbert
Simon:
Natural science impinges on an artifact through two of the three terms
of the relation that characterizes it: the structure of the artifact itself and
the environment in which it performs.5 Whether a clock will in fact tell
time depends on its internal construction and where it is placed. Whether
a knife will cut depends on the material of its blade and the hardness
of the substance to which it is applied. . . . An artifact can be thought
of as. . . an interface. . . between an inner environment, the substance
and organization of the artifact itself, and an outer environment, the
surroundings in which it operates. If the inner environment is appropriate
to the outer environment, or vice versa, the artifact will serve its intended
purpose. [Simon, 1996b, 6]
And, presumably, otherwise not.
But more can be said, perhaps relating to the third term mentioned in the footnote
to this quotation: Consider a clock. Lewis Carroll (of Alice in Wonderland fame) once
observed that a(n analog) clock whose hands dont work at all is better than one that
loses 1 minute a day, because the former is correct twice a day, whereas the latter is
correct only once every 2 years [Carroll, 1848].
4 I am indebted to John Cases lectures (at SUNY Buffalo, ca. 1983) on theory of computation for this
phrasing.
5 The third term of the relation is an artifacts purpose or goal [Simon, 1996b, 5].
10.4. WHAT IS A PROCEDURE? 417
Here is the thought experiment: What about the clock that has stopped and one that
is consistently 5 minutes slow? The latter is never correctunless you simply move
the numerals on its face 5 minutes counterclockwise; then it would always be correct!6
The (users?) interpretation of the clocks output seems to be what counts.7
10.4 What Is a Procedure?

DISCUSS SHAPIROS NOTION [Shapiro, 2001] HERE
10.5 Challenges to the Church-Turing Computability

Thesis
Some philosophers have challenged the Church-Turing Computability Thesis, arguing
that there are things that are intuitively algorithms but that are not Turing machines.
10.5.1 Carol Cleland: Some Effective Procedures Are Not Turing

Machines
In a series of papers, Carol Cleland has argued that there are effective procedures that
are not Turing machines [Cleland, 1993, Cleland, 1995, Cleland, 2001, Cleland, 2002,
Cleland, 2004]. By effective procedure, she means a mundane (or quotidian, that
is, ordinary or everyday; [Cleland, 2002, 177n.1]) procedure that generates a causal
process, that is, a procedure that physically causes (or effects) a change in the world.8
The bigger issue here concerns the relation of a Turing machine to the interpretation
of its output. We will return to this aspect of the issue in Ch. 17.
According to Cleland, there are three interpretations of the Church-Turing
1. It applies only to functions of integers (or, possibly, also to anything
representable bythat is, codable intointegers).
2. It applies to all functions (including real-valued functions).
3. It applies to the production of mental and physical phenomena, such as is
envisaged by AI or in robotics.
She agrees that it cannot be proved but that it can be falsified.9 It cant be proved,
because one of the two notions that the Church-Turing Computability Thesis says are
6 The actual number is probably closer to 5 12 minutes, but this is only a thought experiment!
7 Recall Searles emphasis on the users interpretation, 9.3.
8 There may be an unintentional pun here. The word effective as used in the phrase effective procedure
is a semi-technical term that is roughly synonymous with algorithmic. The verb to effect, as used in the
phrase to effect a change (in something), is roughly synonymous with the verb to cause, and it is not
directly related to effective in the previous sense. (For the difference between the verb to effect and the
noun an effect (as well as the similar-sounding verb and noun affect), see [Rapaport, 2013, 4.2.0, affect
vs. effect].)
9 Recall our discussion of falsifiability in science, 4.7.1.
equivalent is an informal notion, hence not capable of occurring in a formal proof.

There are two possbilities for why it can be falsified: According to Cleland, the Church-
Turing Computability Thesis can be falsified by exhibiting an intuitively effective
procedure, but not in Turings sense, that is more powerful than a Turing machine
[Cleland, 1993, 285]. Presumably, the qualification but not in Turings sense simply
means that it must be intuitively effective yet not capable of being carried out by a
Turing machine, because, after all, thats what Turing thought his Turing machines
could do, namely, carry out any intuitively effective procedure.
But she also suggests another sense in which the Church-Turing Computability
Thesis might be falsifiable: By exhibiting a procedure that is intuitively effective in
Turings sense yet is not Turing-machine-computable. In other words, there might be
two different kinds of counterexamples to the Church-Turing Computability Thesis: If
Turing were alive, we could show him an intuitively effective procedure that we claim
is not Turing-machine-computable, and he might agree; or we could show him one,
and, either he would not agree that it was intuitively effective (thus denying that it
was a possible counterexample), or he could show that, indeed, it was Turing-machine-
computable (showing how it is not a counterexample at all). Cleland seems to be opting
for the first of these.
Curiously, however, she goes on to say that her mundane procedures are
going to be effective in Turings sense [Cleland, 1993, 286]! In any case, they
differ from Turing-machine-computable procedures by being causal. (When reading
Clelands article, you should continually ask yourself two questions: Are her mundane
procedures causal? Are Turing machines not causal?) Here is her argument, with
comments after some of the premises:
1. A procedure is a specification of something to be followed [Cleland, 1993,

287]. This includes recipes as well as computer programs. Note that recipes can
be notoriously vague (exactly how much is a pinch of salt?) [Sheraton, 1981],
whereas computer programs must be excruciatingly precise.
Her characterization of a procedure as something to be followed puts

a focus on imperatives: You can follow an instruction that says, Do
this! But there are other ways to characterize procedures. Stuart C.
Shapiro [Shapiro, 2001], for example, describes a procedure as a way to
do something. Here, the focus is on the goal or end product; the way to do
itthe way to accomplish that goalmight be to evaluate a function or to
determine the truth value of a proposition, not necessarily to follow an
imperative command.
You should also recall (from 8.10.1) that Turing machines dont normally
follow any instructions! The Turing machine table is a description of the
Turing machine, but it is not something that the Turing machine consults
and then executes. That only happens in a universal Turing machine, but,
in that case, there are two different programs to consider: There is the
program encoded on the universal Turing machines tape; that program
is consulted and followed. But there is the fetch-execute procedure that
constitutes the universal Turing machines machine table; that one is not
10.5. CHALLENGES TO THE CHURCH-TURING COMPUTABILITY THESIS419
consulted or followed.
2. To say that a mundane procedure is effective means, by definition, that
following it always results in a certain kind of outcome [Cleland, 1993, 291].
The semi-technical notion of effective as it is used in the theory of
computation is, as we have seen (7.5), somewhat ambiguous. As Cleland
notes [Cleland, 1993, 291], Minsky calls an algorithm effective if it is
precisely described [Minsky, 1967]. And [Church, 1936b, 50ff; cf. 357]
calls an algorithm effective if there is a formal system that takes a formal
analogue of the algorithms input, precisely manipulates it, and yields a
formal analogue of its output. Churchs notion seems to combine aspects
of both Minskys and Clelands notions.
A non-terminating program (either one that erroneously goes into an
infinite loop or else one that computes all the digits in the decimal
expansion of a real number) can be effective at each step even though
it never halts. Well return to this notion in Ch. 11, when we look at the
notion of interactive computing.
3. The steps of a recipe can be precisely described (hence, they can be effective in
Minskys sense).
This is certainly a controversial claim; we will explore it in more detail in
10.5.2.
4. A procedure is effective for producing a specific output. For example, a
procedure for producing fire or hollandaise sauce might not be effective for
producing chocolate.
In other words, being effective is not a property of a procedure but a
relation between a procedure and a kind of output. This might seem to be
reasonable, but a procedure for producing the truth table for conjunction
might also be effective for producing the truth table for disjunction by
suitably reinterpreting the symbols.10
10 Here is a truth-table for conjunction, using 0 to represent false and 1 to represent true:
0 0 0
0 1 0
1 0 0
1 1 1
Note that, in the 3rd column, which represents the conjunction of the first two columns, there are three 0s
and one 1, which occurs only in the line where both inputs are 1.
And here is the analogous truth-table for disjunction:
0 0 0
0 1 1
1 0 1
1 1 1
Note that the third column has three 1s and only one 0, which occurs only in the line where both inputs
are 0.
Now suppose, instead, that we represent true by 0 and false by 1. Then the first table represents
Here is another example: A procedure that is effective for simulating a

battle in the Civil War might also be effective for simulating a particular
game of chess [Fodor, 1978, 232]. Or a procedure that is effective
for computing with a mathematical lattice might also be effective for
computing with a chemical lattice.11
In cases such as these, the notion of effectiveness might not be Churchs,
because of the possibility of interpreting the output differently. How
important to the notions of (intuitively) effective computation and formal
computation is the interpretation of the output symbols?
5. The effectiveness of a recipe for producing hollandaise sauce depends on causal
processes in the actual world, and these causal processes are independent of the
recipe (the mundane procedure) [Cleland, 1993, 294].
Compare: The effectiveness of a clocks telling the correct time depends
just as much on the orientation of the clock face as it does on the position
of the hands and the internal clockwork mechanism. That is, it depends, in
part, on a situation in the external world. And the orientation of the clock
face is independent of the clocks internal mechanism (or procedure).
6. Therefore, mundane processes can (must?) be effective for a given output P in
the actual world yet not be effective for P in some other possible world.
This is also plausible. Consider a blocks-world computer program that
instructs a robot how to pick up blocks and move them onto or off of other
blocks. I once saw a live demo of such a program. Unfortunately, the robot
failed to pick up one of the blocks, because it was not correctly placed, yet
the program continued to execute perfectly even though the output was not
what was intended.
7. Turing machines are equally effective in all possible worlds, because they are
causally inert.
But here we have a potential equivocation on effective. Turing machines
are effective in the sense of being step-by-step algorithms that are precisely
specified, but they are not necessarily effective for an intended output P: It
depends on the interpretation of P in the possible world!
8. Therefore, there are mundane procedures (such as recipes for hollandaise sauce)
that can produce hollandaise sauce because they result in appropriate causal
processes, but there are no Turing machines that can produce hollandaise sauce,
because Turing machines are purely formal and therefore causally inert. QED
disjunction, and the second one represents conjunction! (A similar point is made in [Peacocke, 1995, 1,
p. 231].)
11 Nicolaas de Bruijn once told me roughly the following anecdote: Some chemists were talking about
a certain molecular structure, expressing difficulty in understanding it. De Bruijn, overhearing them,
thought they were talking about mathematical lattice theory, since everything they said could beand was
interpreted by him as being about the mathematical, rather than the chemical, domain. He told them the
solution of their problem in terms of lattice theory. They, of course, understood it in terms of chemistry.
Were de Bruijn and the chemists talking about the same thing? [Rapaport, 1995, 2.5.1, p. 63]
To the objection that physical implementations of Turing machines could be

causally ert (so to speak),12 Cleland replies as follows: The embodied Turing
machines actions are not physical actions but action-kinds; therefore, they are
causally inert.
But an embodied Turing machine does act: Embodied action-kinds are causal
actions. Alternatively, a Turing machines 0-1 outputs can be interpreted by a device
that does have causal effects, such as a light switch or a thermostat.
Perhaps a procedure or algorithm that is effective for P is better understood as
an algorithm simpliciter. In the actual world, it does P. In some other possible world,
perhaps it does Q (6= P). In yet another possible world, perhaps it does nothing (or
loops forever). And so on. Hence, mundane procedures are interpreted Turing machine
programs, so they are computable.
Lets consider Clelands example of a recipe for hollandaise sauce in more detail:
Lets suppose that we have an algorithm (a recipe) that tells us to mix eggs and oil, and
that outputs hollandaise sauce. Suppose that, in the actual world, the result of mixing
the egg and oil is an emulsion that is, in fact, hollandaise sauce. And let us suppose that,
in possible world w1 whose chemistry differs from that in the actual world, mixing eggs
and oil does not result in an emulsion, so that no hollandaise sauce is output (instead,
the output is a messy mixture of eggs and oil). Compare those situations with the
following: Suppose that we have an algorithm (a recipe) that tells us to mix eggs and
oil until an emulsion is produced, and that outputs hollandaise sauce. In the actual
world, an emulsion is indeed produced, and hollandaise sauce is output. But in w1 , this
algorithm goes into an infinite loop; nothing (and, in particular, no hollandaise sauce)
is ever output.
One problem with this is that the until clause (until an emulsion is produced)
is not clearly algorithmic. How would the computer tell if an emulsion has been
produced? This is not a clearly algorithmic Boolean condition whose truth value can
be determined by the computer simply by checking one of its switch settings (that is,
a value stored in some variable). It would need sensors to determine what the external
world is like. But that is a form of interactive computing, which well discuss in Ch. 11.
10.5.2 Beth Preston: Are Recipes Algorithms?

Introductory computer science courses often use the analogy of recipes to explain what
algorithms are. Recipes are clearly procedures of some kind [Sheraton, 1981]. But are
recipes really good models of algorithms?
Cleland has assumed that they are. Beth Preston [Preston, 2006] has a different
take. She is interested in the nature of artifacts and how they are produced (or
implemented) from plans (such as blueprints). Compare this to how a program (which
is like a plan) is actually executed.
According to Preston, the classical view is that of centralized control. The
etymology of the word control is of interest here: A counter-roll is a duplicate
register; to control something originally meant to faithfully copy it for the sake of
12 Inert comes from the Latin prefix in-, meaning not, and the Latin artem, meaning skill; so, if
inert means lacking causal power, then perhaps the non-word ert could mean having causal power;
see the OEDs entry on inert [http://www.oed.com/view/Entry/94999].
Figure 10.3: 2006,

c Universal Uclick
verification or to regulate it.13 So, to implement a plan is to copy an abstract design

into reality, i.e., to control it.
There is a mental design of the artifact in someones mind. This mental design
specifies all the relevant features of the artifact (the copy) (p. 30), along with a set
of instructions for construction (p. 39). Then the actual construction of the artifact
(that is, the copying of the mental design) is a process that faithfully follows the
instructions of the construction plan, and by so doing reproduces in a material medium
the features of the product specified in the design. This faithful copying relationship
between the design and construction phases of production is the control aspect of the
model (pp. 3031). Compare the way in which a program controls the operations and
output of a computer.
But there is a problem: a faithful copy requires that all relevant features of the
artifact be specified in the design. In other words, the design is ideally supposed to be
an algorithm (effective procedure) for realizing both the construction process and the
product (p. 39). Recipes, however, show that this ideal model doesnt describe reality,
according to Preston.
So, recipes differ from algorithms in several ways:
13 See the OED definitions of control as a noun (http://www.oed.com/view/Entry/40562) and as a verb
(http://www.oed.com/view/Entry/40563).
1. Recipes leave details open (e.g., of ingredients, which play the same role in
recipes that data structures do in programs):
(a) Recipes provide for alternatives: use either yogurt or sour cream.
But couldnt a recipe, or a program for that matter, simply call for a
data-analogue of a typed variable or a subroutine here? The recipe
might call, not for yogurt or else sour cream, but, instead, for a white,
sour, milk-based substance.
And what about non-deterministic algorithms? Here is an example of
a non-deterministic procedure for computing the absolute value of x
using a guarded if control structure:14
if x 0 then return x;
if x 0 then return x;
Here, if x = 0, it does not matter which line of the program is executed.
In such procedures, a detail is left open, yet we still have an algorithm.
Or consider a program that simply says to input an integer, without
specifying anything else about the integer. (As a magician might say,
Think of a number, any number. . . ). This could still be an algorithm.
(b) Recipes specify some ingredients generically: use frosting, without
specifying what kind of frosting.
Here, again, compare typed variables or named subroutines. It does
not matter how the subroutine works; all that matters is its input-output
behavior (what it does). And it doesnt matter what value the variable
takes, as long as it is of the correct type. Typed variables and named
subroutines are generic, yet they appear in algorithms.
(c) Recipes provide for optional ingredients (use chopped nuts if desired).
But compare any conditional statement in a program that is based
on user input. (On the other hand, user input may raise issues for
interactive computing; see Ch. 11.)
2. Recipes leave construction steps (= control structure?) open: For instance, the
order of steps is not necessarily specified (at best, a partial order is given):
cream the butter and sugar, then add the rest of the ingredients (where no order
for adding is given for the rest of the ingredients).
Again, compare non-deterministic statements, such as the guarded-if

command in the above example, or programs written in languages like Lisp,
where the order of the functions in the program is not related in any way
to the order in which they are evaluated when the program is executed. (A
Lisp program is an (unordered) set of functions, not a(n ordered) sequence
of instructions.)
14 In an ordinary if statement, when more than one Boolean condition is satsified, the first one is
executed. In a guarded if statement, it doesnt matter which one is executed. See [Gries, 1981, Ch. 10] or
http://en.wikipedia.org/wiki/Guarded Command Language#Selection: if
3. In recipes, some necessary steps can be omitted (that is, go unmentioned):

put cookies on baking sheet. But should the baking sheet be greased? A
knowledgeable chef would know that it has to be, so a recipe written for such a
chef need not mention the obvious.
But the same kind of thing can occur in a program, with preconditions that
are assumed but not spelled out, or details hidden in a subroutine. (Perhaps
put cookies on baking sheet is a call to a subroutine of the form: grease
baking sheet; put cookies on it.)
4. Recipes can provide alternative ways to do something: roll in powdered sugar

or shake in bag with powdered sugar
Again, non-determinism is similar, as are subroutines: To say multiply

x and y is not to specify how; to say coat in powdered sugar is not to
specify whether this should be done by rolling or shaking.
Preston claims that the cook (i.e., the CPU) is expected to do some of the design
work, to supply the missing parts. So, not everything is in the design. She claims
that cooks dont faithfully follow recipes; instead, they improvise. They can even
change other (fixed) parts of the recipe, because of their background knowledge,
based on experience. For example, they can substitute walnuts for pecans in a recipe
for pecan pie. Therefore, the constructor/executor is typically intelligent, in contrast
to an unintelligent CPU. Here, however, compare an intelligent student hand-tracing
a program. DISCUSS DARWINS DANGEROUS INVERSION? [Dennett, 2009].
Preston notes that the control ideal assumes that the world is stable over time and
homogeneous over agents.
But here is a different interpretation of her analysis: Her centralized control
model might describe an algorithm together with a CPU producing a process (i.e.,
an algorithm being executed). But her theory of collaborative improvisation might
describe a specification together with a programmer producing an algorithm. That is,
the development of an algorithm by a programmer from a specification is improvisatory
and collaborative, but the output of an algorithm by a computer from executing the
algorithm is centralized control.
So, recipes are more like design specifications for computer programs than they are
like algorithms. In fact, my counterexamples to differentiate between algorithms and
recipes just show that either recipes are high-level programs requiring implementation
in a lower level, or that recipes are specifications.
10.5.3 Discussion
The Church-Turing Computability Thesis, in any of its various forms, states that the
informal notion of effective computation (or algorithm, or whatever) is equivalent to the
formal notion of a Turing-machine program (or a -definable function, or a recursive
function, or whatever). The arguments in favor of the Church-Turing Computability
Thesis are generally of the form that (1) all known informal algorithms are Turing-
machine-computable15 and (2) all of the formal, mathematical versions of computation

are logically equivalent to each other (see 7.6.4, above).
It has also been argued that the Thesis cannot be formally proved because one
side of it is informal, hence not capable of being part of a formal proof. Some have
suggested (Gandy CITE, Sieg CITE, and [Dershowitz and Gurevich, 2008]) that the
thesis is capable of being proved by providing a set of formal axioms for the informal
notion, and then proving that Turing machines satisfy those axioms. Although this is an
interesting exercise, it is not obvious that this proves the Church-Turing Computability
Thesis. Rather, it seems to replace that Thesis with a new one, namely, that the informal
notion is indeed captured by the formal axioms. But that thesis likewise cannot be
proved for the same reason that the Church-Turing Computability Thesis cannot: To
prove it would require using an informal notion that cannot be part of a formal proof.
Others have argued that neither (1) nor (2) are even non-deductively good
arguments for the Church-Turing Computability Thesis. Against (1), it can be argued
that, just because all known informal algorithms are Turing-machine-computable, it
does not follow that all informal algorithms are. After all, just because Aristotles
theory of physics lasted for some 2000 years until Newton came along, it did not follow
that Aristotles physics was correct, and just because Newtons theory lasted for some
200 years until Einstein came along, it did not follow that Newtons theory was correct.
So, there is no inductive reason to think that the Church-Turing Computability Thesis
is correct any more than there is to think that Einsteins theory is.16
Against (2), it can be argued that, even though the common theme underlying the
equivalence of Turing machines, -definability, recursive functions, etc., is robust
and of great mathematical interest, that is not reason enough to think that there might
not be any other theory of effective computation. (And, indeed, there are alternatives;
see Ch. 11.)
So, are there good reasons for seriously doubting the Church-Turing Computability
Thesis? We have just seen two candidates: Cleland argues that certain mundane
procedures are effectively computable but not Turing-machine-computable, and
Preston suggests that certain recipe-like procedures of the sort typically cited as
examples of effective procedures are not really algorithmic.
But against Clelands example, we have seen that there may be a problem in how
one determines what the proper output of an algorithm is, or, to put it another way, what
the problem that an algorithm is supposed to solve is. Consider a recipe for hollandaise
sauce that, when correctly executed on Earth, produces hollandaise sauce but, when
correctly executed on the moon, does not. Is that recipe therefore not Turing-machine-
computable? It would seem to be Turing-machine-computable on earth, but not on the
moon. Or is it Turing-machine-computable simpliciter (for example, no matter where
it is executed), but conditions having nothing to do with the algorithm or recipe itself
conspire to make it unsuccessful as a recipe for hollandaise sauce on the moon? Is the
algorithm or recipe itself any different?
15 This puts it positively. To put it negatively, no one has yet found a universally convincing example of an
informally computable function that is not also Turing-machine computable

16 On the question of whether any scientific theory is correct, on the grounds that they are all only
falsifiable, see CITE.

And against Prestons example, we have seen that recipes are, perhaps, more like
specifications for algorithms than they are like algorithms.
In the next chapter, we will look at other potential counterexamples to the Church-
Turing Computability Thesis.

1. Shapiros more general notion of procedure (Farkas 1999 may be relevant here,
and some of this may more properly belong in Ch. 3):
[Shapiro, 2001] characterizes procedure as the most general term for
the way of going about the accomplishment of something , citing the
Merriam-Webster Third New International Dictionary (see http://www.merriam-
webster.com/dictionary/procedure). This includes serial algorithms, parallel
algorithms (which are not step by step, or serial), operating systems (which
dont halt), heuristics (which are not guaranteed to produce the correct
answer), musical scores, and recipes. Thus, Turing machines (or Turing-
machine programs), that is, algorithms as analyzed in 7.5, are a special case
of procedures. Given this definition, he claims that procedures are natural
phenomena that may be, and are, objectively measured, principally in terms of
the amount of time they take. . . and in terms of the amount of resources they
require. He gives no argument for the claim that they are natural phenomena,
but this seems reasonable. First, they dont seem to be social phenomena in
the sense in which institutions such as money or governments are (cite Searle).
Some of them might be, but the concept of a procedure in general surely seems
to be logically prior to any specific procedure (such as the procedure for doing
long division or for baking a cake). Second, there are surely some procedures
in nature, such as a birds procedure for building a nest. Insofar as a natural
science is one that deals with objectively measurable phenomena (again citing
Webster; http://www.merriam-webster.com/dictionary/natural%20science), it
follows trivially that computer science is the natural science of procedures.
In this chapter, we are focusing on this more general notion of procedure
2. On the relation between an algorithm and its output:
Fregean senses dont determine reference (because of the possibility of multiple
realization, or slippage between sense and reference [Putnam, 1975]). But
senses do tell you how to find a referent: They are algorithms. The referent is
like the purpose or task of the algorithm. But there is always the possibility of
slippage: An algorithm can be intended for one purpose (say, playing chess),
but be used for another (say, simulating a Civil War battle [Fodor, 1978, 232])
by changing the interpretation of the inputs and outputs. (But discuss Rescorla!;
see also [Pylyshyn, 1992].)
Here is another relevant quotation:
[John von Neumanns]. . . basic insight was that the geometry of
the vectors in a Hilbert space has the same formal properties as
the structure of the states of a quantum-mechanical system. Once
that is accepted, the difference between a quantum physicist and
a mathematical operator-theorist becomes one of language and
emphasis only. [Halmos, 1973, 384]
First, this is a nice statement of mathematical structuralism. Second, this is
also a nice statement of the kind of phenomenon illustrated by Fodors Civil
War-chess example. A computer program designed to operate on Hilbert-space

vectors could be used or adapted to operate on quantum-mechanical states with
no change, except possible for input-output coding. It also illustrates where the
intensionality of computation might be relevant: A Civil War or Hilbert-space
program might be intensionally distinct from a chess or quantum-mechanics
program, yet they could be extensionally equivalent.
3. Re: Cleland:
[Sloman, 2002, 3.2] makes a possibly useful distinction between internal and
external processes: The former include manipulation of cogs, levers, pulleys,
strings, etc. The latter include movements or rearrangements of various kinds
of physical objects. So, a computer on earth that is programmed to make
hollandaise sauce and one on the moon that is identically programmed will have
the same internal processes, but different external ones (because of differences
in the external environment).
A related distinction is that between competence and performance: A
computer might be competent to make hollandaise because of its internal
processes, yet fail to do so because of performance limitations due to external-
environmental limitations.
4. Does the ability of a machine to do something that is non-TM-computable mean
that it can compute something that is non-TM-computable? What does physical
performance have to do with computation? Surely we want to say that whatever
a robot can do is computable, even if that includes cooking. But surely thats
because of the internal processes, not the external ones.
5. . . . Turing machines are not so relevant [to AI] intrinsically as
machines that are designed from the start to have interfaces to external
sensors and motors with which they can interact online, unlike Turing
machines which at least in their main form are totally self contained,
and are designed primarily to run in ballistic mode once set up with
an intial machine table and tape configuration. [Sloman, 2002, 4.2]
This seems to be a distinction between abstract TMs and robots. And Clelands
arguments seem more relevant to robots than to TMs, hence have nothing really
to say about the Church-Turing Computability Thesis (which only concerns
TMs).
6. There may be a connection between Clelands use of hollandaise recipes to
counter the Church-Turing Computability Thesis, Prestons use of recipes as
(bad?) examples of algorithms, and Wegners emphasis on interactive computing
as a counterexample to CTCT:
One example of a problem that is not algorithmic is the following

instruction from a recipe [quote Knuth, 1968]: toss lightly until
the mixture is crumbly. This problem is not algorithmic because
it is impossible for a computer to know how long to mix: this may
depend on conditions such as humidity that cannot be predicted with
certainty ahead of time. In the function-based mathematical world-

view, all inputs must be specified at the start of the computation,
preventing the kind of feedback that would be necessary to determine
when its time to stop mixing. [Goldin and Wegner, 2005], as cited in
[Cooper, 2012, 78].
7. On Following Rules THIS PARA NEEDS TO BE CLARIFIED:

What does it mean to follow a rule? Consider three examples of following
rules: intelligently hand-tracing a program, crossing at a green light, and obeying
the law of gravity (as a cartoon once said, its not just a good idea; its the law!).
In the first example, you are literally following the programs rulesthe program
tells you to do A, so you do A (and if it tells you how to do A, then you do A in
that way. Compare how a universal Turing machine executes a program stored
on its tape.
In the second example, you know that there is a rule (a legal law) that says that
you are only supposed to cross the street when you have a green light. You could
choose to jaywalk, but you decide to obey the law (to follow the rule). But this
is different from obeying the (descriptive) law of gravity; with that law, you
have no choice. You simply behave in accordance with the law of gravity.
Compare describing these as following a rule with describing them as
behaving in accordance with a rulea rule that describes independent
behaviorand with executing a rule in the way that a Turing machine does. On
the latter, compare executing a rule the way that a Turing machine does with the
following rule (which can, but need not, be described in advance as having the
goal of adding two numbers in a spreadsheet): enter a number in cell 1; enter a
number in cell 2; enter the string =sum(cell1,cell2) in cell 3 (or, even better?:
enter a number in cell1; enter a number in cell2; enter the string =hclick on
cell1i;hclick on cell2i in cell3), and compare these with intelligently adding
by following the normal addition algorithm. The former is Turings strange
inversion [Dennett, 2009].17
8. The following quote can go either here or in an earlier chapter on the CTCT. It
is Turings statement of one of the principle justifications for the thesis.
Several definitions have been given to express an exact meaning

corresponding to the intuitive idea of effective calculability as
applied for instance to functions of positive integers. The purpose of
the present paper is to show that the computable functions introduced
by the author are identical with the -definable functions of Church
and the general recursive functions due to Herbrand and Godel and
17 This is based on a real-life example. One of the TAs for a Great Ideas in Computer Science course that
I taught (aimed at students who were computer-phobic) was trying to show me how a spreadsheet worked
(at the time, I had never used one, so, with respect to this, I was as naive as any of the students in the
course!). She gave me those instructions, but I had no idea why I was supposed to do those things. Yet
I could faithfully follow her instructions without understanding them (that is, without understanding their
purpose); cf. [Dennett, 2009].
developed by Kleene. It is shown that every -definable function is

computable and that every computable function is general recursive.
. . . If these results are taken in conjunction with an already available
proof that every general recursive function is -definable we shall
have the required equivalence of computability with -definability. . . .
...
The identification of effectively calculable functions with
computable functions is possibly more convincing than an
identification with the -definable or general recursive functions. For
those who take this view the formal proof of equivalence provides a
justification for Churchs calculus, and allows the machines which
generate computable functions to be replaced by the more convenient
-definitions. [Turing, 1937, 153]
Turing cites [Church, 1936b] for the definition of -definability, [Kleene, 1936a]
for the definition of general recursiveness, and [Kleene, 1936b] for the proof
of their equivalence. Godel was among those who found Turings computable
functions to be more convincing than the other formalisms.
9. CTCT can be viewed as an explication of the informal notion of effective
calculability or algorithm [Montague, 1960, 430]. An explication is the
replacement of an informal or vague term with a formal and more precise
one.18 [Mendelson, 1990, 229] calls it a rational reconstruction (a term also
due to Carnap): a precise, scientific concept that is offered as an equivalent
of a prescientific, intuitive, imprecise notion. Mendelson goes on to argue
that CTCT has the same status as the definition of a function as a certain set
of ordered pairs (see 7.4.1, above) and other mathematical notions (logical
validity, Tarskis definition of truth, the - defintion of limits, etc.). Mendelson
then claims that it is completely unwarranted to say that CT is unprovable
just because it states an equivalence between a vague, imprecise notion. . . and
a precise mathematical notion [Mendelson, 1990, 232]. One reason he gives
is that both sides of the equivalence are equally vague! He points out that the
concept of set is no clearer than that of function. Another is that the argument
that all Turing-machine programs are (informally) computable is considered to
be a proof, yet it involves a vague, informal concept. (Note that it is the claim
that all informally computable functions are Turing-machine computable that is
usually considered incapable of proof on these grounds.)
By the way, Mendelsons paper has one of the clearest discussions of the nature
of CTCT.
For responses to Mendelson, see Bringsjord 1993, [Shapiro, 1993].
10. Not sure where this belongs (this chapter? chapter on hypercomputation?):
[Montague, 1960, 433] suggests thatfor a more general notion of computation
than a mere Turing machine (one that would apply to both digital and analog
computation)a computer needs an output signal that indicates when the
18 The concept is due to the philosopher Rudolf Carnap; see, for example, [Carnap, 1956, 78].
computation is finished. In Turings theory of computation, the machine simply

halts. But, of course, how do you know that a machine has halted rather than
merely being in an infinite loop?
11. On the notion of a procedure:

[Simon, 1962, 479] offers two kinds of descriptions of phenomena in the world:
state descriptions and process descriptions. The former characterize the world
as sensed; they provide the criteria for identifying objects. . . . The latter
characterize the world as acted upon; they provide the means for producing
or generating objects having the desired characteristics given by a state
description. His example of a state description is A circle is the locus of all
points equidistant from a given point; his example of a process description is
To construct a circle, rotate a compass with one arm fixed until the other arm has
returned to its staring point. Process descriptions describe procedures. (State
descriptions seem to be part of science, whereas process descriptions seem to
be part of engineering.)
12. Another argument against CTCT (which I wont elaborate on):

[Bowie, 1973] argues that Churchs thesis is false on the grounds that the
informal notion of computability is intensional but the notion of recursive
functions is extensional (see p. 67). (On the difference between intensional and
extensional, see [Rapaport, 2009].) He also argues against it on what would
now be considered hypercomputational grounds, roughly, that, if you give a
computable function a non-computable input (intensionally represented), it will
compute a non-computable output (see p. 75).
[Ross, 1974] is a reply to [Bowie, 1973]: He points out that Bowies definition of
computable is non-standard, and argues that the standard definition (essentially
the one given in Ch. 7) is not intensional.
13. Gandys statement of Churchs thesis:

What is effectively calculable [by an abstract human being using some
mechanical aids (such as paper and pencil)] is computable, where
computable. . . mean[s] computable by a Turing machine , abstract
indicates that the argument makes no appeal to the existence of practical limits
on time and space [and]. . . effective in the thesis serves to emphasize that the
process of calculation is deterministicnot dependent on guessworkand that
it must terminate in a finite time. [Gandy, 1980, 123124]
Gandy goes on to be concerned with a mechanical version of CTCT, whether
What can be calculated by a machine is computable [Gandy, 1980, 124], where
by machine he says that he us using the term with its nineteenth century
meaning; the reader may like to imagine some glorious contraption of gleaming
brass and polished mahogany, or he [sic] may choose to inspect the parts of
Babbages Analytical Engine which are preserved in the Science Museum at
South Kensington. [Gandy, 1980, 125]. One difference between a human (even
an abstract one) and a machine is that the latter can easily perform paralle
operations such as printing an arbitrary number of symbols simultaneously
[Gandy, 1980, 125]. We will not explore this issue here, but some references are
listed in the Further Sources (also see [Deutsch, 1985], who discusses CTCT
in the context of quantum computation).
14. Godels objection to the CTCT:
Godel has objected, against Turings arguments, that the human mind
may, by its grasp of abstract objects [by insight and imaginative
grasp], be able to transcend mechanism. [Gandy, 1980, 124]
As Gandy notes, we need to know much more about the nature and limits of the
human mind than we do in order to adequately evaluate this objection. After all,
what seems to be non-mechanical intelligence need not be. A more detailed
reply to Godel is given in [Kleene, 1987, 492494], the bottom line of which is
this:
The notion of an effective calculation procedure or algorithm
(for which I believe Churchs thesis) involves its being possible to
convey a complete description of the effective procedure or algorithm
by a finite communication, in advance of performing computations in
accordance with it. (p. 493.)
And, according to Kleene, Godels objection fails to satisfy that criterion (as does
the objection raised in [Kalmar, 1959]).
15. On the importance of a physical CTCT:
[Newell, 1980, 148] points out that physical systems are limited in speed, space,
and reliability. Hence, a physical device that implements a Turing machine
will inevitably be more limited than that abstract, ideal machine. So, some
functions that are theoretically computable on a Turing machine might not be
physically computable on a real computer. Hence, a physical version of the
CTCT is going to differ in plausibility from the abstract CTCT. The abstract
version can be made slightly more realistic, perhaps, by redefining, or placing
limits on, some of the terms:
[R]ather than talk about memory being actually unbounded, we will
talk about it being open, which is to say available up to some
point, which then bounds the performance, both qualitatively and
quantitatively. Limited, in opposition to open, will imply that the limit
is not only finite, but small enough to force concern. Correspondingly,
universal can be taken to require only sufficiently open memory, not
unbounded memory. [Newell, 1980, 161]
16. According to many, a computer simulation of a hurricane is not a hurricane,
because it does not get people wet CITE. But it could get simulated people
simulatedly wet. The difference is between a real hurricane and a simulated
one has to do, in part, with the nature of the inputs and outputs. But, as
[Carleton, 1984, 222223] notes, The input to a fire simulation is not oxygen,
tinder, and catalyst. That is, it is not the same input as goes into a natural
system which produces fire. . . . [I]t is by virtue of dealing in the right kinds
of input and output that one system can play the role in a situation that is
played by a system we acknowledge to literally undergo the [activity]. . . our
system simulates. Clelands hollandaise-sauce-making program may differ in
output when executed on the moon than on earth; it has the wrong output. But
a hurricane-simulator, a fire-simulator, and a hollandaise-sauce program each
exhibit their relevant behaviors if you ignore the input and the output. So, the
question becomes how central to what it is that a computer program is (supposed
to be) doing is the nature of the inputs and outputs (in general, of the environment
in which it is being executed)? This is relevant to the issues concerning embodied
cognition as well as to Rescorlas emphasis on semantics in computation.
How important are the input and output? Here is a definition from
[Deutsch, 1985, 2]:
Two classical deterministic computing machines are

computationally equivalent under given labellings of their input and
output states if they compute the same function under those labellings.
But wouldnt they be computationally equivalent (in a less-restricted sense) if

those labellings were ignored? On this definition, Fodors Civil War and chess
computers are not computationally equivalent. Rescorla, on the other hand,
might favor this definition.
17. [Goodman, 1987, 482] raises the question of whether the CTCT is a reduction
of the informal mathematical notion of algorithm. . . [to] the formal set-theoretic
notion of a Turing machine program. The former is intensional in nature,
because An algorithm, in the informal mathematical sense, is a specific
procedure for solving a particular kind of mathematical problem (my italics; see
also p. 483). But the latter is extensional, because the Turing machine program
does not tell you what the program is for. . . . [Only the] documentation contains
the intensional content which is missing from the bare machine code. . . and
brings the program closer to the algorithm which it is intended to implement
(p. 483). This seems to nicely capture Fodors Civil War-chess puzzle: That
a particular program is a computational model of a Civil War battle rather
than of a chess match is a matter of its intensional description; extensionally,
they are the same program (modulo input-output encoding). Ah, but perhaps
that is where the intensionality can be put back in the Turing machine: in the
input-output encoding. But that is not really part of the machine; it is part
of a users interaction with the machine or how a user treats or interprets the
machines output, how a user interprets what the machine is doing. Here, what
the machine is doing is potentially ambiguous between, say, modeling a Civil
War battle (a description from Dennetts intentional stance) and some design-
stance description that would be neutral between modeling a Civil War battle
and modeling a chess match (see [Goodman, 1987, 483] for a very similar
discussion).
However, Goodman does not believe that the CTCT is
an analysis of the informal concept of algorithm. It at most

provides a necessary condition for the existence of an algorithm.
That is, a problem which no Turing machine can solve cannot be
solved algorithmically. However, a Turing machine program without
additional explanation is not an algorithm, and an algorithm is not as
it stands a Turing machine program. . . . My contention is rather that
not all of the content of our informal intensional talk about algorithms
is captured by extensional talk about Turing machine programs.
From this point of view, then, there is nothing reductionistic about
Churchs thesis, since it does not assert that any previously known
objects can be eliminated in favor of any new objects.
18. Churchs superthesis, an assertion that for any algorithm

whatsoever, there is a Turing machine that computes the same
function the same way [Shapiro, 1993, 65], citing Kreisel, Georg
(1969), Some Reasons for Generalizing Recursion Theory, Logic
Colloquium 69 (Amsterdam: North-Holland): 138198.
19. On Rescorlas emphasis on semantics in computation, and on Clelands emphasis

on the stated goal of a computation, compare this passage from Piccinini:
[Stored-program computers have the ability to respond to (non-

semantically individuated) strings of tokens stored in their memory by
executing sequences of primitive operations, which in turn generate
new strings of tokens that get stored in memory. [Note that this
is basically a description of how computers work, or of what
computation isnote added by WJR.] Different bits and pieces [that
is, substringsWJR] of these strings of tokens have different effects
on the machine. . . . An accurate description of how tokens can be
compounded into sub-strings, and sub-strings can be compounded
into strings, which does not presuppose that the strings of tokens
have any content, may be called the syntax of the system of strings
manipulated by the computer. [Compare my description of syntax
in 14.4 and 19.5.3.2WJR.] . . . [T]he effect of a string on the
computer is assigned to it [that is, the stringWJR] as its content.
This assignment constitutes an internal semantics of a computer. An
internal semantics assigns as contents to a system its own internal
components and activities, whereas an ordinary (external) semantics
assigns as contents to a system objects and properies in the systems
environment. . . . None of this entails that computer languages have
any external semantics, i.e. any content. . . , although it is compatible
with their having one. . . .
[In order to understand computing mechanisms and how they
work (as opposed to why they are built and how they are used), there
is no need to invoke content. . . . [Piccinini, 2004b, 401402, 404] (see
also [Piccinini, 2006a, 2]
On this view, compatible with my own, it is the internal workings of the

computer that count, not the external interpretation of its inputs and outputs
(or even the external interpretation of its internal mechanisms or symbol
manipulations). This is the sense in which a Civil War computer and a chess
computer are computing the same thing.
Note the hedge in the last sentence of Piccininis quote: Cleland and Rescorla
might be quite right in terms of their emphasis on why or how a particular
computer or program is being used. Thats an intentional/intensional aspect of
computation, but doesnt necessarily vitiate the CTCT.
20. Related to the above point is Slomans claim that one of the primary features of
computers (and of brains) is Coupling to environment via physical transducers
[Sloman, 2002, 5, #F6, pp. 1718]. This allows for perceptual processes that
control or modify actions and is how internal information manipulation often
leads to external behaviour. The former (perceptual input) would seem to
be relevant to interactive computing (see Ch. 11). The lattersomething also
emphasized by [Shapiro, 2001]is related to the issues that Cleland discusses.
A computer without one or the other of these would be solipsistic. But
computation can be kept separate from interaction with the world. It is not
that the latter is unimportant or secondary, but that it is a separate thing. (So,
insofar as Cleland-like arguments work against the CTCT, perhaps there are two
versions of the CTCT: an internal or purely syntactic one, which is true, and
an external or semantic one, that might be false.
In fact, [Sloman, 2002, 3.2, p. 9] says as much:
In all these machines [that is, precursors to computersWJR] we
can, to a first approximation, divide the processes produced by the
machine into two main categories: internal and external.
The former correspond to what [Piccinini, 2004b] and I would call purely
syntactic computation; the latter correspond to the kinds of things that Cleland
is more concerned with.
And [Sloman, 2002, 4.2, p. 12] believes that Turing machines are, therefore,
less relevant to AI than machines that are designed from the start to have
interfaces to external sensors and motors. . . , unlike Turing machines which at
least in their main forma re totally self contained, and are designed primarily
to run in ballistic mode once set up with an initial machine table and tape
configuration.
21. On Clelands problem:
According to [Piccinini, 2006a, 2],
In the mathematical theory of computation, abstract computing
mechanisms are individuated by means of formal descriptions,
some of which are called programs. Programs and other formal
descriptions of computing mechanisms specify which inputs may
enter the mechanism, how the inputs affect the internal states of the
mechanism, and which outputs come out of the mechanism under

which conditions. Inputs and outputs are strings of letters from a
finite alphabet, often called symbols.
In computability theory, symbols are typically marks on paper
individuated by their geometrical shape (as opposed to their semantic
properties). Symbols and strings of symbols may or may not be
assigned an interpretation; if they are interpreted, the same string may
be interpreted differently. . . . In these computational descriptions,
the identity of the computing mechanism does not hinge on how the
strings are interpreted.
By individuated, Piccinini is talking about how one decides whether what

appear to be two programs (say, one for a Civil War battle and one for a chess
match) are, in fact, two distinct programs or really just one program (perhaps
being described differently). Here, he suggests that it is not how the inputs
and outputs are interpreted (their semantics) that matters, but what the inputs
and outputs look like (their syntax). So, for Piccinini, the Civil War and chess
programs are the same. For Cleland, they would be different. For Piccinini, the
hollandaise-sauce program running on the Moon works just as well as the one
running on Earth; for Cleland, only the latter does what it is supposed to do.
So, the question Which Turing machine is this? has only one answer, given in
terms of its syntax (determined by [its] instructions, not by [its] interpretations
[Piccinini, 2006a, 2]. But the question What does this Turing machine do?
has n + 1 answers: one syntactic answer and n semantic answers (one for each
of n different interpretations).
One is tempted to ask which of these is the right way to look at it. But it is
important to see that both are right in different contexts. If one is concerned only
with the internal mechanisms, then Piccininis way of looking at things is more
appropriate. If one wants to buy a chess program (but not a Civil War program)
or a hollandaise-sauce maker, then Clelands way is more appropriate.
22. On [Rescorla, 2007]:
(a) A Turing machine manipulates syntactic entities: strings

consisting of strokes and blanks. . . . Our main interest is
not string-theoretic functions but number-theoretic functions.
We want to investigate computable functions from the natural
numbers to the natural numbers. To do so, we must correlate
strings of strokes with numbers. (p. 253.)
This focuses the issues very clearly. Which are we really interested
in: syntactic/symbolic/numeral computation or semantic/ontological/-
number computation? Piccinini is interested in the former, Cleland in the
latter. Hilbert, I think, was interested in the former, for, after all, we humans
can only do the latter via the former. Is that a limitation? Perhaps, but
it also gives us a freedom, because symbols can represent anything, not
just numbers. Can numbers represent things? Yes, but only insofar as we
humans who use the numbers to represent things are thereby using them as
if they were symbols.
Is computation just about numbers? Functions are not limited to numbers,
so why should computation be thus limited? The broader notion of
computation (not limited to numbers) includes number computation.
(b) Lets grant Rescorlas point that (semantic) computation of
number-theoretic functions is distinct from (syntactic) computation of
numeral/symbolic-theoretic functions in the Turing-machine sense. After
all, Turing machines manipulate marks on paper, not numbers (at least, not
directly). But now lets imagine a Platonic, number-theoretic, Turing-like
machine as follows: The machine contains an infinite supply of boxes,
each of which can contain a natural number. (If you prefer, each box can
only contain the number 1, and other natural numbers would be represented
by sequences of boxes containing 1s.) The machine can look into a box to
see what number (if any) is in it, and it can place any number in any box
(putting a number in a box that already has a number in it just replaces
the previous number). It can move one box to the left or one box to
the right of its current location. In other words, it can do exactly what a
normal Turing machine can do, except that, instead of dealing with marks
on paper, it deals with numbers in boxes. If we consider an abstract Turing
machine to be a (perhaps set-theoretically defined) mathematical structure,
then both numeral-based and number-based Turing machines are (abstract)
Turing machines. The CTCT can be interpreted to be saying that informal
computation is Turing-machine computation in any of these three senses of
Turing machine. Is there anything a number-theoretic Turing machine can
or cant do that a numerical-based Turing machine can or cant do? If they
are equally powerful (which I believe that they are), then I dont think
that Rescorlas argument here is significant.
On second thought, maybe not. The version of a number-theoretic Turing
machine that allows any real number to be placed in a box may be too
powerful: How would you add two numbers? I dont see how you could
unless you had addition as a primitive operation (which it isnt). On the
other hand, if you only allow the number 1 to be placed in a box, and you
represent other numbers by sequences of boxes of 1s, then you just have
a version of the numeral-theoretic Turing machine with the number 1 in
a box representing a mark on paper (or representing the numeral 1). In
that case, the number-theoretic Turing machine is just a very odd kind of
numeral-theoretic Turing machine.
(c) A nice example of the which Turing machine is this?/what does this Turing
machine do? contrast is this example of Rescorlas. (Note: The symbol
x represents a sequence of x strokes, where x is a natural number.)
Different textbooks emply different correlations between Turing
machines syntax and the natural numbers. The following three
correlations are among the most popular:
d1 (n) = n.
d2 (n + 1) = n.
d3 (n + 1) = n, as an input.
d3 (n) = n, as an output.
A machine that doubles the number of strokes computes f (n) =

2n under d1 , g(n) = 2n + 1 under d2 , and h(n) = 2n + 2 under
d3 . Thus, the same Turing machine computes different numerical
functions relative to different correlations between symbols and
numbers. [Rescorla, 2007, 254].
Do we really have one machine that does three different things? Piccinini
would agree that there is only one machine. What it does (in one sense of
that phrase) depends on how its input and output are interpreted, that is, on
the environment in which it is working. In different environments, it does
different things, or so Cleland would say.
(d) . . . Turings analysis concerns only human mechanical activity,
not general mechanical activity. . . . For Turing, computable
means computable by a human, not computable by some
possible machine. [Rescorla, 2007, 257].
Exactly! This is what Hilbert was concerned with: How can we humans
compute answers to certain problems? We can only directly compute on
strings, not numbers (we can only indirectly compute on numbers via direct
computation on strings).
(e) [Rescorla, 2007, 272274] argues against exactly this point:
One argument runs as follows: humans and computers directly
manipulate symbols, not numbers; thus, what humans and
computers really compute are string-theoretic functions, not
number-theoretic functions. . . . The argument is fallacious.
. . . At best, the premise establishes that our computations of
number-theoretic functions are mediated by our computations
of string-theoretic functions. It does not follow that all we
really. . . compute are string-theoretic functions. To conclude this
would be analogous to the inference sometimes drawn by the
British empiricists that, since our ideas mediate our perception
of the external world, all we really perceive are our ideas.
Now, I happen to agree with the British empiricists, so I disagree with
Rescorla. But Rescorlas broader point is that, to the extent that we
want to compute over numbers, not numerals, we need a (computable)
interpretation function between the numbers and the numerals.
(f) Against the Hilbertian point, Rescorla is less convincing. It strkes me

as rather slanted (p. 273) is the best he can do. Essentially, his point
comes down to this: Even if Hilbert, Turing, et al., really were only
interested in syntactic computation, that doesnt mean that we shouldnt
also be interested in semantic computation.
ADD RESCORLA TO RECOMMENDED READING FOR THIS
CHAPTER? AND PICCININI?
(g) . . . Turings analysis concerns human mechanical activity, not
human cognition in general. [Rescorla, 2007, 257].
To a certain extent, this is correct. The extent to which is might be
misleading is this: It remains an open question whether human cognition in
general can be completely captured by human mechanical activity. This is
the topic of computational cognitive science, which well explore in Ch. 19.
Rescorla continues:
[Turings] constraints. . . do not purport to goven all possible
processes for determining some functions value. Imagine
a computor [that is, a human clerk who is performing a
computation] who possesses a mystgerious cognitive faculty,
which enables him [sic] to determine some uncomputable
functions value upon any input. . . . By employing the faculty,
the computor introduces an essentially nonmechanical element
into his mathematical activity. He implements a nonalgorithmic
cognitive strategy. . . . But we can expunge them from our account
of computation.
This faculty is precisely the kind of oracle that Turing introduced in his
doctoral thesis. But computation relative to an oracle can still be considered
to be a kind of computation, only one that adds to the set of primitive
operations such an extra faculty. Well return to this in Ch. 11.
(h) [Copeland and Proudfoot, 2010] is a reply to [Rescorla, 2007], and is a
good paper in general on the problem of Turing machines being syntactic.
They argue that deviant encodings. . . that appear to enable Turing
machines to perform impossible tasks, such as solving the halting
problem (p. 247, col. 1) can be avoided by respecting Turings restriction
that the input tape must be empty [Turing, 1936, 232]:
In effect, the blank-
tape restriction formalizes the requirement. . . that the computer
be allowed no access to numbers or statements produced by an
oracle or by a human mathematician working creatively. The
blank-tape restriction provides a simple and elegant solution to
the problem of deviant I[nput]-encodings. All I-encodings are
precluded, deviant or otherwise. Prior tape-markingsinputs of
any formare simply banned where the objective is to provide a
formal analysis of computability. (p. 250, col. 1)
Note that this blocks a Turing machine from being a hypercomputer (see
Ch. 11.
23. Here is another nice example for the Cleland issue:
a loom programmed to weave a certain pattern will weave that pattern

regardless of what kinds of thread it is weaving. The properties of
the threads make no difference to the pattern being weaved. In other
words, the weaving process is insensitive to the properties of the
input. [Piccinini, 2008, 39]
As Piccinini goes on to point out, the output might have different colors
depending on the colors of the input threads, but the pattern will remain the
same. The pattern is internal; the colors are external, to use the terminology
above. If you want to weave an American flag, you had better use red, white, and
blue threads in the appropriate ways. But even if you use puce, purple, and plum
threads, you will weave an American-flag pattern. Which is more important: the
pattern or the colors? Thats probably not exactly the right question. Rather, the
issue is this: If you want a certain pattern, this program will give it to you; if you
want a certain pattern with certain colors, then you need to have the right inputs
(you need to use the program in the right environment). (This is related to issues
in the philosophy of mathematics concerning structuralism: Is the pattern, or
structure, of the natural numbers all that matters? Or does it also matter what the
natural numbers in the pattern really are? CITE)
24. This could go either in this chapter or earlier (Ch. 7?):

Its worth comparing the explication of the informal (or folk?) notion of
algorithm as a Turing machine (or recursive functions, etc.), with the attempt
to define life in scientific terms.
(a) As with any attempt at a formal explication of an informal concept (as

we discussed in Chs. 3 and 9), there is never any guarantee that the
formal explanation will satisfactorily capture the informal notion (usually
because the informal notion is informal or fuzzy). The formal explication
might include some things that are, pre-theoretically at least, not obviously
included in the informal concept, and it might exclude some things that
are, pre-theoretically, included. Many of the attempts to show that there
is something wrong with the CTCT fall along these lines. Similarly, in
the case of life, we have alleged counterexamples to various scientific
definitions such as Artificial Life, etc.
One difference between the two cases is this: There are many non-
equivalent scientific definitions of life. But in the case of algorithm, there
are many equivalent formalizations: Turing machines, recursive functions,
lambda-calculations, etc. What might have happened to the informal notion
if these had not turned out to be equivalent?
For discussion of some of these issues in the case of life, see
[Machery, 2012].
(b) This issue is also related to Darwins and Turings strange inversion.
Compare this quote:
The possibility of the deliberate creation of living organisms from
elementary materials that are not themselves alive has engaged the
human imagination for a very long time. [Lewontin, 2014, 22]
to this paraphrase:
The possibility of the deliberate creation of intelligent behavior
from elementary operations that are not themselves intelligent has
engaged the human imagination for a very long time.
25. On the Cleland issue of the relevance of the environment/external world:
Since we can arbitrarily vary inherited meanings relative to syntactic

machinations, inherited meanings do not make a difference to those
machinations. They are imposed upon an underlying causal structure.
[Rescorla, 2014a, 181]
On this view, the hollandaise-sauce-making computer does its thing whether

its on the earth or the moon (whether its output is hollandaise sauce or not).
(Perhaps its output is some kind of generalized, abstract hollandaise-sauce type,
whose implementations/instantiations/tokens on the moon are some sort of goop,
but whose implementations/instantiations/tokens on earth are what are normally
considered to be (successful) hollandaise sauce.) And the chess/Civil War
computers are doing the same thing.
26. How do you know when a thing just begins to boil? How can you be
sure that the milk has scorched but not burned? Or touch something
too hot to touch, or tell firm peaks from stiff peaks? How do you
define chopped? [Gopnik, 2009b, 106]
27. Hopcroft & Ullman (CITE) distinguish a procedure, which they vaguely define
(their terminology!) as a finite sequence of instructions that can be mechanically
carried out, such as a computer program (to be formally defined in their
Ch. 6), from an algorithm, which they define as A procedure which always
terminates. (See pp. 23 of their text.)
28. [Copeland and Sylvan, 1999, 46] (see also [Copeland, 1997]) distinguish
between two interpretations of the CTCT. The one that they claim was actually
stated by Church and by Turing concerned the functions taht are in principle
computable by an idealised human being unaided by machinery. This one,
they claim, is correct. The other interpretation is that the class of well-defined
computations is exhausted by the computations that can be carried out by Turing
machines.
So, one possible objection to Cleland is that cooking (for example) is not
something that can be carried out by an idealized human being unaided by
machinery, hence the failure of a hollandaise sauce recipe on the moon is

irrelevant to the correct interpretation of the CTCT.
The point about the other interpretation is what leads to hypercomputation
theories (of which the most plausible are interaction machines and trial-and-error
machines).
29. [Church, 1936a, 40, my italics] calls his 1936 statement of what we now
name Churchs thesis a definition of the commonly used term effectively
calculable .
30. On improvisation and interactive computing:
Music is indeed a useful analogy [to computer programs], because

its a flexible form that moves through time. But certainly computer
porgrams are much richer in this respect, becasue programs not only
move through time, they also interact with people and they can even
modify themselves. [Knuth, 2001, 167]
Of course, improvisational music (for example, certain kinds of jazz) also have
tht kind of interactive nature.
31. Modifications of the CTCT, such as the physical CTCT, really arent concerned
with the kind of computation that Hilbert cared about, any more than Euclid
would have cared about geometric constructions that went beyond compass and
straightedge.
32. An interesting argument to the effect that Churchs Thesis should be

distinguished from Turings Thesis can be found in [Rescorla, 2007]: Churchs
Thesis asserts that intuitively computable number-theoretic functions are
recursive. Turings Thesis asserts that intuitively computable string-theoretic
functions are Turing-computable. We can only combine these into a Church-
Turing Computability Thesis by adding a requirement that there be a computable
semantic interpretation function between strings and numbers. (And this raises
the further question whether that semantic interpretation function is intuitively
computable or recursive.)
33. [Shapiro, 2001] argues that computer science is the natural science of
procedures. In support of this claim, we might consider this related claim of
[Rescorla, 2014b, 2, p. 1279]:
To formulate. . . [Euclids GCD algorithm], Knuth uses natural

language augmented with some mathematical symbols. For most
of human history, this was basically the only way to formulate
mechanical instructions. The computer revolution marked the advent
of rigorous computational formalisms, which allow one to state
mechanical instructions in a precise, unambiguous, canonical way.
In other words, computer science developed formal methods for making the
notion and expression of procedures mathematically precise. Thats what makes
is a science of procedures.
34. . . . if we represent the natural number n by a string of n consecutive 1s,

and start the program with the read-write head scanning the leftmost
1 of the string, then the program,
q0 1 1 R q0
q0 0 1 R q1
will scroll the head to the right across the input string, then add a single 1
to the end. It can, therefore, be taken to compute the successor function.
[Aizawa, 2010, 229] But if the environment (the tape) is not a string of n 1s
followed by a 0, then this does not compute the successor function. Compare
this to Clelands Hollandaise sauce recipe being executed on the moon.
35. [Egan, 2010] argue[s] that representational content is to be understood as a

gloss on the computational characterization of a cognitive process (p. 253,
Abstract). She cites David Marrs [Marr, 1982] work on a computational
theory of visual perception. According to Egans interpretation of Marr, a
component of the early visual processing sytem filters the retinal image; the
theoretically important characterization, from a computational point of view,
is a mathematical description: the device computes the Laplacean convolved
with the Gaussian. . . . As it happens, it takes as input light intensity values
at points in the retinal image, and calculates the rate of change of intensity
over the image [Egan, 2010, 255, col. 1, my italics]. But, considered solely
as a computational device, it does not matter that input values represent
light intensities and output values the rate of change of light intensity. The
computational theory characterizes the visual filter as a member of a well
understood class of mathematical devices that have nothing essentially to do with
the transduction of light [Egan, 2010, 255, cols. 12, original italics].
Compare this to the chess-Civil War example: To paraphrase, the theoretically

important characterization from a computational point of view is a mathematical
description: the device computes some mathematical function that, as it happens,
can be interpreted as a chess match or else as a Civil War battle; but, considered
solely as a computational device, it does not matter that input values represent
(say) chess moves or battle positionsthe computational theory characterizes
the device as a member of a well understood class of mathematical devices that
have nohting essentially to do with chess or the Civil War. It is environment
neutral [Egan, 2010, 256, col. 1].
[Egan, 2014, 122] points out, with respect to [Shadmehr and Wise, 2005]s
characterization of the motor-control mechanism [for moving our hand to grasp
an objectsee comments below] allows us to see that a mariner who knew
the distance and bearing from his home port to his present location and the
distance and bearing from his home port to a buried treasure could perform the
same computationvector subtractionto compute the course from his present
location to the treasure. Similarly, a Civil War buff might be able to use the
same computer program that describes a particular battle to help her win a chess
game:
A crucial feature of the function-theoretic characterization [that is,

the characterization that focuses solely on the mathematical function
being computed and not on the purpose or external environment] is
that it is environment neutral: the task is characterized in terms that
prescind from the environment in which the mechanism is normally
deployed. The mechanism described by Marr would compute the
Laplacean of a Gaussian even if it were to appear (per mirabile) in
an environment where light behaves very differently that it does on
earth, or as part of an envatted brain. . . . [Egan, 2014, 122]
This relates to Clelands hollandaise sauce example: Egan says that the visual
filter would compute the same mathematical function in any environment, but
only in some environments would its doing so enable the organism to see
[Egan, 2010, 256, col. 2, my italics]. Similarly, Clelands recipe would compute
the same (culinary?) function in any environment, but only on Earth (and not on
the Moon) would it doing so result in Hollandaise sauce.
When the computational characterization is accompanied by an

appropriate cognitive interpretation, in terms of distal objects and
properties, we can see how the mechanism that computes a certain
mathematical function can, in a particular context, subserve a
cognitive function such as vision. . . . [Egan, 2010, 256, col. 2] (see
also p. 257, cols. 1-2)
That is, we can see what it implements by looking at its semantic interpretation.
(We will discuss this further in 14.4.)
If the program is put in a chess environment, it plays chess; if it is put in

a Civil War environment, it simulates a battle. This issue is also related to
Putnams Twin Earth puzzle (MAYBE PUT THIS IN A FOOTNOTE IF IT IS
NOT DISCUSSED ANYWHERE ELSE): If I use the word water on Earth, it
denotes H2 O; if I (or twin-I) use it on Twin Earth, it denotes XYZ.
Does this mean that algorithms are not intentional, as Hill argues when she says
that all algorithms must have the form: To accomplish task X, do this.
[Egan, 2014] takes up these themes again, in more detail. She gives another
example, due to [Shadmehr and Wise, 2005], concerning how we are able to
move our hand to grasp an object. Note that our brain presumably actually
computes such mathematical functions as those in this example, and thereby

our hand moves correctly; there is a causal connection between the brains
computation and our hands motion. Recall our earlier discussion of whether
the solar system computes Keplers laws of motion (9.7.2). In what ways are
these two casesthe brain computing various functions that cause (or, at least,
result in) our hand to move in a certain way, and the solar system (possibly)
computing various functions that (possibly) result in (if not cause) the planets to
move as they do. Is it possible that our brain causes our hands to move correctly
but doesnt really compute those functions (in the same way that the solar system
doesnt really compute Keplers laws)? It seems reasonable to say, in the brain
case, that the brain really does compute, but not so reasonable to say, in the
solar system case, that the solar system computes. We seem to be willing to be
scientific realists in the one case, but not the other. What, if anything, is the
difference between the two cases?
36. On the conjunction-disjunction issue:
(a) Be sure to cite Shagrir.
(b) [Sprevak, 2010, 3.3, pp. 268269] has a similar example.

(a) Abramson, Darren (2011), Philosophy of Mind Is (in Part) Philosophy of
Computer Science, Minds and Machines 21: 203219.
Argues that the CTCT is relevant to questions about the Turing test for AI.
(b) Blass, Andreas; & Gurevich, Yuri (2003), Algorithms: A Quest for Absolute
Definitions, Bulletin of the European Association for Theoretical Computer
Science (EATCS), No. 81 (October): 195225.
Pre-print accessed 22 April 2014 from:
http://research.microsoft.com/en-us/um/people/gurevich/opera/164.pdf
Reprinted in Olszewksi et al. 2006 (see below), pp. 2457.
The [Church-Turing] thesis was a great step toward understanding
algorithms, but it did not solve the problem what an algorithm is.
(c) Bringsjord, Selmer (1993), The Narrational Case against Churchs Thesis,
accessed 17 April 2014 from:
http://homepages.rpi.edu/brings/SELPAP/CT/ct/
(d) Cleland, Carol E. (2004), The Concept of Computability, Theoretical
Computer Science 317: 209225.
(e) Daly, Ian (2010), Just Like Mombot Used to Make, New York Times, Dining
section (24 February): D1, D5.
http://www.nytimes.com/2010/02/24/dining/24robots.html
On robots that not only follow recipes, but actually cook the meals: Fear not,
humans, these robots are here to be our friends. To prove it, they serve you
food.
Interesting reading in connection with what Cleland 1993 and Preston 2006
have to say about the computability of recipes.
(f) Davis, Martin (2004), The Myth of Hypercomputation, in C. Teuscher (ed.),
212.
Contains a good discussion of the Church-Turing Thesis.
(g) Folina, Janet (1998), Churchs Thesis: Prelude to a Proof, Philosophia
Mathematica 6(3): 302323.
Argues against [Gandy, 1988, Mendelson, 1990, Shapiro, 1993, Sieg, 1994]
(and others) that the thesis is unprovable but true.
(h) Kearns, John (1997), Thinking Machines: Some Fundamental Confusions, Minds
and Machines 7(2): 269287.
A discussion of the Church-Turing Computability Thesis, arguing that it has
not interesting consequences for human cognition because carrying out a
procedure is a purposive, intentional activity. No actual machine does, or can
do, as much.
(i) Kripke, Saul A. (2013), The Church-Turing Thesis as a Special Corollary of

Godels Completeness Theorem, in B. Jack Copeland, Carl J. Posy, & Oron
Shagrir (eds.), Computability: Turing, Godel, Church, and Beyond (Cambridge,
MA: MIT Press): 77104.
An argument by one of the great logicians that the CTCT can be proved. Also
contains a lot of useful historical remarks.
(j) Kroes, Peter (2010), Engineering and the Dual Nature of Technical Artefacts,
properties
(k) Olszewski, Adam; Wolenski, Jan; & Janusz, Robert (eds.) (2006), Churchs Thesis
after 70 Years (Frankfurt: Ontos Verlag).
Includes essays by Selmer Bringsjord, Carol Cleland, B. Jack Copeland, Janet
Folina, Yuri Gurevich, Andrew Hodges, Charles McCarty, Elliott Mendelson,
Oron Shagrir, Stewart Shapiro, and Wilfrid Sieg, among many others.
(l) Petroski, Henry (2010), Occasional Design, American Scientist 98(1) (January-
February): 1619.
An everyday challenge provides lessons in the processes of planning and
execution
A nice follow-up to Preston 2006.
(m) Rey, Georges (2012), The Turing Thesis vs. the Turing Test, The Philosophers
(n) Shapiro, Stewart (2013), The Open Texture of Computability, in B. Jack
Copeland, Carl J. Posy, & Oron Shagrir (eds.), Computability: Turing, Godel,
Church, and Beyond (Cambridge, MA: MIT Press): 153181.
A very readable discussion of the provability of the CTCT; of the nature
of the informality, vagueness, or open texture of the notion of
computability; and of the difference between human, mechanical, and
mathematical computability, with observations on many of the other readings
discussed or mentioned in this chapter.
(o) Shapiro, Stuart C. (2001), Computer Science: The Study of Procedures
http://www.cse.buffalo.edu/shapiro/Papers/whatiscs.pdf
(p) Sieg, Wilfried (1994), Mechanical Procedures and Mathematical Experience, in
George Alexander (ed.), Mathematics and Mind (New York: Oxford University
Press): 71117.
23 are especially good on the history of the Church-Turing Computability
Thesis.
(q) Sieg, Wilfried (2000), Calculations by Man and Machine: Conceptual Analysis,
Technical Report CMU-PHIL-104 (Pittsburgh, PA: Carnegie-Mellon University
Department of Philosophy); http://repository.cmu.edu/philosophy/178.
The (informal) calculability of number-theoretic functions is first analyzed

into calculability by humans satisfying boundedness and locality conditions;
that is analyzed into computability by string machine; and the latter is
analyzed into computability by a Turing machine. Sieg identifies Turings
thesis as the analysis of the first of these by the last.
(r) Soare, Robert I. (2009), Turing Oracle Machines, Online Computing, and
Three Displacements in Computability Theory, Annals of Pure and Applied
Logic 160: 368399.
http://www.people.cs.uchicago.edu/ soare/History/turing.pdf
and published version accessed 2 July 2014 from:
main.pdf? tid=8258a7e2-01ef-11e4-9636-
00000aab0f6b&acdnat=1404309072 f745d1632bb6fdd95f711397fda63ee2
2, 3, and 12 (Origins of Computabilty and Incomputablity, Turing
Breaks the Stalemate, and Renaming It the Computability Thesis )
contain a good summary of the history of the CTCT.
(s) Special Issue on Churchs Thesis (1987), Notre Dame Journal of Formal Logic
28(4) (October).
Accessed 9 April 2014 from:
http://projecteuclid.org/euclid.ndjfl/1093637642
(t) Turner, Raymond (2011), Specification, Minds and Machines 21(2): 135152.
2. Websites:
Does a function in mathematics change anything?
Accessed 8 November 2012 from http://www.askphilosophers.org/question/1877
Humorous Algorithms
http://www.cse.buffalo.edu/rapaport/510/alg-cartoons.html
Chapter 11
The Church-Turing
II. What Is Hypercomputation?
Speculation that there may be physical processesand so, potentially,

machine-operationswhose behaviour cannot be simulated by the
universal Turing machine of 1936 stretches back over a number of
decades. Could a machine, or natural system, deserving the name
hypercomputer really exist? Indeed, is the mindor the brainsome
form of hypercomputer? [Copeland, 2002, 462]1
We now know both that hypercomputation

(or super-recursive computation) is mathematically well-understood, and
that it provides a theory that according to some accounts for some real-
life computation. . . [is] better than the standard theory of computation at
and below the Turing Limit. . . . [S]ince its mathematically possible that
human minds are hypercomputers, such minds are in fact hypercomputers.
[Bringsjord and Arkoudas, 2004, 167]
The editors have kindly invited me to write an introduction to this special

issue devoted to hypercomputation despite their perfect awareness of my
belief that there is no such subject. [Davis, 2006c, 4]
According to a 1992 paper, a computer operating in a Malament-Hogarth

spacetime or in orbit around a rotating black hole could theoretically
perform non-Turing computations.
http://en.wikipedia.org/wiki/Hypercomputation
1 Recall our discussion of the mind-body problem, 2.8.
449
450 CHAPTER 11. WHAT IS HYPERCOMPUTATION?
(accessed 26 May 2014). [At least, its something, as someone once

posted in reply.]

1. Required:
(a) Copeland, B. Jack (2002), Hypercomputation, Minds and Machines 12(4)
(November): 461502.
A good overview and survey.
Read pp. 461465.
Read 1.5, on Putnam & Gold
Read 1.7, on Boolos & Jeffrey
Read 1.13, on Kugel
Read 1.181.18.1, on Penrose, Turing, & Godel
Read 1.24, on Cleland
Read all of 23
(b) Wegner, Peter (1997), Why Interaction Is More Powerful than Algorithms,
Communications of the ACM 40(5) (May): 8091.
Many of Wegners papers are online in various formats at his homepage:
http://www.cs.brown.edu/people/pw/home.html
(accessed 15 May 2014).
(c) Kugel, Peter (2002), Computing Machines Cant Be Intelligent (. . . and Turing
Said So), Minds and Machines 12(4) (November): 563579.
(d) Davis, Martin (2004), The Myth of Hypercomputation, in C. Teuscher (ed.), Alan
Turing: The Life and Legacy of a Great Thinker (Berlin: Springer): 195212.

(a) Eberbach, Eugene; & Wegner, Peter (2003), Beyond Turing Machines, Bulletin
of the European Association for Theoretical Computer Science (EATCS), No. 81
(October): 279304
http://www.eatcs.org/images/bulletin/beatcs81.pdf
Another useful survey, but it opens with some seriously misleading comments
about the history of computation and the nature of logic.2
(b) On Putnam-Gold machines:
John Cases COLT Page
accessed 8 November 2012 from: http://www.eecis.udel.edu/case/colt.html
2 For example, contrary to what the authors say, (1) [Turing, 1936] was not primarily a mathematical
demonstration that computers could not model mathematical problem solving (p. 279) (although it did
include a mathematical demonstration that there were some mathematical problems that a computer could
not solve); and (2) Hilbert did not take the Entscheidungsproblem as a basis that all mathematical problems
could be proved true or false by logic (p. 279), simply because mathematical problems cannot be proved
true or false (rather, mathematical propositions might be provable or not, and they might be true or false,
but truth and falsity are semantic notions incapable of proof, whereas provability is a syntactic notion).
Nowak, Martin A.; Komarova, Natalia L.; & Niyogi, Partha (2002),
Computational and Evolutionary Aspects of Language, Nature 417 (6 June):
611617.
Accessed 4 June 2014 from:
http://people.cs.uchicago.edu/niyogi/papersps/NKNnature.pdf
11.2 Introduction
[A] computer linked with a human mind is a more powerful tool than
an unassisted human mind. One hope of many computer scientists is to
demonstrate. . . that the computer/human team can actually accomplish far
more than a human alone. [Forsythe, 1967a, 3].
One might also ask whether that team could accomplish far more than a computer
alone, say by interacting with the computer while it is computing [Lohr, 2013,
Steed, 2013].
We have seen that it can be argued that computer science is the systematic study of
computing, that computing is the mathematical study of algorithms, and that algorithms
are best understood mathematically in terms of Turing machines (or anything logically
equivalent to Turing machines).
But are they best understood that way? We saw in the previous chapter that it can
be argued that there might be other kinds of algorithms that are not computable by a
Turing machine. In this chapter, we will look at some of the arguments for this point
of view and at some of the kinds of algorithms that are allegedly not computable by a
Turing machine.
11.3 Copelands Theory of Hypercomputation

In a series of papers, the philosopher Jack Copeland has contrasted Turing-
machine computation with what he has called hypercomputation (see especially
[Copeland, 2002]). He describes a Turing-machine computation as a computation of
functions or numbers. . . with paper and pencil [such that each digit of the numerical
representation of the number or each input-output pair of the function is produced]
in a finite number of steps by a human clerk working effectively. And he defines
hypercomputation as the computation of functions or numbers that cannot be
computed by a Turing machine [Copeland, 2002, 461].
So, if hypercomputable functions or numbers cannot be computed by a Turing
machine, how can they be computed? He cites the following possibilities:
The primitive operations of the computation might not be executable by a human

working alone, in the way that Turings 1936 paper described.
MOVE THIS TO A SECTION ON ORACLE MACHINES:
For example, in Turings Ph.D. dissertation at Princeton (which he began after
his classic 1936 paper was published, and which was completed in 1938; Church
was his dissertation advisor), Turing described what he called oracle machines,
or o-machines. These were an extension of Turing machines that allowed
additional primitive operations that are not computable by an ordinary Turing
machine (that is, by what Turing called an a-machine). An oracle machine
consists of a Turing machine plus a black box that outputs the value of
a function that is not computable by a Turing machine. If a function g is
computable by an oracle machine with an oracle that outputs the value of a
(non-Turing-machine-computable) function f , then it is said that g is computable

relative to f .
Some data could be supplied during the computation, rather than being pre-
stored on the Turing machine tape. This is what happens during interactive
computing.
MOVE THIS TO A SECTION ON INTERACTIVE COMPUTING:
Peter Wegner [Wegner, 1997] has argued that interaction machines (or
coupled Turing machines) are strictly more powerful than Turing machines.
Turing machines accept no new input while computing. But interaction machines
are coupled to a finite number of streams of input symbols (and can have a
finite number of output channels). If an interaction machine halts, then it can be
simulated by a universal Turing machine (by the Substitution Property discussed
below). But if it does not halt (as is the case, for example, for an ATM;3 ATMs
only halt when they are broken or being repaired), then, if the unending input
stream is a number computable by a universal Turing machine (e.g., ), then the
interaction machine can also be simulated by a universal Turing machine, but,
if the unending input stream is not computable by a universal Turing machine,
then the interaction machine is a hypercomputer.
But is it? Or is it just an oracle machine? We will see in a moment why it is not
obvious that oracle-machine computation is hyper in any interesting sense.
Why might such a non-halting, non-computable, interaction machine be a
hypercomputer? Its input stream might be random, in which case, it would not be
computable by a Turing machine (this was proved by Church) (but Davis objects
to this; see below WHERE?). Or its input stream might not be random, but just
non-computable.
Here is a potential counterexample to the idea that interactive computation
is not computable by a Turing machine. In the theory of (Turing machine)
computation, there is a theorem called the Substitution Property:
( recursive function s)(x, y, z N)[x (y, z) = s(x,y) (z)]
In this statement, a (b) represents the ath Turing machine (in some numbering
scheme for Turing machines), and b is its input. (How do you number Turing
machines? One way is to take the numbering that Turing used and that we
discussed in 8.12.)
So, the Substitution Property says that there exists a recursive function s (i.e.,
a function s that is computable by a Turing machine) that has the following
property: For any three natural numbers x, y, z, the following is true: The xth
algorithm (the xth Turing machine), when given both y and z as inputs produces
the same output that the s(x, y)th algorithm (that is, the s(x, y)th Turing machine)
does when given z as input.
Here is another way to say this: First, enumerate all of the partial recursive
functions. Second, let x be the xth partial recursive function. Suppose that it
3 That is, an automated teller machine.
11.3. COPELANDS THEORY OF HYPERCOMPUTATION 455
takes two inputs: x and y (another way to say this is that its single input is the
ordered pair (y, z)). Then there exists another partial recursive function s(x,y)
that is, we can find another partial recursive function that depends on s (two)
inputs (and the dependence is itself a recursive function, s)such that s(x,y) is
input-outputequivalent to x when y is fixed, and which is such that s(x,y) is a
function with y (that is, with part of x s input) stored internally as data.
Here is an example of these two kinds of Turing machines (with a program for
x (y, z) on the left and a program for s(x,y) (z) on the right:
begin begin
input(y, z); constant := y;
output := process(y, z); input(z);
print(output) output := process(constant, z);
end. print(output);
end.
In other words, data can be stored effectively (that is, algorithmically) in
programs; the data need not be input from the external world.
In this way, any interactive program could, in principle, be shown to be logically
equivalent to a non-interactive program. That is, any interactive program can be
simulated by an ordinary Turing machine by pre-storing the external input.
The catch here is whether you need to know in advance what the external input
is going to be. But the Substitution Property theorem does seem to say that, once
you know what that input is, you need only an ordinary Turing machine, not an
interactive hypercomputer.
The constraint of data as symbols on paper could be relaxed.

MOVE THIS TO A SECTION ON CLELAND?:
For example, Clelands mundane hollandaise-sauce recipe does not take such
symbols as either input or output; instead, the input are certain food ingredients
and the output is a certain food preparation.
The computers rules could be allowed to change during the computation.

MOVE THIS TO A SECTION THAT ELABORATES ON IT?:
But what about self-modifying programs? Surely, they are computable by a
Turing machine, yet they can change their own programs. More precisely, a
universal Turing machine with a program P stored on its tape could, as part of
executing P, change P, yet continue executing. This is a consequence of the idea
that program and data are interchangeable.
The constraints about a finite number of steps or a finite amount of time can be
relaxed.
MOVE THIS TO A SECTION THAT ELABORATES ON IT?:
This would be the case with what has been called a Zeus machine or
an accelerating Turing machine [Boolos and Jeffrey, 1974]. Turing placed
temporal patterning restrictions on his model because he wanted to model

human computation. If we relax those restrictions, then we could devise a
machine that would calculate each digit of a real numbers decimal expansion in
half of the time of the previous digits calculation. Thus, an infinite calculation,
including the Halting Problem, could be computed in a finite amount of time.
(As Bertrand Russell observed of a very similar example [Russell, 1936, 143],
this is not logically impossible, just medically impossible!)
There are several basic questions that need to be considered:
Are these really alternative models of humanly effective procedures? (And does
effective need to mean humanly effective?)
Are they counterexamples to the Church-Turing Computability Thesis?
Or are they just other models of computation (perhaps of some more general
notion of computation)?
How realistic are they? Can such hypercomputers really exist?
Is the mind or brain a hypercomputer (instead of a Turing-machine computer)?
Martin Davis thinks that most of these hypercomputers are either wrong or just
plain silly [Davis, 2004], [Davis, 2006c]: Consider machines that allow for non-
computable input: Of course, if you allow for non-computable input, you should
expect to be able to get non-computable output. (Compare the garbage in/garbage
out principle: If you allow for incorrect input, you should expect to be able to get
incorrect output.)
But might there be some intuitively effective, yet not Turing-machineeffective,
computations that dont take non-computable input? One potential example of this are
so-called trial-and-error machines, to which we now turn.
11.4 Trial-and-Error Machines

One example of such a machine is what are called trial-and-error machines
[Putnam, 1965], [Gold, 1967], [Hintikka and Mutanen, 1997]. Here, no constraints
need to be relaxed; rather, we change our interpretation of what counts as the output of
the computation.
Recall our discussion of the Halting Problem: We contrasted two algorithms:
begin
if C(i) halts
then output halts
else output loops
end.
This version of the algorithm can be converted to the self-referential H (see 7.8)
and thereby used in order to show that the Halting Problem is not Turing-machine
computable.
11.4. TRIAL-AND-ERROR MACHINES 457
But there is another version of this algorithm, which we saw could not be so
converted:

begin
output loops; {i.e., make an initial guess that C loops}
if C(i) halts
then output halts; {i.e., revise your guess}
end.
H2 is an example of a trial-and-error machine: You make an initial guess about

the desired output, and then keep running the program on a number of trials. If the
trials produce errors or dont come up with a desired response, then continue to run
more trials.
In general, a trial-and-error machine is a Turing machine with input i that outputs
a sequence of yes/no responses such that it is the last output that is the desired
output of the machine (rather than the first, or only, output). But you dont allow any
way to tell effectively if youve actually achieved the desired output, that is, if the
machine has really halted.
As Hilary Putnam says,
[I]if the machine has most recently printed yes, then we know that
the. . . [input has been accepted] unless the machine is going to change its
mind; but we have no procedure for telling whether it will change its mind
or not. The. . . [problems] for which there exist decision procedures in
this widened sense are decidable by empirical meansfor, if we always
posit that the most recently generated answer is correct, we will make a
finite number of mistakes, but we will eventually get the correct answer.
(Note, however, that even if we have gotten to the correct answer (the end
of the finite sequence) we are never sure that we have the correct answer.)
[Putnam, 1965, 49, italics in original]
Such a machine, as weve seen, can decideor computethe Halting

Problem!
11.4.1 Does Intelligence Require Trial-and-Error Machines?

NOTE: THIS SECTION NEEDS TO BE REVISED AFTER A CAREFUL RE-
READING OF KUGEL 2002.
Peter Kugel [Kugel, 2002] has argued that trial-and-error machines can give a
computer initiative, allowing them to achieve real AI. If so, then this would show,
not that AI is not computable, but that it is not a computable function. There are certain
aspects of Kugels argument that make his version of hypercomputation more plausible
than some of the others.
Kugel argues that AI will be possible, using digital computers (and not requiring
fancy, quantum computers or other kinds of non-digital computers), but by using those
digital computers in a non-computational way. He begins his argument by observing
that intelligence in general, and artifical intelligence in particular, requires initiative,

which he roughly identifies with the absences of discipline, defined, in turn, as the
ability to follow orders. Thus, perhaps, intelligence and AI require the ability to break
rules! Computation, on the other hand, requires this kind of discipline (after all, as we
have seen, computation certainly includes the ability to follow orders, or, at least, to
behave in accordance with orders).
Computers compute. But, Kugel points out, they can do other things, too. For
example, they can change their mind. Turing 1947 (CITE) claimed that infallible
entities could not be intelligent, but that fallibility allows for intelligence. And,
according to Kugel, [Turing, 1950] said that computers could probably be programmed
to fake intelligence, but probably could not be programmed to be intelligent. We
will see if this is a reasonable interpretation of [Turing, 1950] when we look at the
philosophy of AI in Chapter 19, but I will note here that most interpretations of
[Turing, 1950] take it that a computer might be able to fool a human interrogator into
confusing a human with a computer because both of them would be intelligent!
The core of Kugels argument is the observation that a distinction can be made
between a Turing machine and Turing machinery. Turing machinery with a finite
memory (either a finite tape or even no tape at all) is not a Turing machine but a finite
automaton, a mathematical structure with less than the power of a Turing machine.
SAY MORE ABOUT FINITE AUTOMATA.
In a Turing machine, the first output is the result of its computation. But there
is nothing preventing the use of Turing machinery (such as a finite automaton) and
taking the last output of its operation as its result. (You cant say that the operation
of such a finite automaton is computation, if you accept the CTCT, which identifies
computation with the operation of a Turing machine.) This, as we have seen, is a
Putnam-Gold trial-and-error machine.
Kugel next argues that Turing-machine computation does not suffice for
intelligence, on the grounds that, if it did, it would not be able to survive! So, suppose
(by way of reductio) that Turing-machine computation did suffice for intelligence. And
suppose that a mind is a universal Turing-machine with instincts (that is, with some
built-in programs) and that is capable of learning (that is, capable of computing new
programs). To learn (that is, to compute a new program), it could either compute a
total computable program (that is, one defined on all inputs) or else compute a partial
computable program (that is, one that is undefined on some inputs).
Next, Kugel defines a total machine to be one that computes only total computable
functions, and a universal machine to be one that computes all total computable
functions. Is a universal Turing machine total or universal in this sense? According
to Kugel, it cant be both: the set of total computable functions is enumerable:
f1 , f2 , . . .. Let Pi be a program that computes fi , and let P be a program (machine?)
that runs each Pi . Next, let P0 compute fn + 1. Then P0 is a total computable function,
but it is not among the Pi , hence it is not computed by P. That is, if P computes only
total functions, then it cant compute all of them.
According to Kugel, a universal Turing machine is a universal machine (so it also
computes partial functions). If the mind is a universal Turing machine, then there are
functions whose values it cant compute for some inputs. And this, says Kugel would
be detrimental to survival. If the mind were total, then there would be functions that it
11.5. SUMMARY 459
coudnt compute at all (namely, partial ones). This would be equally detrimental.
But, says Kugel, there is a third option: Let the mind be a universal Turing machine
with pre-computed or default values for those undefined inputs. Such a machine is
not a Turing machine, because it could solve the Halting Problem! Give it a program C
and input i, and simply let it immediately output loops and only then start determining
whether C halts on i. If C does halt on i, let this machine output halts. That is, its
real output is the last thing that it outputs, not the first thing. It is a trial-and-error
machine.
(It is interesting to compare such machines to heuristic programming. In Chapter 7,
we saw that a heuristic program was one that gave an approximate solution to a
problem. It would seem that trial-and-error machines do just that: Each output is an
approximation to the last output. Ideally, each output is a closer approximation than
each previous output.)
Again, Putnam-Gold machines (NOTE: DECIDE WHETHER TO CALL THEM
PGM OR T&EM) are such that, for each input, they allow an unlimited, but finite,
number of outputs (trials). The last output (not any of the intermediate guesses)
is considered to be the computed value of the input. (If there is no first output, or no
last output, then the result is technically undefined.) Both Putnam-Gold machines and
Turing machines use the same machinery, but they use it in different ways.
Is there an aspect of intelligence that is not Turing-machinecomputable, but that is
Putnam-Goldcomputable? Yes: learning a function from its values. In Golds version,
you are given the initial outputs of a function f : f (1), f (2), . . ., and you are asked if
you can infer, or guess, or compute what function f is. This is called computational
learning theory or inductive inference. It is an abstract way of describing the
problem that a child faces when learning its native language: f (i) is what the child
hears at time t = i, and f is the grammar of the language. Putnam-Gold machines are
precisely defined to model this. ELABORATE ON GOLDS THEORY
11.5 Summary
Martin Davis seems to have a good refutation of some hypercomputation theories,
namely, of course you should expect to get non-computable output if you allow non-
computable input.
Only two theories of hypercomputation seem useful: Interaction machines and
trial-and-error machines. But both of these seem to be only minor extensions
of the Turing analysis of computation. Interaction machines give rise to relative
computation, which still seems to be a version of Turings original analysis. The only
case in which an interaction machine differs significantly from a Turing machine is
when the interactive machine doesnt halt and its input stream allows for an oracle-
supplied, non-computable number. But this seems to fall under Daviss objection to
hypercomputation. Notably (as we will see in Ch. 19), the Turing test is interactive;
insofar as interaction is not modeled by a Turing machine, this would seem to be a
trivial gap.
Trial-and-error machines, on the other hand, seem to have potential, but,
importantlydespite being technically not the computation of a resultthey abide by
all of Turings constraints on what counts as computing, with the possible exception of
knowing whether it has halted. So, perhaps intelligence is not a computable function,
not because it is not computable, but because it is not a function!4
4 Some of the observations in this summary are based on a personal conversation with Jonathan Buss,
6 March 2007.

1. Possibly organize this chapter in terms of procedures that relax different
conditions on the notion of algorithm.
For example, [Hopcroft and Ullman, 1969, 2] characterize a procedure as a

finite sequence of instructions that can be mechanically carried out, such as a
computer program and algorithm as a procedure that halts (p. 91). They also
characterize Churchs hypothesis as the claim that any process which could
naturally be called a procedure can be realized by a Turing machine (p. 80).
Since procedures in their sense need not halt, neither need Turing machines.
2. Need to say something about Hilberts original constraints (finiteness,

etc.), which would seem to require computation to be humanly possible
computation.
3. Say something about quantum computing.
4. Need to elaborate on the notion of interactive computing and its relation to oracle
machines.
5. Wegner is a proponent of interactive computing. There is also something

called reactive systems, which may be the same thing, but neither body of
literature seems to cite the other. Their relationships may be worth exploring.
See:
http://en.wikipedia.org/wiki/Reactive system
accessed 8 November 2012. Check with Selman on this.
One of the pioneers of reactive systems, Amir Pnueli, characterizes them as

systems whose role is to maintain an ongoing interaction with their environment
rather than produce some final value upon termination. Typical examples
of reactive systems are air traffic control system[s], programs controlling
mechanical devices such as a train, a plane, or ongoing processes such as a
nuclear reactor (from the syllabus for his Fall 2002 NYU course, Analysis
of Reactive Systems). In the lecture notes for that course, he wrote: There are
two classes of programs: Computational Programs: Run in order to produce
a final result on termination. Can be modeled as a black box. Specified
in terms of input/output relations. Example: The program which computes
y = 1 + 3 + . . . + (2x 1) can be specified by the requirement y = x2 . Reactive
Programs: Programs whose role is to maintain an ongoing interaction with their
environments. . . . Such programs must be specified and verified in terms of their
behaviors. (from Lecture 1,
http://www.cs.nyu.edu/courses/fall02/G22.3033-004/index.htm) also at:
http://cs.nyu.edu/courses/fall02/G22.3033-004/ accessed 18 June 2014
6. Some games that involve tossing dice might be like oracle or interactive Turing
machines. Still, by the Subst Thm, any such game/Turing machine, after it has
been played, can be simulated by a deterministic Turing machine. I.e., for each
such oracle Turing machine, there is some deterministic Turing machine whose
i-o and algorithmic behavior is identical, except it cant be determined in advance
which one it is.
7. There are things. . . bees can do that humans cannot and vice versa
[Sloman, 2002, 3.3]. Does that mean that bees can do non-computable
tasks? Iff do means something different from compute, such as physical
performance. Cf. Cleland
8. See [Hayes, 1990], as discussed in 11, suggesting that interactive and classical
Turing computation differ only in the source of input information.
9. On Turing on fallibility, related to 11.4.1:
[O]ne can show that however the machine [that is, a computer] is
constructed there are bound to be cases where the machine fails to
give an answer [to a mathematical question], but a mathematician
would be able to. On the other hand, the machine has certain
advantages over the mathematician. Whatever it does can be
relied upon, assuming no mechanical breakdown, whereas the
mathematician makes a certain proportion of mistakes. I believe that
this danger of the mathematician making mistakes is an unavoidable
corollary of his [sic] power of sometimes hitting upon an entirely new
method. [Turing, 1951, 256]
This gives support Kugels claims about fallibility. Such trade-offs are common;
for example, as Godel showed, certain formal arithmetic systems can be
consistent (infallible?) or else complete (truthful?), but not both. An analogy
is this: In the early days of cable TV (the late 1970s), there were typically two
sources of information about what shows were onTV Guide magazine and
the local newspaper. The former was consistent/infallible in the sense that
everything that it said was on TV was, indeed, on TV; but it was incomplete,
because it did not list any cable-TV shows. The local newspaper, on the other
hand, was complete in the sense that it included all broadcast as well as all
cable-TV shows, but it was inconsistent/fallible, because it also erroneously
included shows that were not on TV or cable (but there was no way of knowing
which was which except by being disappointed when an advertised show was
not actually on).
In a later passage, Turing suggests one feature that. . . should be incorporated in

the machines, and that is a random element (p. 259). This turns the computer
into a kind of interactive o-machine that would result in the behaviour of the
machine not being by any means completely determined by the experiences to
which it was subjected (p. 259), suggesting that Turing realized that it would
make it a kind of hypercomputer, but, presumably, one that would be only (small)
extension of a Turing machine.
(It is also of interest to note that, in this talk, Turing envisaged what has come to
be known as The Singularity: it seems probably tht once the machine thinking
method had started, it would not take long to outstrip our feeble powers. There
would be no question of the machines dying, and they would be able to converse
with each other to sharpen their wits. At some stage therefore we should have
to expect the machines to take control, in the way that is mentioned in Samuel
Butlers Erewhon (pp. 259260).)
10. Concurrent computation can be thought of in two ways. In the first way,
a single computer (say, a (universal) Turing machine) executes two different
algorithms, not one after the other (sequentially), but in an interleaved fashion,
in much the way that a parallel computation can be simulated on a serial
computer: On a serial computer (again, think of a Turing machine), only one
operation can be executed at a given time; on a parallel computer, two (or
more) operations can be executed simultaneously. But one can simulate parallel
computation on a serial computer by interleaving the the two operations. The
difference between parallel and concurrent computation is that the latter can also
be thought of in the second way mentioned above: Two (or more) computers
(say, non-universal Turing machines) can each be executing two different, single
algorithms simultaneously in such a way that outputs from one can be used as
inputs to the other.5
Is concurrent computation a kind of hypercomputation? It is, if interactive

computation is the same as that second way of thinking about concurrent
computation, and if interactive computation is hypercomputation. At least
one important theoretician of concurrent computing has said that concurrent
computation. . . is in a sense the whole of our subjectcontaining sequential
computing as a well-behaved special area [Milner, 1993, 78] This is consistent
with those views of hypercomputation that take it to be a generalization of
(classical) computation. Where (classical) computation is based on an analysis
of mathematical functions, Milner argues that concurrent computation must be
based on a different mathematical model; that also suggests that it is, in some
sense, more than mere classical computation. Interestingly, however, he also
says that this different mathematical model must match the functional calculus
not by copying its constructions, but by emulating two of its attributes: It is
syntheticwe build systems in it, because the structure of terms represents the
structure of processes; and it is computationalits basic semantic notion is a
step of computation [Milner, 1993, 82; italics in original, my boldface]. Again,
it appears to be a generalization of classical computation.
Unlike certain models of hypercomputation that either try to do something

that is physically (or medically) impossible or can only exist in black holes,
5 See http://en.wikipedia.org/wiki/Concurrent computing
and http://en.wikipedia.org/wiki/Concurrency (computer science),

both accessed 26 May 2014.
concurrent computation seems more down to earthmore of a model of

humanly possible processing, merely analyzing what happens when Turing
machines interact.
In fact, heres an interesting way to think about it: A single Turing machine
can be thought of as a model of a human computer solving a single task. But
humans can work on more than one task at a time (concurrently, if not in
parallel). And two or more humans can work on a single problem at the same
time. Moreover, there is no reason why one humans insights (outputs) from
work on one problem couldnt be used (as inputs) in that same humans work
on a distinct problem. And there is surely no reason why one humans insights
from work on a given problem couldnt be used by another humans work on
that same problem. Surely even Hilbert would have accepted a proof of some
mathematical proposition that was done jointly by two mathematicians. So,
insofar as concurrent computation is a kind of hypersomputation, it seems to
be a benign kind.
Is it the same as Wegners interactive computation? Perhaps. One possible

difference concerns the source of intermediate input. If such input comes from
an oracle that can solve non-classically-computable problems, then it is surely
different.
11. [Wegner, 1995, 45] identifies interaction machines with Turing machines with
oracles and with modeling reactive processes (citing work by Pneuli).
Moreover, Churchs thesis. . . becomes invalid if the intuitive notion of
computability is extended to include interactive computation. Church and Turing
identify effective computability with purely automatic computation, while
interaction machines model patterns of distributed interactive computations
coordinated by protocols [Wegner, 1995, 46]. Here, I think, we can
begin to see where a disagreement with Wegner over the invalidity of
the CTCT even for someone who might agree with him about the way in
which interactive computing might be hyper: I wouldnt want to say that
CTCT identifies effective computability with purely automatic computation,
because I would distinguish the latter notion from the intutive notion, taking
automatic computation as a synonym for -computation or Turing-machine
computation and then understanding the CTCT as identifying the intuitive notion
with the more formal, mathematical notion of automatic/-/Turing-machine
computation.
But heres an interesting question: If you dont allow oracles (or physically
impossible computations, or black-hole computations, etc.), can interactive
computation make the Halting Problem computable? Put another way, the
Halting Problem is not classically computable; is it interactively computable?
12. Automatic computation is like batch computing:

The classic models of computation are analogous to minds

without bodies. For Turings machine, a calculation begins with a
problem on tis tape, and ends with an answer there. . . . How the initial
tape. . . is input, and how the final one is output, are questions neither
asked nor answered. These theories conform to the practice of batch
computing.
Eventually, interactive models of computation emerged,

analogous to minds in bodies. . . . A single input at initiation and a
single output at termination are now superseded by multiple inputs
and outputs distributed in time and space. These theories conform to
the practice of interactive computing.
Interaction is the mind-body problem of computing.

[Wadler, 1997, 240241]
Wadler also notes that such interaction need not be concurrent. The rest of
Wadler 1997 is concerned to show how such interaction can be added to a Turing-
complete programming language. (It is not clear whether to what extent this
makes the language more powerful.)
More fully:
Today, computing scientists face their own version of the mind-
body problem: how can virtual software interact with the real world?
In the beginning, we merely wanted computers to extend our minds:
to calculate trajectories, to sum finances, and to recall addresses. But
as time passed, we also wanted computers to extend our bodies: to
guide missiles, to link telephones, and to proffer menus.
The classic models of computation are analogous to minds
without bodies. For Turings machine, a calculation begins with a
problem on its tape, and ends with an answer there. . . .
Eventually, interactive models of computation emerged,
analogous to minds in bodies. . . . A single input at initiation and
a single output at termination are now superseded by multiple inputs
and outputs distributed in time and space.
13. As [Hintikka and Mutanen, 1997, 181] note, the Halting Problem algorithm in its
trial-and-error form is not recursive, even though it is obviously mechanically
determined in a perfectly natural sense. They also note that this perfectly
natural sense is surely part of the informal notion of computation that the
CTCT asserts is identical to recursiveness, hence that the CTCT is not valid
(p. 180). (Actually, theyre a bit more cautious, claiming that the informal
notion is ambiguous: We doubt that our pretheoretical ideas of mechanical
calculability are so sharp as to allow for only one explication (p. 180).
14. Heres another term for interactive/reactive (etc.) computing:
Turing machines accept no input while operating. A finite amount

of data may be inscribed on the tape before the computation starts,
but thereafter the machine runs in isolation from its environment. A
coupled Turing machine is the result of coupling a Turing machine
to its environment via one or more input channels. Each channel
supplies a stream of symbols to the tape as the machine operates.
[Copeland and Sylvan, 1999, 51]
As with Kugel, such a system clearly uses Turing machinery. The authors
also note that any coupled Turing machine that halts can be simulated by the
universal Turing machine, since the streams of symbols supplied by the coupled
machines input channels are, in this case, finite in length, and so can be inscribed
on the universal machines tape before it commences its task, essentially by
the methods I suggested in 11.3 in our discussion of the Substitution Property.
However, the universal Turing machine is not always able to simulate a coupled
Turing machine thatlike an idealised automatic teller machine or air traffic
controllernever halts (p. 51). And they give a simple proof (p. 52) that there
is a coupled Turing machine that cannot be simulated by the universal Turing
machine. Essentially, the proof involves an oracle that supplies an non-Turing
computable real number, so this falls prey to Daviss objection. It raises the
following question: Can a coupled Turing machine that does not make use of a
non-computable oracle and that does not halt be simulated by a classical Turing
machine?
15. Of course there are kinds of hypocomputation: Consider the Chomsky hierarchy,
FSA, PDA, etc. So, if there is any serious hypercomputation, that puts
classical, Turing-machine computation in the middle. It still holds a central
place, given the equivalence of Turing machines to recursive functions, etc.,
and given its modeling of human computing and its relation to Hilberts
problem. Hypercomputation seems to come in two flavors: what Ill call
weird hypercomputation and what Ill call plausible hypercomputation (to
use neutral terms!). In the former category, Ill put medically impossible
Zeus machines, machines that can only exist near black holes, etc. In the latter
category, Ill put tae-machines and interactive machines. In a grey zone in the
middle, Ill put o-machines. o-machines are clearly a plausible extension of
Turing machines, as even Turing knew. What about interactive machines that
dont use non-computable input?
Related to all this is the role of non-halting programs like operating systems,
ATMs, etc. Some of my colleagues say that those programs dont express
algorithms, because algorithms, by definition, must halt. But even Turings
original Turing machines didnt halt: They computed infinite decimals, after
all! The central idea behind the Halting Problem is to find an algorithm that
distinguishes between programs that halt and those that dont. Whether halting
is a Good Thing is independent of that.
16. [Copeland and Sylvan, 2000, 8.1, esp. pp. 190191] offer an interesting
categorization of hypercomputation. They first carve off analogue computing

as a challenge to classical computation. This is not normally considered to be
part of hypercomputation, so I will ignore it here (but see Chs. 3 and 9 for some
references). Then there are the challenges from newer physics; these include
Zeus machines, relativistic machines, etc. (which strike me as only of theoretical
interest). But then they point out, more interestingly, that computability is
relative not simply to physics, but more generally to systems or frameworks.
They begin with degree theory or Turing reduction, based on Turings
o-machines. I NEED TO SAY MORE ABOUT THIS. As weve seen, this
either falls prey to Daviss objection (non-computability in/non-computability
out) or else simply becomes Wegners interaction machines (Copelands coupled
machines), which are built out of Turing machinery and which may, or may
not, exceed Turing machines in computing power (the only argument Ive seen
to that effect runs afoul of Daviss objection).
They then identify two kinds of relativity: resource relativity and logical
relativity. The former includes relativity in procedures, methods or in
the devices available for computing, which includes the analogue and
physical categories discussed above, as well as o-machines. Different
procedures might include different basic operations or instructions (in
much the same way that different geometric systems might go beyond
straightedge and compass to include measuring devices (or might go below
to replace a collapsible compass with a fixed compasson this, see
http://en.wikipedia.org/wiki/Compass equivalence theorem). But its not clear
if such procedural relativity necessarily goes beyond (if not below) Turing
computability.
Logical relativity concerns the use of non-classical logics, such as relevance

logics. [Copeland and Sylvan, 2000] suggest that these might lead to
hypercomputation. Perhaps; but it is certainly the case that classical computers
can compute using relevance logics [Shapiro and Rapaport, 1987].
17. [Davis, 2006b, 1218] notes that if a set A of natural numbers is Turing reducible
to a set B of natural numbers (written: A t B), and if B is itself a computable
set, the nothing new happens; in such a case A t B just means that A is
computable. But if B is non-computable, then interesting things happen.
According to [Davis, 2006c], of course, one of the uninteresting things is that
A will then turn out to be non-computable. (The interesting things have to do
with degrees of non-computability: can one non-computable set be more non-
computable than another?.
18. In [Knuth, 2014, Question 13], Alfred Aho asks about reactive distributed
systems that maintain an ongoing interaction with their environmentsystems
like the Internet, cloud computing, or even the human brain. Is there a universal
model of computation for these kinds of systems?
Note that phrasing the question this way suggests that computation is a very
general term that includes not only Turing-machine computation but other kinds
of computing as well (perhaps even hypercomputation).
Knuths answer identifies reactive processes with computational methods,

defined in [Knuth, 1973, 4] as a non-terminating, single-processor algorithm (see
7.5, above).
This brief discussion suggests that the proper way to consider such reactive
or distributed or interactive systems is not as some new, or hyper, model
of computation, but simply as the study of what might happen when two or
more Turing machines interact. Theres no reason to expect that they would
not be more powerful in some sense than a single Turing machine. Clearly, if
a Turing machine interacts with something that is not a Turing machine, then
hypercomputation can be the result; this is what Davis notes. Even if it is two
Turing machines that are interacting, and even if that produces the ability to
compute non-Turing-machine-comptable functions, that would not indicate
that there is any limitation to the CTCT (and this may be what Copeland had
in mind when he distinguishes two interpretations of the CTCT).
19. On the Substitution Property:
An interactive system is a system that interacts with the environment
via its input/output behavior. The environment of the system is
generally identified with the components which do not belong to it.
Therefore, an interactive system can be referred to as an open system
because it depends on external information to fulfil its task. If the
external resources are integrated into the system, the system no longer
interacts with the environment and we get a new, closed system. So,
the difference between open and closed systems lies in the eye of the
beholder. [Prasse and Rittgen, 1998, 359, my italics]
The components which do not belong to it would presumably include external
input. The italicized sentence is a nice statement of my interpretation of the
Substitution Property.
[Prasse and Rittgen, 1998], by the way, admirably call attention to the vagueness
and informality of the claims in [Wegner, 1997] (see especially 4.4 and 4.6).
20. On interaction machines:
[Prasse and Rittgen, 1998, 359] consider a program such as the following:
begin
let b be a Boolean variable;
let x be an integer variable;
input(b);
while b = true do
input(x);
output(x2 );
input(b);
end
They say of a program such as this:
Neglecting input/output, each iteration can be interpreted as a
computation performed by a Turing machine. However, the influence
of the (unknown) user input on the control flow makes it hard to
determine what overall function is computed by this procedure (or
if the procedure can be seen as a computation at all). . . .
The input will be determined only at run-time. The overall
function is derived by integrating the user into the computation, thus
closing the system.
It is evident that Turing machines cannot model this behavior
directly due to the missing input/output operations. Therefore, we
need models that take into account inputs and outputs at run-time to
describe computer systems adequately.
Wegner would agree, though Prasse & Rittgens point is that Wegners
interpretation that this violates the CTCT is incorrect. Still, this is a nice example
of the kind of hypercomputation that Wegner has in mind.
Now, we could easily write a Turing-machine program that would be a version

of this while-loop. Consider such a Turing machine with a tape that is initialized
with all of the input (a sequence of bs and xs, encoded appropriately). This
Turing machine clearly is (or executes) an algorithm in the classical sense. Now
consider a Turing machine with the same program (the same machine table), but
with an initially blank tape and a user who inscribes appropriate bs and xs on the
tape just before each step of the proram is executed (so the Turing machine is
not waiting for user input, but the input is inscribed on the tape just in time).
Is there a mathematical difference between these two Turing machines? Is there
anything in [Turing, 1936] that rules this out? CHECK THIS
Interaction machines are defined as Turing machines with input

and output. Therefore, their internal behavior and expressiveness
do not differ from that of equivalent Turing machines. Though
Wegner leaves open the question of how the input/output
mechanism works, it can be assumed that input and output
involve only data transport, without any computational capabilities.
Therefore, the interaction machine itself does not possess greater
computational power than a Turing machine. However, through
communication, the computational capabilities of other machines can
be utilized. Interaction can then be interpreted as a (subroutine) call.
[Prasse and Rittgen, 1998, 361]
Turings o-machines are of this type; the call to a (possibly non-computable)

oracle is simply a call to a (possibly non-computable) subroutine. So, as they
say, the machine itself is just a Turing machine, and, as Davis would say, if a
non-computable input is encoded on its tape, then a non-computable output can
be encoded there, too.
They also note that Wegners concept of computability focuses more on real
computers and systems than on the theoretical, function-oriented view of Church,
Kleene, Markov, Post and Turing (p. 362). This is certainly consistent with
some of Smiths claims, to be discussed in 21.7.1.
21. Consistent with the organizing principle for the chapter of considering lifting
restrictions on Turing machines, if we ignore temporal requirements, then
we can get accelerating (or Zeus) machines: Neither Turing nor Post, in
their descriptions of the devices we now call Turing machines, made much
mention of time. . . . They listed the primitive operations that their devices
perform. . . but they made no mention of the duration of each primitive operation
[Copeland, 1998, 150]6
We might choose to ignore or reject accelerating Turing machines on the grounds

that they are medically (physically) impossible. After all, no (physical) device
can really accelerate in that way. But then, by parity of reasoning, we would have
to reject ordinary Turing machines, on the grounds that they, too, are physically
impossible, because, after all no (physical) device can really have an infinite
tape or even an arbitrarily extendible tape. So all physical Turing machines are
limited to a tape that is no larger than the universe (large, but still finite) and
hence simulable by a hypocomputer (such as a PDA). So, if an abstract Turing
machine is mathematically possible, then, surely, so is an accelerating Turing
machine.
22. A tae-machine (such as the tae version of the Halting Problem program), if
implemented on an accelerating Turing machine, will always let us know when
it is finished guessing:
To produce a machine that can compute the values of the

halting function. . . one simply equips an accelerated universal Turing
machine wiht a signalling devicea hooter, saysuch that the hooter
blows when and only when the machine halts. . . . [I]f the hooter blows
within two moments of the start of the simulation then H(p, d) = 1,
and if it remains quiet then H(p, d) = 0. [Copeland, 1998, 153]
(where H(p, d) is the halting function such that H(p, d) = 1 if Turing machine p
halts on input d and H(p, d) = 0 otherwise).
6 [Copeland, 1998, 151, fn 2] distinguishes between accelerating Turing machines and Zeus
machines.
[Copeland, 1998, 154] observes that the hooter-enhanced machine is not a Turing
machine.
23. [Shagrir, 1999, 2, pp. 132133] observes that The Church-Turing thesis is
the claim that certain classes of computing machines, e.g., Turing-machines,
compute effectively computable functions. It is not the claim that all computing
machines compute solely effectively computable functions. This is consistent
with Kugels views on tae-machines; it is also consistent with the view that what
makes interaction machines more powerful than Turing machines is merely how
they are coupled with either other machines or with the environment, and not
with the machinery itself. Shagrirs goal is to argue for a distinction between
algorithms and computation, with the latter being a wider notion.
24. On interaction:
We interact with the world through programs all the time. Very often,
programs we use interact with a network of unseen programs so that
their ends and ours may be realized. We are acted on by programs all
the time. These entities are not just agents in the sense of being able
to take actions; they are also agents in the representative sense, taking
autonomous actions on our behalf. [Chopra, 2014]
Similarly, the nature of ubiquitous computing seems both to clearly call
for such interaction as well as to consider it part of computing (not
hypercomputing); see, for example, [Dublon and Paradiso, 2014].
25. Some have argued that Turings model of computation is based on what humans
can do. Yet it is an idealized human who is modeled, for example, one that
has no limits on space (recall that the tape is infinite). [Cleland, 2004, 212]
points out that, in that case, one could allow other idealizations, such as no
limits on speed of computation. That would make a Zeus machine at least as
plausible as a Turing machine. (This fits in with the idea that hypercomputation,
in general, investigates the lifting of restrictions on the classical informal notion
of algorithm.) However, [Cockshott and Michaelson, 2007, 2.5, p. 235] reject
these relaxations out of hand.
26. For a refutation of Wegner, see:
http://blog.computationalcomplexity.org/2006/07/principles-of-problem-
solving-tcs.html
27. My point about a Turing machine being
able to simulate an interactive/reactive/coupled computer (via the Substitution
Property) is also mentioned in [Teuscher and Sipper, 2002].
28. [Fortnow, 2010] nicely refutes three of the major arguments in favor of
hypercomputation (including analog computation). Of most interest to us is this
passage, inspired by Turings comment that The real question at issue is What
are the possible processes which can be carried out in computing a number?
[Turing, 1936, 9, p. 249] (see 8.7.2.1, above):
Computation is about process, about the transitions made from one

state of the machine to another. Computation is not about the input
and the output, point A and point B, but the journey. Turing uses the
computable numbers as a way to analyze the power and limitations of
computation but they do not reflect computation itself. You can feed a
Turing machine an infinite digits of a real number. . . , have computers
interact with each other. . . , or have a computer that perform an
infinite series of tasks. . . but in all these cases the process remains
the same, each step following Turings model. . . . So yes Virginia,
the Earth is round, man has walked on the moon, Elvis is dead and
everything computable is computable by a Turing machine. (pp. 3, 5,
my italics.)
29. Good idea for intro section: [Cockshott and Michaelson, 2007, 2.5, p. 235]
wonder whether hypercomputation is a Kuhnian revolutionary challenge to the
normal science of computer science.
30. For more on oracles and relative computability, see

[Dershowitz and Gurevich, 2008, 5, pp. 329f].
31. Almost all the results in theoretical computability use relative

reducibility and o-machines rather than a-machines and most
computing processes in the real world are potentially online or
interactive. Therefore, we argue that Turing o-machines, relative
computability, and online computing are the most important concepts
in the subject, more so that Turing a-machines and standard
computable functions since they are special cases of the former and
are presented first oly for pedagogical clarity to beginning students.
[Soare, 2009, Abstract, p. 368]7
This is an interesting passage, because it could be interpreted by
hypercomputation advocates as supporting their position and by anti-
hypercomputationalists as supporting theirs!
In fact, a later comment in the same paper suggests the pro-hypercomputational

reading:
The Turing o-machine is a better model to study them [that is, to study
online computing devices which can access or interact with some
external database or other device] because the Turing a-machine
lacks this online capacity. [Soare, 2009, 9, p. 387]
He even cites, seemingly approvingly, passages from the introduction to a book

on the topic edited by Wegner and Goldin [Goldin et al., 2006].
7A slightly different version of the same paper appears as [Soare, 2013].

He goes on to say, The theory of relative computability developed by Turing [in

his doctoral disseration] and [Emil] Post [for example, in [Post, 1943] and the o-
machines provide a precise mathematical framework for database [or interactive]
or online computing just as Turing a-machines provide one for offline computing
processes such as batch processing [Soare, 2009, 1.3, pp. 370371]. This
certainly suggests that some of the things that Copeland and Wegner say
about hypercomputation are a bit hyperbolic; it suggests that both the kind of
hypercomputation that takes non-computable input (supplied by an oracle) to
produce non-computable output as well as the kind that is interactive are both
well-studied and simple extensions of classical computation theory. Oracles
can model both client-server interaction as well as communication with the Web
[Soare, 2009, 1.3, p. 371].
He also discusses Putnams tae-machines, which he calls limit computable

functions [Soare, 2009, 9.2, p. 388].
However, the interesting point is that all of these are extensions of Turing
machines, not entirely new notions. Moreover, Soare does not disparage, object
to, or try to refute the CTCT; rather, he celebrates it [Soare, 2009, 12].
His basic point on this topic seems to be this:

Conclusion 14.3 The subject is primarily about incomputable objects
not computable ones, and has been since the 1930s. The single
most important concept is that of relative computability to relate
incomputable objects. [Soare, 2009, 14, p. 395]
This is certainly in the spirit of hypercomputation without denigrating the CTCT.
32. On Gold and tae machines, cite [Hauser et al., 2002, 1577].
33. [Milner, 1993, 80] offers an example of how a theory of concurrency
and interaction requires a new conceptual frameworkthough that is not
necessarily to say that it is a form of hypercomputation or that it refutes the
CTCT:
[T]here are programs P1 and P2 which have the same relational
meaning,8 but which behave differently when each runs in parallel
with a third program Q. . . :
Program P1 : x := 1 ; x := x + 1
Program P2 : x := 2
In the absence of interference,9 P1 and P2 both transformt he initial
memory by replacing the value of x by 2, so they have the same
meaning. But if you take the program
8 In the case of the example below, this simply means that the two programs have the same input-output
behavior.
9 This can be thought of as interaction with the environment in the form of intermediate input.
Program Q: x := 3
and run it in parallel with P1 and P2 in turn:
Program R1 : P1 par Q Program R2 : P2 par Q
then the programs R1 and R2 have different meaning. (Even if an
assignment statement is executed indivisibly, R1 can end up with x
equal to 2, 3, or 4, while R2 can only end up with x equal to 2 or
3.) So a compositional semantics must be more refined; it has to take
account of the way tghat a program interacts with the memory.
To see why R1 can have any of those three possible values, consider that Q
could be thought of as being executed before x := 1, immediately after it,
or after x := x + 1. To make the relevance to Wegners claims clearer, note
that Q could have been input(x), instead of an assignment statement. But
note Milners conclusion: It is not that such interactive programming refutes the
CTCT, nor that this is a form of computation that goes beyond classical Turing-
machine computation; rather, it is merely the claim that a more refined form
of semantics is needed. (Wegner & Goldin 2003 chastise Milner for avoid[ing]
the question.)
34. Nice quote on interactive computing:
I can design a program that never crashes if I dont give the user any
options. And if I allow the user to choose from only a small number
of options, limited to htings that appear on a menu, I can be sure that
nothing anomalous will happen, becasue each option can be foreseen
in advance and its effects can be checked. But if I give the user the
ability to write programs that will combine with my own program, all
hell might break loose. [Knuth, 2001, 189190]
35. A quote sympathetic to Wegner:
Turing reducibility [developed by Post from Turings oracle

machine]. . . is the most important concept in computability theory.
Today, the notion of a locla machine interacting with a remote
database or remote machine is central to practical computing
[Soare, 2012, 3290]
36. One of the constraints on Turings analysis of computation, as cited by

[Gurevich, 2012, 4], is that of being Isolated Computation is self-contained.
No oracle is consulted, and nobody interferes with the computation either during
a computation step or in between steps. The whole computation of the algorithm
is determined by the initial state. This certainly suggests that interactive
computation is not Turing-machine computation. On the other hand, it could
also be interpreted to mean merely that computation must be mechanical or
automatic, and surely10 this could include the mechanical or automatic
10 A word that any philosopher should surely(?) take with a grain of salt!
use of input from an external source (including an oracle).
Cf. also [Gurevich, 1999, 93, 98], which talks about algorithms that are closed
in the sense that they do not interact with their environments and ones that
do so interact (5, pp. 111115). Gurevich also notes that, in the case of
non-deterministic algorithms, the active environment will make the choices
(p. 116).
[Ekdahl, 1999, 3, pp. 262263] has a nice example that illustrates how
interactive computing is modeled by o-machines and relative computability. The
essence of the example considers a simplified version of an airline-reservation
program. Such a program is a standard example of the kind of interactive
program that Wegner claims is not Turing-machine computable, yet it is not
obviously an o-machine, because it does not obviously ask an oracle for the
solution to a non-computable problem. Suppose our simplified reservation
program is this:
begin
while true do:
input(passenger, destination);
output(ticket(passenger, destination))
end while
end.
Ekdahl observes that, although writing the passenger and destination information
on the input tape is computable and can equally well be done by another Turing
machine, when our reservation program then asks for two new inputs,
which [inputs are] going to [be written] on the tape is not a recursive process.
. . . So, the input of [passenger and destination] can be regarded as a question to
an oracle. An oracle answers questions known in adavance but the answers are
not possible to reckon in advance (italics in original; my boldface).
37. On trial and error machines:
someone who wishes to know the correct answer would have no way
to know at any given time whether the latest output is the correct
output (Quoted by [Nayebi, 2014, 286] from:
Davis, Martin (2006), The Church-Turing Thesis: Consensus and
Opposition, Lecture Notes in Computer Science 3988: 125132.)
That is, a trial and error machine does compute the uncomputable, but we
cant reliably use the result. Really? What about language learning and heuristic
satisficing?
38. On trial and error machines:

William Jamess discussion of two different approaches to knowledge seems
relevant:
But the faith that truth exists, and that our minds can find it, may
be held in two ways. We may talk of the empiricist way and of the
absolutist way of believing in truth. The absolutists in this matter say
that we not only can attain to knowing truth, but we can know when we
have attained to knowing it; whilst the empiricists think that although
we may attain it, we cannot infallibly know when. To know is one
thing, and to know for certain that we know is another. [James, 1897,
The Will to Believe, V, p. 465]
Compare: The faith that a problem has a computable (or algorithmic) solution
exists, and that our computers can find it, may be held in two ways. We may
talk of the trial-and-error way and of the Turing-algorithmic way of solving
a problem. The Turing algorithmists in this matter say that we (or Turing
machines) not only can solve computable problems, but we can know when
we (or they) have solved them; while the trial-and-error hypercomputationalists
think that although we (or our computers) may solve them, we cannot infallibly
know when. For a computer to produce a solution is one thing, and for us to
know for certain that it has done so is another.
39. [Feferman, 1992, 315] claims that notions of relative (rather than absolute)
computability (that is, notions based on Turings o-machines rather than on
his a-machines) have primary significance for practice and that these relative
notions are to be understood as generalization[s]. . . of computability [and of
the Church-Turing Thesis] to arbitrary structures. So this seems to fly in the
face of Wegners claims that interaction is something new while agreeing with
the substance of his claims that interaction is more central to modern computing
than Turing machines are.
40. [Feferman, 1992, 321] notes that o-machines (where o is an oracle that solves a
non-computable problem) can be generalized to that of a B-machine for any set
B:
In the register machine model, one simply adds to the basic
instructions, ones of the form:
(1) r j := 1 if rk is in B, else r j := 0.
Essentially, this adds primitive operations to a Turing machine (or a register

machine). If these operations can be simulated by the standard primitive
operations of the Turing machine, then we havent increased its power (only
its expressivity, essentially by the use of named subroutines). But if B contains
problems not solvable by the Turing machine, then, of course, we have increased
its power.
On named subroutines: [Feferman, 1992, 339340] says that the built-in

functions of actual computers are given by a black boxwhich is just
another name for an oracleand a program to compute a function f from one
or more of these built-in functions is really an algorithm for computation of f
relative to those built-in functions.
More importantly, Feferman goes on to observe that the arguments for the
Church-Turing Thesis lead one strongly to accept a relativized version: (C-T)r [a
set] A is effectively computable from B if (and only if) A is Turing computable
from (or Turing reducible to) B. To say that A is Turing computable from (or
Turing reducible to) B is to say that x is in A iff the B-machine outputs 1 when its
input is x (where output 1 means yes, x is in A). This is called Feferman then
says that Turing reducibility gives the most general concept of relative effective
computability.
And here is Feferman on the crucial matter:
Uniform global recursion provides a much more realistic picture of

computing over finite data structures than the absolute computability
picture, for finite data bases are constantly being updated.
As examples, we may consider. . . airline reservation systems.
[Feferman, 1992, 342]
He does, however, go on to say that while notions of relativized (as compared

to absolute) computability theory are essentially involved in actual hardware and
software design, the bulk of methods and results of recursion theory have so
far proved to be irrlevant to practice [Feferman, 1992, 343]. That certainly is
congenial to Wegners complaints.
41. Perhaps the issue is not so much whether it is possible to compute the
uncomputable (by extending or weakening the notion of Turing-machine
computation), but whether it is practical to do so. [Davis, 2006a, 126] finds
this to be ironic:
. . . computer scientists have had to struggle wih the all-too-evident

fact that from a practical point of view, Turing computability does not
suffice. . . . With these [that is, with NP-complete] problems Turing
computability doesnt help because, in each case, the number of steps
required by the best algorithms available grows exponentially with
the length of the input, making their use in practive problematical.
How strange that despite this clear evidence that computability alone
does not suffice for practical purposes, a movement has develoope
under the banner of hypercomputation proposing the practicality of
computing the non-computable.
42. Trial and error computation is equivalent to computations by o-machines that

solve the halting problem:
If the computation is to determine whether or not a natural number n

as input belongs to some set S, then it turns out that sets for which
such trial and error computation is available are exactly those. . .
that are computable relative to. . . a oracle that provides correct

answers to queries concerning whether a given Turing machine. . . will
eventually halt. [Davis, 2006a, 128]
43. O-machines show us that not all that is studied in computation theory
is Turing-equivalent. [Aizawa, 2010, 230]
Note the subtle difference between saying this and saying something like: All
computation is equivalent to Turing-machine computation (which is a version of
the CTCT).
44. [Weizenbaum, 1976, Ch. 5, p. 135] interestingly distinguishes between
computers and robots, where the latter (but not the former) have
perceptors. . . and effectors. This could be interpreted as a way of distinguishing
between Turing-equivalent computers and interactive computers, the latter being
Weizenbaums robots. On the other hand, even Turing machines have to have
perceptors and effectors in the sense of having a read-write head.
45. On oracle computation:
One technique for simulating solutions to the N-body problem is to use an
ordinary computer linked to a graphics processing unit that is far more powerful
than the ordinary computer. Is such a combination like a Turing machine with
an oracle? See [Hayes, 2015, esp. pp. 9293]

MAYBE REORGANIZE THIS TO INCLUDE SEPARATE SECTIONS OF PRO AND
CON HYPERCOMPUTATION PAPERS.
Aaronson, Scott (2008), The Limits of Quantum Computers, Scientific American
(March): 6269.
Contains some comments casting doubt on certain models of
hypercomputation.
Aaronson, Scott (2013), Quantum Computing Since Democritus (New York:
Cambridge University Press).
Contains discussions of both quantum computing and hypercomputation; see
Aaronson 2006, below.
Aaronson, Scott (2014), Quantum Randomness, American Scientist 102(4) (July-
August): 266271.
(a) Accessed 8 September from:
http://www.americanscientist.org/issues/pub/quantum-randomness
(b) If theres no predeterminism in quantum mechanics, can it output numbers
that truly have no pattern?
Arkoudas, Konstantine (2008), Computation, Hypercomputation, and Physical
Science, Journal of Applied Logic 6(4) (December): 461475.
Preprint accessed 30 June 2014 from:
http://people.csail.mit.edu/kostas/papers/jal.pdf
Argues against Copeland that a version of the CTCT that is immune to
hypercomputation objections follows from the systematic predictability of
observable behavior (and that even o-machines are, in fact, deterministic
digital computers) (see 3 of his paper).
Brassard, Gilles (1995), Time for Another Paradigm Shift, ACM Computing
Surveys 27(1) (March): 1921.
A commentary on [Hartmanis, 1995a], arguing that probabilistic and
especiallyquantum computers may be qualitatively more powerful than
classical machines; see [Hartmanis, 1995b] for a reply.
Bringsjord, Selmer; & Arkoudas, Konstantine (2004), The Modal Argument for
Hypercomputing Minds, Theoretical Computer Science 317: 167190.
http://research.cs.queensu.ca/home/akl/cisc879/papers/PAPERS FROM THEORETICAL COMPUTER SCIENCE/05051
[W]e give herein a novel, formal modal argument showing that since its
mathematically possible that human minds are hypercomputers, such minds
are in fact hypercomputers. (from the abstract)
Bringsjord, Selmer; & Zenzen, Michael (2002), Toward a Formal Philosophy of
Hypercomputation, Minds and Machines 12(2) (May): 259280.
Button, Tim (2009), SAD Computers and Two Versions of the Church-Turing
Thesis, British Journal for the Philosophy of Science 60: 765792.
Chaitin, Gregory (2006), How Real Are Real Numbers?, International

Journal of Bifurcation and Chaos
2006 version accessed 6 November 2012 from:
http://www.cs.auckland.ac.nz/chaitin/olympia.pdf
2009 version accessed 6 November 2012 from:
http://www.umcs.maine.edu/chaitin/wlu.html
We discuss mathematical and physical arguments against continuity and in
favor of discreteness (from the abstract). In addition to this mathematical
soul-searching regarding [the reality of] real nuymbers, some physicists are
beginning to suspect that the physical universe is. . . perhaps even a giant
computer [. . . Wolfram, 2002] (from 1).
Chaitin, Gregory (2006), The Limits of Reason, Scientific American (March):
7481.
http://www.umcs.maine.edu/chaitin/sciamer3.pdf
Cockshott, Paul; & Michaelson, Greg (2007), Are There New Models of
Computation? Reply to Wegner and Eberbach, The Computer Journal 50(2): 232
247.
http://www.dcs.gla.ac.uk/wpc/reports/wegner25aug.pdf
Argues that Wegners interaction machines do not exceed Turing machines in
computational power; also argues against a number of other hypercomputation
proposals, and contains an excellent summary of Turing machines and
complexity theory.
Cooper, S. Barry (2012), Incomputability after Alan Turing, Notices of the AMS
59(6): 776784.
Accessed 21 July from:
http://www.ams.org/notices/201206/rtx120600776p.pdf
An interesting, illustrated, historical survey.
Copeland, B. Jack (2002), Accelerating Turing Machines, Minds and Machines
12(2) (May): 303326.
Copeland, B. Jack (guest ed.) (2002), Special Issue on Hypercomputation, Minds
and Machines 12(4) (November).
Includes Bringsjord & Zenzen 2002; Copeland 2002, Hypercomputation;
Kugel 2002; Steinhart 2002.
Complete table of contents accessed 5 February 2014 from:
http://link.springer.com/journal/11023/12/4/
Copeland, B. Jack (guest ed.) (2003), Special Issue on Hypercomputation
(continued), Minds and Machines 13(1) (February).
Complete table of contents accessed 5 February 2014 from:
http://link.springer.com/journal/11023/13/1/
Copeland, B. Jack; & Proudfoot, Diane (1999), Alan Turings Forgotten Ideas in
Computer Science, Scientific American (April): 98103.

http://www.cs.virginia.edu/robins/Alan Turings Forgotten Ideas.pdf
For a rebuttal, see:
Hodges, Andrew, The Professors and the Brainstorms
accessed 8 November 2012 from
http://www.turing.org.uk/philosophy/sciam.html
Copeland, B. Jack; & Shagrir, Oron (2011), Do Accelerating Turing Machines
Compute the Uncomputable?, Minds and Machines 21(2): 221239.
Copeland, B. Jack; & Shagrir, Oron (2013), Turing versus Godel on Computability
and the Mind, in B. Jack Copeland, Carl J. Posy, & Oron Shagrir (eds.),
Computability: Turing, Godel, Church, and Beyond (Cambridge, MA: MIT Press):
133, esp. 1.2.3 (pp. 1215).
Cotogno, P. (2003), Hypercomputation and the Physical Church-Turing Thesis,
British Journal for the Philosophy of Science 54(2): 181224.
Argues that hypercomputation does not refute the CTCT.
For a reply, see:
Welch, P.D. (2004), On the Possibility, or Otherwise, of Hypercomputation,
British Journal for the Philosophy of Science 55: 739746.
Davies, E.B. (2001), Building Infinite Machines, British Journal for the
Philosophy of Science 52: 671682.
(a) Preprint accessed 4 June 2014 from:
http://www.mth.kcl.ac.uk/staff/eb davies/jphilsci.pdf
(b) Argues that an accelerating computer could be built in a continuous
Newtonian universe (as opposed to the Hogarth-Malament spacetime
mentioned in the motto to this chapter), though not in the real universe.
Davis, Martin (2006), Why There Is No Such Discipline as
Hypercomputation, Applied Mathematics and Computation 178: 47
http://tinyurl.com/Davis200611
Dresner, Eli (2008), Turing-, Human- and Physical Computability: An Unasked
Question, Minds and Machines 18(3) (Fall): 349355.
Frenkel, Karen (1993), An Interview with Robin Milner, Communications of the
ACM 36(1) (January): 9097.
http://delivery.acm.org/10.1145/160000/151241/a1991-frenkel.pdf
A companion piece to [Milner, 1993], with interesting observations on AI, the
semantics of programming languages, program verification, and the nature of
computer science.
Gold, E. Mark (1965), Limiting Recursion, Journal of Symbolic Logic 30(1):
2848.
11 http://research.cs.queensu.ca/home/akl/cisc879/papers/PAPERS FROM APPLIED MATHEMATICS AND COMPUTATION/
Special Issue on Hypercomputation/davis%5b1%5d.pdf

Presents the mathematics behind tae machines: A class of problems is called

decidable if there is an algorithm which will give the answer to any problem of
the class after a finite length of time. The purpose of this paper is to discuss the
classes of problems that can be solved by infinitely long decision procedures
in the following sense: An algorithm is given which, for any problem of the
class, generates an infinitely long sequence of guesses. The problem will be
said to be solved in the limit if, after some finite point in the sequence, all the
guesses are correct and the same. . . (from the abstract, my italics).
Gold, E. Mark (1967), Language Identification in the Limit, Information and
Control 10: 447474.
http://web.mit.edu/6.863/www/spring2009/readings/gold67limit.pdf
For an excellent guide to understanding Golds paper, see:
Johnson, Kent (2004), Golds Theorem and Cognitive Science, Philosophy
of Science 71 (October): 571592.
http://www.afhalifax.ca/magazine/wp-content/sciences/for-
emgold/Johnson.GoldsTheorem.pdf
Goldin, Dina Q.; Smolka, Scott A.; Attie, Paul C.; & Sonderegger, Elaine L.
(2004), Turing Machines, Transition Systems, and Interaction, Information and
Computation 194(2) (November), Special Issue Commemorating the 5oth Birthday
Anniversary of Paris C. Kanellakis, pp. 101128.
http://www.cse.uconn.edu/dqg/papers/its.pdf
We present Persistent Turing Machines (PTMs), a new way of interpreting
Turing-machine computation, one that is both interactive and persistent. A
PTM is a nondeterministic 3-tape Turing machine that upon receiving an input
token, computes for a while and then outputs the result, and this process is
repeated forever. It is persistent in the sense that the work-tape contents are
maintained from one computation. . . to the next. (from the abstract)
Goldin, Dina; & Wegner, Peter (2004), The Origins of the Turing Thesis Myth,
https://web.archive.org/web/20120212032143/http://www.engr.uconn.edu/dqg/papers/myth.pdf
A sequel to [Wegner, 1997] and Wegner & Goldin 2003.
Goldin, Dina; & Wegner, Peter (2008), The Interactive Nature of Computing:
Refuting the Strong Church-Turing Thesis, Minds and Machines 18(1) (Spring):
1738.
Hintikka, Jaakko; & Mutanen, Arto (1997), An Alternative Concept of
Computability, in Jaakko Hintikka (ed.), Language, Truth, and Logic in
Mathematics (Dordrecht, The Netherlands: Springer): 174188.
Hodges, Andrew (2012), Beyond Turings Machines, Science 336 (13 April):
163164.
http://211.144.68.84:9998/91keshi/Public/File/41/336-6078/pdf/163.full.pdf
Discusses whether all types of computationincluding that of our own

mindscan be modeled as computer programs, in the context of quantum
computation.
Johnson, Kent (2004), Golds Theorem and Cognitive Science, Philosophy of
Science 71 (October): 571592.
Kugel, Peter (1986), Thinking May Be More than Computing, Cognition 22(2)
(March): 137198.
Kugel, Peter (1986), When Is a Computer Not a Computer?, Cognition 23(1)
(June): 8994.
Kugel, Peter (2004), Toward a Theory of Intelligence, Theoretical Computer
Science 317: 1330.
Kugel, Peter (2005), Its Time to Think outside the Computational Box,
Communications of the ACM 48(11) (November): 3337.
For a response, see:
Wegner, Peter; & Goldin, Dina (2006), Forum, Communications of teh ACM
49(3): 11.
Lloyd, Seth (2000), Ultimate Physical Limits to Computation, Nature 406
(31 August): 10471054.
(a) Pre-prints accessed 4 June 2014 from:
http://arxiv.org/pdf/quant-ph/9908043.pdf?origin=publicationDetail
and
http://cds.cern.ch/record/396654/files/9908043.pdf
(b) Investigates the fundamental physical limits of computation as determined by
the speed of light. . . , the quantum scale. . . and the gravitational constant.
Milner, Robin (1993), Elements of Interaction, Communications of the ACM
36(1) (January): 7889.
Parker, Matthew W. (2009), Computing the Uncomputable; or, The Discrete
Charm of Second-Order Simulacra, Synthese 169: 447463.
http://philsci-archive.pitt.edu/3905/1/Discrete Charm i.pdf
Argues that non-computable behavior in a model. . . [can be] revealed by
computer simulation (which is, of course, computable).
Piccinini, Gualtiero (2003), Alan Turing and the Mathematical Objection, Minds
and Machines 13: 2348,
Primarily about Turings views on AI (see Ch. 19), but also discusses his
theory of computation and the role of oracle machines.
Piccinini, Gualtiero (2011), The Physical Church-Turing Thesis: Modest or
Bold?, British Journal for the Philosophy of Science 62: 733769.
http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=CEE1AFD492550C99D72CE9A7D90D4E37?doi=10.1.1.366.94
4 discusses hypercomputational challenges to the modest CTCT, which
states that any function that is physically computable is Turing computable
(p. 734).)
Schachter, Vincent (1999), How Does Concurrency Extend the Paradigm of

Computation?, The Monist 82(1) (January): 3757.
Sloman, Aaron (1996), Beyond Turing Equivalence, in P.J.R. Millican & A. Clark
(eds.), Machines and Thought: The Legacy of Alan Turing, Vol. I, Ch. 10, pp. 179
219.
http://www.cs.bham.ac.uk/research/projects/cogaff/Sloman.turing90.pdf
Clarifies how hypercomputation can show how some aspects of human
cognition might not be Turing-machine computable. Nevertheless, the
question remains whether cognition in general is Turing-machine computable
(or can be approximated by a Turing machine).
Steinhart, Eric (2002), Logically Possible Machines, Minds and Machines 12(2)
(May): 281301.
van Leeuwen, Jan; & Wiedermann, Jir (2000), The Turing Machine Paradigm
in Contemporary Computing, in Bjorn Engquist & Wilfried Schmid (eds.),
Mathematics Unlimited2001 and Beyond (Berlin: Springer-Verlag): 11391155.
(a) Preprint accessed 4 June 2014 from:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.32.3925&rep=rep1&type=pdf
(b) Takes up the challenge of formalizing Wegners ideas about interaction
machines; argues that interactive Turing machines with advice. . . are more
powerful than ordinary Turing machines.
Wegner, Peter (1999), Towards Empirical Computer Science, The Monist 82(1)
(January): 58108.
Wegner, Peter; & Goldin, Dina (1999), Interaction, Computability, and Churchs
Thesis, preprint accessed 4 June 2014 from:
http://cs.brown.edu/pw/papers/bcj1.pdf
An (unpublished) attempt at formalizing some of Wegners claims about
interaction machines.
Wegner, Peter; & Goldin, Dina (2003), Computation beyond Turing Machines,
Communications of the ACM 46(4) (April): 100102.
http://www.pld.ttu.ee/vadim/AIRT/8 computation beyond turing machines.pdf
Wegner, Peter; & Goldin, Dina (2006), Principles of Problem Solving,
For a response, arguing that Wegner and Goldin are wrong, see:
Fortnow, Lance (2006), Principles of Problem Solving: A TCS Response,
iComputational Complexity/i (blog), accessed 27 June 2014 from:
http://blog.computationalcomplexity.org/2006/07/principles-of-problem-
solving-tcs.html
Wells, Benjamin (2004), Hypercomputation by Definition, Theoretical Computer
Science 317: 191207.
http://research.cs.queensu.ca/home/akl/cisc879/papers/PAPERS FROM THEORETICAL COMPUTER SCIENCE/0505
2 contains a discussion of [Cleland, 1995, Cleland, 2001].
3 contains a discussion of Tarskis views on the relation between decidability

and recursiveness, and on the relation between the CTCT and the P=NP
problem.
On Quantum Computers:
Aaronson, Scott (2008), The Limits of Quantum Computers, Scientific
American (March): 6269.
Quantum computers would be exceptionally fast at a few specific
tasks, but it appears that for most problems they would outclass todays
computers only modestly. This realization may lead to a new fundamental
physical principle.
Bacon, Dave; & van Dam, Wim (2010), Recent Progress in Quantum
Algorithms, Communications of the ACM 53(2) (February): 8493.
Deutsch, David (1985), Quantum Theory, the Church-Turing Principle and
the Universal Computer, Proceedings of the Royal Society of London A 400:
97117
Preprint accessed 8 April 2014 from:
http://www.cs.berkeley.edu/christos/classics/Deutsch quantum theory.pdf
Gover, Lov K. (1999), Quantum Computing, The Sciences (July/August):
2430.
http://cryptome.org/qc-grover.htm
Hayes, Brian (2014), Programming Your Quantum Computer, American
Scientist 102(1) (January-February): 2225.
Accessed 19 January 2014 from:
https://www.americanscientist.org/issues/pub/programming-your-
quantum-computer/
The quantum system serves as an oracle, answering questions that can
be posed in a format suitable for qubit computations (p. 24).
Monroe, Christopher R.; & Wineland, David J. (2008), Quantum Computing
with Ions, Scientific American (August): 6471.
http://www.cs.virginia.edu/robins/Quantum Computing with Ions.pdf
2. Websites:
Aaronson, Scott (2006), PHYS771 Lecture 4: Minds and Machines,

http://www.scottaaronson.com/democritus/lec4.html (accessed 28 June 2014).
Part of a course on Quantum Computing Since Democritus. The first part
discusses oracles and Turing reducibility in a very clear (but elementary) way,
concluding that hypercomputation is not a serious objection to the CTCT; later
parts discuss AI.
Some of the material also appears in Aaronson 2013.
Aaronson, Scott (2012), The Toaster-Enhanced Turing Machine, iShtetl-
Optimized/i,
http://www.scottaaronson.com/blog/?p=1121
Argues against hypercomputation via a parallel argument that, because Turing

machines cant toast bread, a toaster-enhanced Turing machine that allows
bread as a possible input and includes toasting it as a primitive operation
would be more powerful than a classic Turing machine.
American Mathematical Society (2003), AMS Sectional Meeting Full Program
http://www.ams.org/meetings/sectional/2096 progfull.html
Contains the titles and authors of many papers on hypercomputation that
were presented in four separate sessions on Beyond Classical Boundaries
of Computability.
Published versions of many of the papers (including papers by Bringsjord &
Arkoudas, Cleland, Copeland, Kugel, and Shagrir) are online at:
http://research.cs.queensu.ca/home/akl/cisc879/papers/PAPERS FROM THEORETICAL COMPUTER SC
Burgin, Mark, & Wegner, Peter (organizers) (2003), Special Sessions on Beyond
Classical Boundaries of Computability (Parts I, II, III, & IV), 2003 Spring Western
Section Meeting, American Mathematical Society.
http://www.ams.org/meetings/sectional/2096 progfull.html
E-mail discussion between Carl Hewitt (a major figure in concurrent computing)
and Peter Wegner (the major advocate of interactive computing as a counterexample
to the CTCT), accessed 26 May 2014 from:
http://www.cse.buffalo.edu/rapaport/510/actors-vs-church-thesis.txt
Interactive Computation Is More Powerful Than Non Interactive, accessed 26 May
2014 from:
http://c2.com/cgi/wiki?InteractiveComputationIsMorePowerfulThanNonInteractive
An debate between a hypercomputation skeptic and a believer.
On Hypocomputation: If hypercomputation is computing that is above the
Turing speed limit, so to speak, then hypocomputation is computing below
it:
Lindell, Steven (2004), Revisiting Finite-Visit Computations, accessed
5 February 2014 from: http://tinyurl.com/Lindell200412
Lindell, Steven (2006), A Physical Analysis of Mechanical Computability,
accessed 5 February 2014 from: http://tinyurl.com/Lindell200613
12 http://www.haverford.edu/cmsc/slindell/Presentations/Revisiting%20finite-visit%20computations.pdf
13 http://www.haverford.edu/cmsc/slindell/Presentations/A%20physical%20analysis%20of%20mechanical%20computability.pdf
Chapter 12
What Is a Computer Program?

I. Software vs. Hardware
Version of 2 October 2015
program: n. 1. A magic spell cast over a computer allowing it to turn

ones input into error messages.. . .
The Jargon Lexicon, http://www.jargon.net/jargonfile/p/program.html
Figure 12.1: 2010,

c Berkeley Breathed
487
488 CHAPTER 12. SOFTWARE VS. HARDWARE

1. Required:
Moor, James H. (1978), Three Myths of Computer Science, British Journal for
the Philosophy of Science 29(3) (September): 213222.
For the purposes of Ch. 12, concentrate on 12.
Colburn, Timothy R. (1999), Software, Abstraction, and Ontology, The Monist
82 (1): 319.
Reprinted (in slightly different form) as Ch. 12 of Colburn, Timothy R. (2000),
Philosophy and Computer Science (Armonk, NY: M.E. Sharpe), pp. 198209.
Suber, Peter (1988), What Is Software?, Journal of Speculative Philosophy 2(2):
89119.
Revised version accessed 8 November 2012 from:
http://www.earlham.edu/peters/writing/software.htm
12.2 Introduction
[NOTE FOR NEXT DRAFT: This section could be re-written as a general introduction
to the Part on What Is a Computer Program?, whose first chapter would be
Software vs. Hardware.]
We have just been exploring what an algorithm is. We are now going to ask about
the relation between algorithms and computer programs: In the course of the next few
chapters, we will ask these questions:
What is a computer program?
Do computer programs implement algorithms?
What is the nature of implementation?
What are software and hardware, and how are they related?
Can computer programs be copyrighted, or should they be patented instead?
Can (some) computer programs be considered to be scientific theories?
Typically, one tends to consider a computer program as an expression, in some

language, of an algorithm. The language is typically a programming language such
as Java, or Lisp, or Fortran. The algorithm is something more abstract, whereas
the program is something that is more concrete. Perhaps the relationship between
an algorithm and a program is something like the relationship between a number and
a numeral: Just as a single algorithm, such as the algorithm for binary search, can
be expressed in many different programming languages, so the number two can be
expressed with many different numerals, such as 2 or II, and many different words,
such as two, deux, zwei, and so on.
In fact, we cant really talk about numbers or algorithms without using some kind of
language, so maybe there really arent any of these abstract things called algorithms.
Maybe the only things are programs, some of which might be written in programming
languages that can be directly used to cause a computer to execute the program (or
execute the algorithm?), and some of which might be written in a natural language,
such as English.
In the early days of computers programs were hard-wired into the computer
(perhaps certain physical switches were set in certain ways). In a sense, these programs
were physical parts of the computers hardware. The program could be changed by re-
wiring the computer (perhaps by re-setting the switches). Was such wiring (or switch-
setting) a computer program? Yet computer programs are typically considered to be
software, not hardware.
And what about a program written on a piece of paper? Does it differ from the very
same program written on a computer file? The former just sits there doing nothing. So
does the latter, but the latter can be used as input to other programs on the computer
that will use the information in the program to set the switches so that the computer
can execute the program. But is the medium on which the program is written the only
difference between these two programs?
12.3 What Is Software and Its Relation to Hardware?

Our first main issue concerns the dual nature of software: It can be considered to be
both text and machine (or mechanism).
To clarify this dual nature, consider this problem:
. . . Bruce Schneier authored a book entitled Applied Cryptography, which

discusses many commonly used ciphers and included source code for
a number of algorithms. The State Department decided that the book
was freely exportable because it had been openly published but refused
permission for export of a floppy disk containing the same source code
printed in the book. The books appendices on disk are apparently
munitions legally indistinguishable from a cluster bomb or laser-guided
missile. . . . [The] disk cannot legally leave the country, even though the
original book has long since passed overseas and all the code in it is
available on the Internet. [Wallich, 1997, 42]1
How can a program written on paper be considered a different thing from the very same
program written on a floppy disk? What if the paper that the program was written on
was Hayess magic paper that was discussed in 9.4? But isnt that exactly what a
floppy disk is, at least, when it is being read by a computer?
12.3.1 Moors Theory of the Nature of Software
Figure 12.2: 2011,

c Universal Uclick
Lets begin our discussion with an examination of James H. Moors theory of the
nature of software [Moor, 1978]. According to Moor, both computers and computer
programs can be understood on different levels: They can be understood as physical
objects, subject to the laws of physics, electronics, and so on; a computer disk
containing a program would be a clear example of this level. But they can also be
understood on a symbolic level: A computer can be considered as a calculating device,
and a computer program can be considered as a set of instructions; the text of the
computer program that is engraved on the disk would be a clear example of this level.
1 This case is discussed at length in [Colburn, 1999].
12.3. WHAT IS SOFTWARE AND ITS RELATION TO HARDWARE? 491
So far, we merely have a restatement of the dual nature. But the larger point
is that, for very many phenomena, a single entity can be viewed from multiple
perspectives or stances: The philosopher Daniel C. Dennett has suggested that a
chess-playing computer or its computer program can be understood in three different
ways [Dennett, 1971]:
physically:
From the physical stance, its behavior can be predicted or explained on the basis
of its physical construction together with physical laws. Thus, we might say that
it made (or failed to make) a certain move because its logic gates numbered 5, 7,
and 8 were open, or because transistor #41 was broken.
semantically:
From the design stance, its behavior can be predicted or explained on the
basis of information or assumptions about how it was designed or how it is
expected to behave, assuming that it was designed to behave that way and isnt
malfunctioning. Thus, we might say that it made (or failed to make) a certain
move because line #73 of its program has an if-then-else statement with an
infinite loop.
intentionally:
From the intentional stance, its behavior can be predicted or explained on the
basis of the language of folk psychology: ordinary peoples informal (and not
necessarily scientific) theories of why people behave the way they do, expressed
in the language of beliefs, desires, and intentions. For instance, I might explain
your behavior by saying that (a) you desired a piece of chocolate, (b) you
believed that someone would give you chocolate if you asked them for it, so
(c) you formed the intention of asking me for some chocolate. Similarly, we
might say that the chess-playing computer made a certain move because (a) it
desired to put my king in check, (b) it believed that moving its knight to a certain
square would put my king in check, and so (c) it formed the intention of moving
its knight to that position.
Each of these stances has different advantages for dealing with the chess-playing
computer: If the computer is physically broken, then the physical stance can help us
repair it. If the computer is playing poorly, then the design stance can help us debug
its program. If I am playing chess against the computer, then the intentional stance can
help me figure out a way to beat it.
Moor offers a definition of computer program that is neutral with respect to the
software-hardware duality:
a computer program is a set of instructions which a computer can follow

(or at least there is an acknowledged effective procedure for putting
them into a form which the computer can follow) to perform an activity
[Moor, 1978, 214].
Lets make this a bit more precise:

Definition 1:
Let C be a computer.
Let P be a set of instructions.
Then P is a computer program for C =de f
1. there is an effective procedure for putting P in a form. . .

2. . . . that C can follow. . .
3. . . . in order to perform an activity.
We could be even more precise:
Definition 2:
Let P be a set of instructions.
Let A be an activity.
Then P is a computer program for C to do A =de f
1. there is an effective procedure for putting P in a form. . .

2. . . . that C can follow. . .
3. . . . in order to perform A.
Definition 2 makes the activity A a bit more perspicuous. We looked at the role of an
algorithms purpose in Ch. 7. But, for now, it might be easier to focus on Definition 1.
On that definition, being a computer program is not simply a property of some
set of instructions. Rather, it is a binary relation between a set of instructions and a
computer.2 A set of instructions that is a computer program for one computer might
not be a computer program for a different computer, perhaps because the second one
lacks an effective procedure for knowing how to follow it.
For instance, the Microsoft Word program that is written for an iMac computer
running MacOSX differs from the Microsoft Word program that is written for a PC
running Windows, because the underlying computers use different operating systems
and different machine languages. This would be so even if the two programs look
and feel (that is, what the user sees on the screen and how the user interacts with the
program) were identical.
There are several questions we can ask about this definition:
Does a computer follow instructions? Or does it merely behave in accordance

with them? As we have seen, Turing machineswhich are models of
hardwired, single-purpose computersmerely behave in accordance with their
machine table. On the other hand, universal Turing machineswhich are
models of programmable, general-purpose computerscan be said to follow
instructions. Compare the human use of natural language: When we speak or
write, do we follow the rules of grammar, in the sense of consulting them (even
if unconsciously) before generating our speech or writing? Or does it make more
2 On Definition 2, it is a ternary relation among a set of instructions, a computer, and an activity. But here
I just want to focus on the role of the computer, which is why were just going to consider Definition 1.
sense to say that the rules of grammar merely describe our linguistic behavior?
Note, however, that a computer programmed to understand and generate natural
language might, in fact, speak and write by explicitly following rules of grammar
that are encoded in its suite of natural-language-processing programs. We have
also seen a similar question when we considered whether the solar system
follows Keplers laws of planetary motion or whether the planets movements
are merely best described by Keplers laws (9.7.2). It might be useful to have a
neutral term for what it is that a computer does when it is doing its processing,
whether it is a Turing machine or a universal Turing machine; perhaps execute
could be used for this purpose. So, we could say that a Turing machine executes
the instructions in its machine table, but doesnt follow them, whereas a universal
Turing machine executes the fetch-execute cycle and thereby follows the program
encoded on its tape.
Is the set of instructions physical, that is, hardwired? Or are the instructions
written in some language? (Could they be drawn, instead, perhaps as a
flowchart? Could they be spoken?) Here, Moors answer seems to be that
it doesnt matter, as long as there is a way for the computer to interpret
the instructions and thus carry them out. Importantly, the way that the
computer interprets the instructions must itself be a computable function (it
must be effective). Otherwise, it might require some kind of built-in, non-
computable method of understanding what it is doing.
Begin Digression
HERE, DISCUSS INTEPRETATION VS COMPILATION OF PROGRAMS,
JUST FOR CLARIFICATION THAT THIS IS *NOT* WHAT I MEAN BY
INTERPRETATION.
End Digression
Figure 12.3: 2009,

c Zits Partnership
Moors next distinction is between software and hardware. The informal and
traditional distinction is that a computer program is software and a computer is
hardware. But this raises the problem of whether the hardwiring in a hardwired
computer is hardware (because it involves physical wires) or software (because it is
the computers program). And, of course, it gives rise to the problem mentioned by
Wallich at the beginning of this chapter. So, Moor suggests a better set of definitions:
For a given person and computer system the software will be those
programs which can be run on the computer system and which contain
instructions the person can change, and the hardware will be that part of
the computer system which is not software [Moor, 1978, 215].
In line with this, [Vahid, 2003, 27, my boldface, original italics] notes that, in the
early days of computing, the frequently changing programs, or software, became
distinguished from the unchaning hardware on which they ran.
Again, lets make this a bit more precise:
Definition 3:
Let P be a person, perhaps Cs programmer.3
Let S be some entity (possibly a part of C).
Then S is software for computer C and person P =de f
1. S is a computer program for C,4 and
2. S is changeable by P.
and H is hardware for C and P =de f
1. H is (a physical) part of C, and
2. H is not software for C and P.
Note that being software is a ternary relation among three entities: a computer
program, a programmer, and a computer;5 it is not a simple property such that
something either is, or else it isnt, software. In other words, software is in the eye
of the beholder: One persons software might be anothers hardware!
And a physical part of a computer will be hardware for a programmer and that
computer if either it is not a computer program for that computer or it is not changeable
by that programmer.
That seems to allow for the following possibility: Suppose that there is a computer
program PJava written in Java that runs on my computer. Even if PJava is changeable
by the programmer who wrote it or the lab technician who operates the computerand
therefore software for that personit will be hardware for me if I dont know Java or
dont have access to the program so that I could change it.
So, if a programmer can rewire a computer, then that computers program is
software, even if it is physical.
Question to Think About:

[Tanenbaum, 2006, 8] points out that hardware and software are logically equivalent
3 Please note that here, P refers to a person, not a program, as in the two previous definitions.
4 This would have to be modified to include activity A if we want to use Definition 2 for computer
program.
5 Or it might be a quaternary relation among four things, if we include activity A.
in the sense that anything doable in hardware is doable in software, and vice versa.
How does this relate to Moors definition?
12.3.2 Subers Theory of the Nature of Software

Peter Suber [Suber, 1988] argues that software is pattern per se, or syntactical form
([Suber, 1988, online abstract]). Any syntactical form? Yes! In other words, everything
is software. (You might ask yourself how this relates to Searles claim that everything
is a computer!)
Does he really mean everythingeven random bits or noise? Why does Suber
say this? He claims that software patterns do not carry their own meanings
[Suber, 1988, 6, p. 103]. In other words, they are purely formal syntax, meaningless
marks, symbols with no intrinsic meanings. (This is consistent with the claim of
[Tenenbaum and Augenstein, 1981, 6] that information has no meaning (see also
http://www.cse.buffalo.edu/rapaport/563S05/data.html).) But they need to be
interpreted by a suitable computer: They need only make a fruitful match with another
pattern (embodied in a machine) [which] we create [Suber, 1988, 6, p. 103, my
italics]. So, software is pure syntax and needs another piece of syntax to interpret
it. This is not far from Moors definition: Both require someone (or something) to
interpret the syntax.
[NOTE FOR NEXT TIME: THIS WILL MAKE MUCH MORE SENSE
IF THERES AN EARLIER CHAPTER IN WHICH I EXPLAIN SYNTACTIC
SEMANTICS AND HOW THEY RELATE (VIZ., TWO SYSTEMS OF SYNTAX
CAN BE SEEN AS INTERPRETING EACH OTHER). See [Rapaport, 1988,
Rapaport, 1995, Rapaport, 2012b].]
How does this relate to Moors definition? Could a hardwired program and a written
program both be software, perhaps because they have the same syntactic form? I think
the answer is yes.
Here is a possible refinement: Software is a pattern that is readable and executable
by a machine. And this is roughly Moors definition of computer program. But, for
Suber, all patterns are readable and executable.
The bottom line is that Subers notion of software is closer to Moors notion of
computer program.
12.3.3 Colburns Theory of the Nature of Software

Finally, [Colburn, 1999, Colburn, 2000] argues that software is not a machine made
out of text. Thus, he would probably disagree with Hayess definition of a computer as
magic paper. Colburn says this because he believes that there is a difference between
softwares medium of description and its medium of execution. The former is the
text in a formal language (an abstraction). The latter consists of circuits and semi-
conductors (it is concrete). Consequently, Colburn says that software is a concrete
abstraction.
He offers several analogies to positions that philosophers have taken on the mind-
body problem (see ]refinanyfield); we might call this the abstract-program/concrete-
program problem. Consider monism, the view that there are not both minds and
brains, but only minds (which is the view called idealism, associated primarily with
the philosopher George Berkeley) or else only brains (the view called materialism).
Similarly, one might hold that either software is abstract or else it is concrete, but it
cannot be both. No matter how strong the arguments for, say, materialism might be as
the best answer so far to the mind-body problem, monism as a solution to the software-
hardware problem fails to account for its dual nature (according to Colburn).
So lets consider dualism. In the mind-body problem, there are several versions
of dualism (the view that there are both minds and brains). They differ in how they
explain the relationship between minds and brains. The most famous version is called
interactionism, due to Descartes. This says that (1) there are minds (things that think,
but are not physical, and that obey only psychological laws), and (2) there are brains
(things that are physically extended in space and that obey only physical and biological
laws), and (3) minds and brains interact. The problem for interactionism as a solution
to the mind-body problem is that there is no good explanation of how they interact.
After all, one is physical and the other isnt. So you cant give a physical explanation
of how they would interact, because the laws of physics dont apply to the mind. And
you cant give a psychological explanation of how they interact, because the laws of
psychology dont apply to the brain. Similarly, according to Colburn, when applied
to the software-hardware problem, an interactionist perspective fails to account for the
relation between abstractions and concrete things, presumably because the relations
themselves are either abstract, in which case they dont apply to concrete things, or
they are concrete and, so, dont apply to abstractions. (There is, however, a possible
way out, which we will explore in depth in Ch. 14, namely, perhaps the relationship
between them is one of implementation, or semantic interpretation, along the lines of
Subers theory.)
There is another version of dualism, called the dual-aspect theory, due to Spinoza.
Here, instead of saying that there are two different kinds of substance, mental
substance and physical substance, it is said that there is a single, more fundamental
kind of substance of which minds and brains are two different aspects. (For Spinoza,
this more fundamental substance, which he believed had more than just two aspects,
though we humans are only cognizant of those two, was nature, which he identified
with God.) As a solution to the software-hardware problem, Colburn points out that
we would need some way to characterize that more fundamental underlying substance,
and he doesnt think that any is forthcoming. Again, however, an alternative possibility
is to think of how a single abstraction can have multiple implementations.
Finally, another dualism is known as parallelism: There are minds, and there
are brains; they are not identical (hence this is a dualistic theory), but they operate
in parallel, and so there is no puzzle about interaction. There are two versions of
parallelism. One, called occasionalism, says that God makes sure that every mental
event corresponds to a parallel brain event (this keeps God awfully busy on very small
matters!). Another, called pre-established harmony, says that God set things up so that
minds and their brains work in parallel, much in the way that two clocks can keep
the same time, even though neither causally influences the other. For Colburn, this
would mean that a concrete or hardwired program would behave the same way as an
abstract or textual program CHECK THIS. It is also consistent with my view that a
single abstraction can have two parallel implementations.
MAYBE FACTOR OUT MY COMMENTS ON IMPLMN AND PUT THEM

HERE. Before exploring this notion of implementation in more detail (in Ch. 14)
we will consider more of the ontological puzzles that arise from the dual nature of
software, this time in the legal realm: Can programs be copyrighted? Or should they
be patented instead?
After all, if a program is a written text, then it can be copyrighted, but, if it is a
machine, then it can be patentedyet nothing can (legally) be both copyrighted and
patented!

1. The basic question, What is a computer program?, is not necessarily answered
by considering the nature of software vs. hardware, copyright vs. patent, etc.
We dealt with some of the issues in Ch. 7, and so some of the ideas from there,
especially those arising from my emails with Robin Hill, might be worth putting,
or rehearsing, in this chapter.
Presumably, a program is a textual expression (or, in the old, hardwired days,
a physical, hardware implementation of) an algorithm, where an algorithm
is something abstract, perhaps a mathematical structure of some kind, and a
program is something concrete (even if only written on non-magic paper). On
this view, perhaps the best analogy is that program is to algorithm as numeral
is to number. Another analogy would be that of a proposition to a sentence;
presumably, the English sentence It is snowing. and the French sentence Il
neige. both express the same (abstract) proposition that it is snowing.
So, for instance, a Turing-machine program for a binary search algorithm is
not identical to that algorithm, because that very same algorithm could have
been expressed in a Lisp program. The algorithm is something more abstract
and independent of any expression of it in a particular programming language.
Moreover, there might be several different Turing-machine programs or Lisp
programs that implement binary search.
One possible difference between these analogies is this: Presumably, there is
only one number one and arbitrarily many numerals representing it. But
given some task, say, sorting, there might be arbitrarily many algorithms that
accomplish it: First, there are many different, standard sorting algorithms:
insertion sort, Quicksort, merge sort, heapsort, etc. Second, for each of these,
there are many different possible variations; for instance, Shell sort is (arguably)
a variation of insertion sort. Third, for any of these, one can always interpolate
a small, do-nothing loop in the middle of the algorithm; is a standard insertion
sort with
begin
x := 5;
while x > 0 do
x := x 1;
end
interpolated somewhere in the middle of it a different algorithm from the same

insertion sort without that loop? So, where the standard understanding of the
number-numeral relationship is that it is a one-many relation, it would seem that
the algorithm-program relationship is many-many.
On the other hand, mathematical structuralists have argued that numbers are
not things (that exist is some Platonic heaven somewhere), but are more like
roles (think of the role of Hamlet in Shakespeares play)) in a mathematical
structure (defined, say, by Peanos axioms) that can be played by many
different things (think of actors like Richard Burton, Laurence Olivier, etc.),
such as different sets, Arabic numerals, etc. In that case, the number-numeral
relationship is also many-many.
2. Turings work clearly showed the extensive interchangeability of
hardware and software in computing. [Hartmanis, 1993, 11]
3. In a Turing machine, is its machine table software, or hardware? Its certainly
hardwired. If you think that it is a kind of category mistake to talk about whether
an abstract, mathematical entity such as a Turing machine can have software or
hardware, then consider this: Suppose you have a hardwired, single-purpose,
physical computer that (lets say) does nothing but accept two integers as input
and produces their sum as output. (This would be a physical implementation of
a Turing machine.) Is the program that runs this adder software or hardware? I
would say that its hardware.
In a universal Turing machine, is its machine table (which is (a description of?)
its fetch-execute cycle) software, or hardware? And what about the program that
is stored on its tape? Again, if you prefer to limit the discussion to physical
computers, then consider this: Suppose that you have a smartphone, one of
whose apps is a calculator that also can add two integers (but, not only can
the calculator do other mathematical operations, the smartphone itself can do
many other things (play music, take pictures, make phone calls, etc.), and it
can download new apps that will allow it to do many other things. (This would
be a physical implementation of a universal Turing machine.) Is the program
that runs the smartphones adder (calculator) software? And is the program that
allows the smartphone to do all of the above hardware? Id say that the program
that is stored on the Turing machines tape (for example, the smartphones
calculator app) is software; therefore, the universal Turing machines machine
table (for example, the smartphones CPU) is hardware; and, therefore, the
Turing machines machine table (for example, the adders program) is also
hardware.
4. On claims that hardware and software are logically equivalent:
Similar or analogous cases of such logical equivalence of distinct things are
Turing machines, the lambda-calculus, and recursive functions. Also, such
an equivalent-but-different situation corresponds to the intensional-extensional
distinction: Two intensionally distinct things can be extensionally identical.
5. On Moors definitions of software and hardware:
I cant modify a Java program, because I dont know Java. Is a Java program
therefore hardware for me? Suppose Pat cant modify the Java program, even
though Pat is a Java programmer (perhaps Pat has no access to the source code).
Is that Java program hardware for Pat? These are probably different senses of
cant.
6. On Subers definitions:
How (if at all) does Subers position relate to Searles [Searle, 1990] claim that
anything can be considered to be a computer, even the wall behind me.
7. By fixed programming we mean the kind of programming which

controls your automatic dishwasher for example. Here the sequence
of operations is fixed and built into the wiring of the control or
sequencing unit. Once started, the dishwasher will proceed through a
regular series of operations, washing, rinsing and drying. Of course,
if one wished, one could change the wiring to alter the program.
[Samuel, 1953, 12261227].
This is like a Turing machines hardwired program. Note, by the way, that
modern dishwashers allow for some programming by pushbuttons that can
alter its operations.
For the universal Turing machines software (program stored on its tape), Samuel
has this analogy:
Suppose you wished to give your assistant a large number of

instructions for manual computations all in advance. You could do
this by supplying him with a prepared set of instructions, or you could
dictate the instructions and have him write them down, perhaps at the
top of the same sheet of paper on which he is later to perform the
computations. Two different situations are here involved, although
at first glance the distinction appears trivial. In the first case the
instructions are stored on a separate instruction form, while in the
second case they are stored by the same medium which is used for
data. Both situations are found to exist in computing machines.
The first case is exemplified by certain machines which use special
program tapes. The second situation is becoming quite common in the
newer machines and is the case usually meant when the term stored
program is used. [Samuel, 1953, 1227]
We can now go back and consider one property of the stored
program method of operation which is rather unique and which
really must be understood to appreciate the full value of such a
concept. This property is that of being able to operate on the
instructions themselves just as if they were ordinary data. This means
that the entire course of a computation can be altered, including the
operations themselves, the choice of data on which the operations
are to be performed, and the location at which the results are to be
stored, and this can all be done on the basis of results obtained during
the course of the calculations through the use of conditional transfer
instructions. [Samuel, 1953, 1228]
Shades of Hayess magic paper!
8. On the difference between software (program) and the process produced by it

when executed:
In software culture, we no longer have documents. Instead, we

have software performances.
I use the word performance because what we are experiencing

is constructed by software in real time. Whether we are exploring a
website, playing a video game, or using an app on a mobile phone
to locate nearby friends or a place to eat, we are engaging with the
dynamic outputs of computation.
Although static documents may be involved, a scholar cannot
simply consult a single PDF or JPEG file the way 20th-century critics
examined a novel, movie, or TV program. . . . For instance, a user of
Google Earth is likely to experience a different earth every time he
or she uses the application. Google could have updated some of the
satellite photographs or added new Street Views and 3D buildings.
[Manovich, 2013, B11]
Not only does this emphasize what a process is, but it also points out an
interesting difference, relevant to copyright issues that we will explore in Ch. 13,
between the static, written text of a program and the potentially dynamic version
engraved on an executable CD. A similar point is made later in Manovichs
essay:
The attraction of reading the code approach for the humanities is
that it creates an illusion that we have a static and definite text we can
studyi.e., a program listing. But we have to accept the fundamental
variability of the actual software performances. So rather than
analyzing the code as an abstract entity, we may instead trace how
it is executed, or performed, in particular user sessions. To use the
terms from linguistics, rather than thinking of the code as language,
we may want to study it as speech. [Manovich, 2013, B13]
9. On Dennetts intentional stance:
See the discussion of levels of description in [Newell, 1980, 6]. One major
difference between Newell and Dennett is that the former holds that these levels
of description do not exist just in the eye of the beholder, but have a reality
in. . . the real world [Newell, 1980, 173].
10. The earliest use of the word software in its modern sense has been traced back
to [Tukey, 1958, 2]:
Today the software comprising the carefully planned interpretive

routines, compilers, and other aspects of automative [sic]
programming are at least as important to the modern electronic
calculator as its hardware of tubes, transistors, wires, tapes and the
like.
(See [Shapiro, 2000a], [Mahoney and Haigh (ed.), 2011,

238, n. 20], and http://www.historyofinformation.com/expanded.php?id=936 for
more information on the history of the word, which is likely older than Tukeys
use.) This use of the term strongly suggests that the things that count as software
are more abstract than the things that count as hardware.
Interestingly, an earlier citation than Tukeys, [Carhart, 1956, 149], equates

software with the people who operate a computer system (where the computer
system is identified as the hardware); programs (or other modern notions
of software) are not mentioned. (An even earlier use, dating to 1850
[Shapiro, 2000a, 69], equates it with vegetable and animal matterseverything
that will decompose in the realm of rubbish-tip pickers; insofar as
decomposition is a form of changeability, this is, at least, consistent with Moors
interpretation!)
11. Exercise for the Reader:

Find a (short) article on the mind-body problem (for example, the Wikipedia
article at http://en.wikipedia.org/wiki/Mind-body problem). Replace all words
like mind, mental, etc., with words relating to software; and replace all
words like body, brain, etc., with words like hardware, computer, etc.
Discuss whether your new paraphrased article makes sense, and what this says
about the similarities (or differences) between the mind-body problem and the
software-hardware problem.6
12. On the idea that some kinds of software, in particular, computer programs (and
the activity of programming) are mathematical in nature, see [Dijkstra, 1974,
Scherlis and Scott, 1983, Hoare, 1986]. Hoare, for example, says that computer
programs are mathematical expressions (p. 115). This suggests that (at
least some kinds of) software are abstract, mathematical entities (or linguistic
expressions thereof, perhaps more along the lines of numerals than of numbers).
13. [Vahid, 2003, 27, 31, 32] suggests that the processors, memories, and buses
what we previously considered a systems unchangeable hardwarecan actually
be quite soft (p. 32). What he seems to mean by this is that embedded
systemshidden computing systems [that] drive the electronic products around
us (p. 27)can be swapped for others in the larger systems that they are
components of, thus becoming changeable in much the way that software is
normally considered to be. But this seems to me to just be the same as the old
rewiring of the early days of programming (except that, instead of changing the
wires or switches, it is entire, but miniaturized, computers that are changed).
14. In an unpublished talk given to the Center for Cognitive Science at SUNY
Buffalo (21 September 2005), Amnon H. Eden offers three ways to think about
what a computer program is. The first is as a mathematical expression (citing
the work of C.A.R. Hoare). This is consistent with the view mentioned elsewhere
in this chapter of a program being like a numeral, where an algorithm is like
a number. But it is a bit different, I think; here, the idea is that programs
themselves are mathematical objects capable of being studied mathematically.
We will return to this theme in Chapter 16.
The second way is as a natural kind (see 3.4). Computer programs exist, so
we should study them as we find them in nature, so to speak. We can try to
6 Thanks to James Geller for this idea.
categorize them, identify their properties, determine there relative merits, and so
on.
The third way is as an engineered artefact. Whereas the study of programs
as natural kinds is more or less descriptive, the study of them as engineered
artifacts is normative: We should try to specify ways in which programs can
be designed so as to be more efficient, more reliable, etc.
Finally, Eden offers a fourth way: as a cognitive artefact: Software is conceived
and designed at a level of abstraction higher than the programming language.
15. An example of the algorithm-program relationship that is along the lines of

the number-numeral relationship is offered by Stewart Shapiro: When trying
to study a dynamic process mathematically,
It is common practice in mathematics. . . [to associate] a class of

(static) mathematical objects with the process in question and then
transferring results about the mathematical objects to the processes.
In mathematical logic, for example, a process of deductive reasoning
is studied through the use of formal deductions; in mathematical
physics, the movement of particles is studied by means of an
associated class of curves in geometrical space. . . . . . . The technique
employed in the theory of computability is to associate an algorithm
with a written set of instructions for computation, usually in a
formal language, and to study the class of algorithmic descriptions
syntactically. [Shapiro, 2000b, 204]
16. On Moors changeability criterion:

[Piccinini, 2008, 3.1.13.1.2, p. 40] distinguishes between
hard programmability and soft programmability: The former refers to the
mechanical modification of. . . [a computers] functional organization; the latter
refers to modification involv[ing] the supply of appropriately arranged digits
(instructions) without manually rewiring any of the components. This can be
viewed either as a further breakdown of software into these two kinds, or else as
limiting the changeability to soft changeability.
17. [Irmak, 2012] argues that software is an abstract artifact, where an

artifact. . . is an intentional product of human activity (pp. 5556) and an
abstract object is one that is non-spatial but that may or may not have
some temporal properties (p. 56). Irmak likens software to another abstract
artifact: musical works (2, pp. 65ff). I agree that there are close similarities.
For instance, I think that a Turing machine (or any hard-wired computer that can
perform only one task) is like a music box that can only play one tune, whereas
a player piano is like a universal Turing machine.
One difference between software and music that Irmak points out concerns
a change or a revision on a musical work once composed (p. 67). This
raises some interesting questions: How should musical adaptations or jazzy
versions of a piece of music be characterized? What aobut different players
interpretations?: One pianists version of, say, Bachs Goldberg Variations will
sound very different from anothers, yet, presumably, they are using the same
software. Are there analogies to these with respect to software? (In addition,
problems about small changes in software seem analogous to issues of personal
identity through time.)
[Irmak, 2012, 68] says, the idea that software and musical works are created
is. . . central to our beliefs. Perhaps, but here there are similarities with issues
in the philosophy of mathematics: Are theorems or proofs similarly created?
Or should we take a mathematically Platonic attitude towards all of these kinds
of things, and revise our manifest image view that they are created to a more
scientific image view that they are discovered? Mendelssohn (or Mill?) is
alleged to have been depressed when he learned that there were only a finite
number of possible combinations of notes, hence only a finite number of possible
musical compositions.
For more on the relationships between software and improvisational music, see
10.5.2.
18. [H]ardware is at least tangible. Software by contrast is elusively
intangible. In essence, it is the behavior of the machines when
running. It is what converts their architecture to action. . . .
[Mahoney and Haigh (ed.), 2011, 33].
On this view, programs (as texts) are part of the hardware; it is the process that is
the software. Whether or not one wants to draw the boundary there, it does point
out the different role of process from both computer and program. As a historian,
Mahoney is also concerned with the fact that processes are ephemeral: They
leave no record, and are difficult, if not impossible, to repeat (legacy software
can only be run on machines that no longer exist).
19. Possible discussion question for end of chapter exercise:
Is the machine table of a Turing machine software or hardware? It is certainly
hardwired, at least in the sense that it is unchangeable: It is, after all,
definitional of that Turing machine, that is, an essential property of it.
Is the machine table of a universal Turing machine software or hardware? The

machine table of a universal Turing machine is simply its fetch-execute cycle.
What about the program stored on its tape?
My answers: The program stored on the tape of a universal Turing machine is

software; its fetch-execute-cyle machine table is hardware; and the machine table
of a (non-universal) Turing machine is also hardware.
20. On Moors definition of software as changeable:
I cant modify a Java program, because I dont know Java; is the Java program
hardware for me? Suppose that Kim cant modify a certain Java program, even
though Kim is a Java programmer, because Kim has no access to the source code;
is that program hardware for Kim? Are these different senses of cant?
21. Not sure where to put this, but possibly relevant to Subers notion that anything
can be considered to be software:
The presentation in [Denning and Martell, 2015, 40, Fig. 3.3] of Shannons
theory of the communication of information suggests that the encoding of a
message is communicated along a channel to be decoded by a receiver. This
is very similar to the idea of input being converted into output. Of special
interest is the idea that noise is any disruption that alters the signal. Wouldnt a
computation performed on an input signal alter it? In that case, one could either
consider such a computation to be noise, or one could consider noise to be a
computation.

Bond, Gregory W. (2005), Software as Art, Communications of the ACM 48(8)
(August): 118124.
1. A programming language is a language after all, albeit a hightly constrained

one. As such, it is a perfect medium for the poet comfortable with other highly
constrained poetic forms like the sonnet or haiku (p. 123).
2. On the literary value of programs, see also:
(a) Chandra, Vikram (2014), Geek Sublime: The Beauty of Code, the Code of
Beauty (Graywolf Press).
See also:
Gleick, James (2014), A Unified Theory (review of Chandra 2014),
New York Times Book Review (24 August): 17; accessed 8 September
from:
http://www.nytimes.com/2014/08/24/books/review/geek-sublime-by-
vikram-chandra.html
Knuth, Donald E. (1984), Literate Programming, The Computer Journal
27(2): 97111.
i. Accessed 29 Jully 2014 from:
http://comjnl.oxfordjournals.org/content/27/2/97.full.pdf+html
ii. Reprinted in:
Knuth, Donald E. (1992), Literate Programming CSLI Lecture Notes 27
(Stanford, CA: Center for the Study of Language and Information),
http://www-cs-faculty.stanford.edu/uno/lp.html
(b) Schneider, David (2007), Neatness Counts: Messy Computer Code Is
Normally Frowned OnBut Not Always, American Scientist (May-June):
213214
http://www.americanscientist.org/issues/pub/neatness-counts
(c) Turner, Zeke (2015), Beautiful Code, The New Yorker (30 March): 2122,
http://www.newyorker.com/magazine/2015/03/30/beautiful-code (accessed
4 April 2015)
On the first art auction of computer code.
3. For some examples of aesthetic or playful programs, see:
(a) Most Adaptable Programs (1991, 1996), accessed 29 July 2014 from:
http://www2.latech.edu/acm/helloworld/multilang.html
(two examples of programs that can be compiled and run in several
different programming-language systems, as if there were a sentence that was
grammatical and meaningful in both English and Frenchis there such a
sentence?)
(b) dodsond1, Text to Pig Latin translator, 12th International Obfuscated C
Code Contest (1995), accessed 29 July 2014 from:
http://www.ioccc.org/1995/dodsond1.c
and
http://www.ioccc.org/years-spoiler.html
(must be seen to be believed!)
Duncan, William (2009), Using Ontological Dependence to Distinguish between

Hardware and Software, accessed 1 August 2014 from:
https://www.jiscmail.ac.uk/cgi-
bin/webadmin?A3=ind1103&L=SWORD&E=base64&P=519469&B=
050201060409000909050704&T=application
Argues that a formal ontology is needed before a useful distinction between

hardware and software can be made, and that, on such an ontology (the
Basic Formal Ontology), a piece of computational hardware is an ontologically
independent entity, whereas a software program is an ontologically dependent
entity.
Eden, Amnon H.; & Turner, Raymond (2007), Problems in the Ontology of Computer
Programs, Applied Ontology 2(1): 1336.

http://www.eden-study.org/articles/2007/problems-ontology-programs ao.pdf
Gemignani, Michael (1981), What Is a Computer Program?, American Mathematical

Monthly 88(3) (March): 185188.
Argues that technical artefacts are physical structures with functional properties
Mahoney, Michael Sean (2011), Histories of Computing, ed. by Thomas Haigh

(Cambridge, MA: Harvard University Press).
Ch. 13, Extracts from Software as ScienceScience as Software, discusses the
history of software; the section titled Extract 3: Software as Science is especially
interesting.
Miller, Christopher A. (guest ed.) (2004), Human-Computer Etiquette: Managing
Expectations with Intentional Agents, Communications of the ACM 47(4) (April): 31
34.
Contains some interesting follow-ups from the perspective of computer science to

Dennetts theory of different stances.
Turner, Raymond; & Eden, Amnon H. (2007), Towards a Programming Language

Ontology, in Gordana Dodig-Crnkovic & Susan Stuart (eds.), Computing, Information,
CognitionThe Nexus and the Liminal (Cambridge, UK: Cambridge Scholars Press):
147159.

http://www.eden-study.org/articles/2007/towards-pl-ontology dodigstuart.pdf
Chapter 13

II. Can Software Be
Copyrighted or Patented?
Figure 13.1: 2010,

c American Scientist
509
510 CHAPTER 13. COPYRIGHT VS. PATENT
Figure 13.2: 2007,

c UFS, Inc.

1. Required:
Newell, Allen (1985-1986), The Models Are Broken, the Models Are Broken,
University of Pittsburgh Law Review 47: 10231031.
http://digitalcollections.library.cmu.edu/awweb/awarchive?type=file&item=356219
For follow-up (including links to commentary by Donald Knuth), see:
Newell 1986: The Models Are Broken, accessed 5 August 2014 from:
https://web.archive.org/web/20120527101117/http://eupat.ffii.org/papers/uplr-
newell86/index.en.html
Samuelson, Pamela (1990), Should Program Algorithms Be Patented?,
Communications of the ACM 33(8) (August): 2327,
https://www.law.berkeley.edu/php-
programs/faculty/facultyPubsPDF.php?facID=346&pubID=94 (accessed
9 November 2015)
Samuelson, Pamela (2015), Software Patents Are Falling Down,
Communications of the ACM 58(11) (November): 2729.
Koepsell, David R.; & Rapaport, William J. (1995), The Ontology of Cyberspace:
Questions and Comments, Technical Report 95-25 (Buffalo: SUNY Buffalo
Department of Computer Science) and Technical Report 95-09 (Buffalo: SUNY
Buffalo Center for Cognitive Science)
http://www.cse.buffalo.edu/rapaport/Papers/cyber.pdf
This document consists of two papers, both originally presented at the Tri-
State Philosophical Association Meeting, St. Bonaventure University, 22 April
1995:
(a) Koepsell, David (1995), The Ontology of Cyberspace: Preliminary
Questions
(b) Rapaport, William J. (1995), Comments on The Ontology of

Cyberspace
See esp. Koepsells 1 and 4, and Rapaports 2 and 3.
13.2 Introduction
We are trying to understand what algorithms, computer programs, and software are,
and how they are related to each other. One way to approach these questions is by
considering the legal issues of whether any of them can be copyrighted or patented.
(There is also a legal notion of trade-secret protection, which we will not explore.)1
One of the first questions is what kind of entity might be copyrighted or patented:
algorithms?
(These seem to be abstract, mathematical entities. Can abstract, mathematical
entitiessuch as numbers, formulas, theorems, proofs, etc.be copyrighted or
patented?)
computer programs?
(These seem to be implementations of algorithms, expressed in a programming
language. We have seen that programs might be analogous to numerals (as
opposed to numbers)can numerals be copyrighted or patented?)
programs written on paper?

(These might seem to be literary works, like poems or novels, which can be
copyrighted but not patented.)2
programs implemented in hardware (that is, machines)?

(If programs are implementations of algorithms, then are hardware
implementations of programs thereby implementations of implementations of
algorithms? And physical implementations might seem to be the kind of thing
that is patentable but not copyrightable.)
software?
(Software might be a neutral term covering both algorithms and programs, or
software might be a more controversial entity not necessarily indistinguishable
from certain kinds of hardware, as we saw in the previous chapter.)
Here are some excerpts from the legal definitions. The full sources are: Copyright
Basics, accessed 5 August 2014 from: http://www.copyright.gov/circs/circ01.pdf
and United States Patent and Trademark Office, accessed 5 August 2014 from
http://www.uspto.gov/patents/index.jsp
Note what is said about computer programs, procedures, methods, and processes!
Copyright:
Copyright is a form of protection provided by the laws of the United States (title 17,
U.S.Code) to the authors of original works of authorship, including literary, dramatic,
musical, artistic, and certain other intellectual works. . . . [T]the owner of copyright [has]
the exclusive right to do and to authorize others to do the following:
1 But see the Further Readings for essays dealing with trade secrets, especially Bender 1985-1986, and
also [Samuelson et al., 1994, 2.2.1,5.3.3].

2 See the Further Readings for Ch. 12 for some references to work on the literary value of programs, at
Bond 2005.
reproduce the work. . .

prepare derivative works based upon the work
distribute copies. . . of the work to the public by sale or other transfer of ownership,
or by rental, lease, or lending
perform the work publicly, in the case of literary, musical, dramatic, and
choreographic works, pantomimes, and motion pictures and other audiovisual
works
display the work publicly, in the case of literary, musical, dramatic, and
choreographic works, pantomimes, and pictorial, graphic, or sculptural works,
including the individual images of a motion picture or other audiovisual work. . .
What Works Are Protected?

Copyright protects original works of authorship that are fixed in a tangible form
of expression. The fixation need not be directly perceptible so long as it may be
communicated with the aid of a machine or device. Copyrightable works include the
following categories:
1. literary works
2. musical works, including any accompanying words
3. dramatic works, including any accompanying music
4. pantomimes and choreographic works
5. pictorial, graphic, and sculptural works
6. motion pictures and other audiovisual works
7. sound recordings
8. architectural works
These categories should be viewed broadly. For example, computer programs and most
compilations may be registered as literary works; maps and architectural plans may be
registered as pictorial, graphic, and sculptural works.
What Is Not Protected by Copyright?

Several categories of material are generally not eligible for federal copyright protection.
These include among others:
works that have not been fixed in a tangible form of expression (for example,
choreographic works that have not been notated or recorded, or improvisational
speeches or performances that have not been written or recorded)
titles, names, short phrases, and slogans; familiar symbols or designs; mere
variations of typographic ornamentation, lettering, or coloring; mere listings of
ingredients or contents
ideas, procedures, methods, systems, processes, concepts, principles, discoveries,
or devices, as distinguished from a description, explanation, or illustration
works consisting entirely of information that is common property and containing
no original authorship (for example: standard calendars, height and weight charts,
tape measures and rulers, and lists or tables taken from public documents or other
common sources)
Patents:
A patent is an intellectual property right granted by the Government of the United States
of America to an inventor to exclude others from making, using, offering for sale, or
selling the invention throughout the United States or importing the invention into the
United States for a limited time in exchange for public disclosure of the invention when
the patent is granted.
There are three types of patents. Utility patents may be granted to anyone who invents
or discovers any new and useful process, machine, article of manufacture, or composition
of matter, or any new and useful improvement thereof. . . . Design patents may be
granted to anyone who invents a new, original, and ornamental design for an article of
manufacture. Plant patents may be granted to anyone who invents or discovers and
asexually reproduces any distinct and new variety of plant.
Furthermore, utility patents include inventions involving computer

programming, computer program listings
(http://www.uspto.gov/patents/resources/types/utility.jsp).
13.3 Preliminary Considerations

Lets begin by trying to distinguish between algorithms and corresponding computer
programs. Recall that algorithms compute functions. So let us begin by considering
some function, that is, some set of input-output pairs that are functionally related. That
function can be implemented by different algorithms.
What do I mean by implemented? Well explore that in Ch. 14. A synonym is
realized (that is, made real). For now, we will just say that implementation is the
relation between a function and an algorithm that computes it: Any such algorithm
implements that function.
Take a look at Figure 13.3 (at the end of this chapter). In the figure, merely as an
example, I consider a function being implemented by two different serial algorithms
and one parallel algorithm.
(For a concrete example, consider a sorting function that takes as input a set of
rational numbers and that yields as output a sequence of those numbers sorted in
increasing order. One serial algorithm that implements this function is Quicksort;
another is Merge Sort. A parallel algorithm that implement the function might be a
parallel version of Merge Sort.)
Take a look at serial algorithm 1. The figure suggests that it might be expressable
in two different programming languages: a parallel programming language and a serial
programming language.
Consider now our original function implemented by serial algorithm 1 that is, in
turn, expressed in serial programming language 2. There might be several ways in
which that program might itself be implemented:
on a virtual machinea theoretical model of computation for our serial
programming language 2,
on an actual computer that uses our serial programming language 2 as its
machine language,
13.4. COPYRIGHT 515
compiled into a serial machine language,

compiled into a different serial machine language,
compiled into a parallel machine language.
And now consider parallel algorithm 3 that implements our original function.
It might be expressed in a serial programming language 3, which itself might be
implemented in a virtual machine, an actual computer, or compiled into several
different machine languages. Or it might be expressed in a parallel programming
language 4, which itself might be compiled into serial machine language 5 or else
into a parallel machine language.
And so on.
Which (if any) of these should be copyrightable?
Which (if any) of these should be patentable?
13.4 Copyright
What does it mean to copyright something? For a person to copyright a text is to give
that person the legal right to make copieshence the term copyrightof that text for
the purpose of protecting the expression of ideas. An idea cannot be copyrighted; only
the expression of an idea can be copyrighted. Ideas are abstract, perhaps in the way that
algorithms are abstract. But if you express that idea in languageif you implement
it in languagethen the expression of that idea in that language can be copyrighted,
perhaps in the way that a computer program that implements an algorithm might be
copyrightable.
Consider [Mooers, 1975, 50] on the distinction between an uncopyrightable idea
and a copyrightable expression of an idea:
Where does the expression leave off, and the idea take over? The
best insight into this matter comes from discussions of copyright as applied
to novels and dramatic productions. In these, expression is considered
to include the choice of incident, the personalities and development of
character, the choice of names, the elaboration of the plot, the choice of
locale, and the many other minor details and gimmicks used to build the
story. In other words, expression is considered to include not only the
marks, words, sentences, and so on in the work, but also all these other
details or structures as they aggregate into larger and larger units to make
up the expression of the entire story.
In other words, after the bare abstract idea has been chosen (e.g.,
boy meets girl, boy loses girl, boy wins girl), the expression to which
copyright applies covers the remaining elements of original choice and
artistry which are supplied by the author in order for him [sic] to develop,
express, and convey his version of the bare idea.
Recall FROM SOME EARLIER CHAPTER ON SYN VS SEM our picture of two
domains: the world and our description of the world in some language. The former is
the realm of a semantic interpretation of the latter, which is the realm of syntax. Perhaps
copyright is something that applies only to the realm of syntax, whereas patents are
things that apply only to entities in the world.
Algorithms, construed abstractly, seem more like ideas than like expressions,
which would suggest that they cannot be copyrighted. On this view, it would be a
computer programa text written in some languagethat would be copyrightable.
Programs are definitely expressions: They are non-dramatic literary works. Why
literary? After all, they dont seem to read like novels! (But see the references cited
at the end of Ch. 7 on literate computing.) But all that literary means in this context
is that they can be written and read. Moreover, programs can be performed, that is,
executed, just like lectures, plays, or movies.
On the other hand, there is an entity that is even more abstract than an algorithm,
and with respect to which it is the algorithm that appears to be a detailed expression,
namely, the function that the algorithm computes. So, one could argue that it is a
function that is more like a bare abstract idea (boy meets girl, etc.) and that it is an
algorithmwhich would have to specify how the boy meets the girl, etc.that is more
like an expression (in this case, an expression of a function). On this view, it is the
algorithm that would be copyrightable!
So, here are two problems to consider:
Problem #1
Consider the input-output behavior of a program-being-executed (a process;
see 3.7). That is, consider how the process looks and feels to a user. A
clever programmer could create a new algorithm (or recreate an old one) and
then implement a new program that will have the same look and feel. Should
that new program be separably copyrightable?
Consider two computer designcompanies; call them Pear Computers and
Macrohard. Pear Computers might write an operating system with windows,
icons, and a mouse. Macrohard, seeing how this operating system works,
might write their own operating system that also uses windows, icons, and a
mouse and that would have the same (or almost the same) functionality and the
same (or almost the same) look and feel, but it would be a different expression
of that same (or almost the same) idea, hence separably copyrightable. Or would
it?3
Problem #2
It is part of copyright law that you cannot copyright an article of utility; thats
something that can only be patented.
IS THAT RIGHT? NEED TO FIND SOURCE OF QUOTE
But surely an executable program is useful; that is, it is an article of utility.
3 This is not far from the actual situation that allegedly existed between the Xerox Palo Alto Research
Center and Apple Computer, on the one hand, and (later) between Apple and Microsoft, on the other. The
story is that Apple borrowed or was inspired by Xerox PARCs creation of the graphical user interface and
used it for its Mac OS, and that Microsoft borrowed or was inspired by Apples version and used it for
Windows. See [Fisher, 1989], [Samuelson et al., 1994, 23342335], [Gladwell, 2011], and
http://en.wikipedia.org/wiki/Apple Computer, Inc. v. Microsoft Corp. (accessed 8 February 2014).
13.5. PATENT 517
But, according to Mooers, it is a translation of a copyrightable text; hence, it,

too, is copyrightable.
One diagnosis of these problems is that there is no legal definition of look and
feel [Samuelson, 1989]. Another diagnosis is that the distinction between an idea
and its expression may not be applicable in the realm of software [Samuleson, 1991],
[Koepsell, 2000]. We will explore this in a later section.
WHERE? NEED A SECTION ON KOEPSELL!
13.5 Patent
Patents are weird: Not just anything can be patented, but lots of strange things can
be! (I have read patents for devices or techniques that claim to be based on carefully
developed scientific theories, but that I know, from personal experience, are not based
on any such thing, and I have read patents that claim to provide mechanisms for
completely solving problems that have not yet been solved, and may never besuch
as fully automatic, high-quality machine translation.)
The purposes of a patent are to foster invention and to promote the disclosure of
an invention. So, the inventors ideas must be made public. (Does that mean that they
have to be implemented?)
It follows that you cannot patent something that would inhibit invention, namely,
you cannot patent any of the following:
abstract ideas
mathematical processes
scientific principles
laws of nature (and yet genes have been patented!)
mental processes
But you can patent any new and useful process, machine, article of manufacture,
or composition of matter, or any new and useful improvement thereof (see the
definition of patent, above). Does process here include software?
Here are some problems to consider:
Each step of certain cognitive algorithms (such as those used in reasoning,
perception, or natural-language generation and understanding) are mental
processes. Does that mean that some algorithms cannot be patented, even
if others can? If so, how does one distinguish between these two kinds of
algorithms?
Lots of software are mathematical processes. Does this mean that some, but not
all, software cannot be patented?
If too much software is patented, that might impede invention, contrary to one of
the explicit purposes of patent law.
13.6 Allen Newells Analysis

In this section, we will look at one persons analysis of these problems. But it is not
just any one person; it is a well-respected computer scientist. And it is not just any
computer scientist; it is one of the founders of the fields of AI and of cognitive science,
and a winner of the Turing Award: Allen Newell. In a paper written for a law journal
[Newell, 1986], Newell argues that the conceptual models of algorithms and their
uses are broken, that is, that they are inadequate for discussions involving patents
(and copyrights).
Newells definition of algorithm is very informal, but perfectly reasonable, not
arcane; we can live with it: An algorithm is an unambiguous specification of a
conditional sequence of steps or operations for solving a class of problems. (See
7.5, above, and [Chisum, 1986, The Definition of an Algorithm, B, pp. 974977]
for other definitions.)
TO BE ADDED, MAYBE IN AN APPENDIX: THESE OTHER DEFS; SEE
http://www.cse.buffalo.edu/rapaport/584/c-vs-pat.html item 3.
Algorithms are mathematical objects, but there are non-mathematical algorithms,
so, Newell points out, you cannot make a decision on the patentability of algorithms on
the basis of a distinction between mathematical processes and non-mathematical ones.
Moreover, humans think by means of algorithms (my italics; CITE). This is a
very strong version of a thesis in the philosophy of mind called computationalism.
It is, in fact, one reason why Newell is so famous: Newell and Nobel-prize winner
Herbert Simon wrote one of the first AI programs, which they interpreted as showing
that humans think by means of algorithms (see [Newell et al., 1958]). Note that
Newell doesnt say what kind of algorithms he has in mind here, nor does he say
whether they are Turing-machineequivalent. His point is that some algorithms are
sequences of mental steps; so, you cant make patentability decisions based on any
alleged distinction between mental processes and algorithms.
He also points out that you cannot patent natural laws or mathematical truths (for
instance, you cannot patent the integers, or addition). Are algorithms natural laws?
(In 10.4, we saw that [Shapiro, 2001] thinks that they are.) Are they mathematical
truths? They are certainly among the items that some mathematicians study. In either
case, algorithms would not be patentable. On the other hand, they are applicable,
hence patentable. But all of computer science concerns application, and algorithms are
certainly part of computer sciences theoretical foundations.
In addition, Newell points out, it is difficult to distinguish an algorithm (an abstract
program) from a computer program (which adds implementation details that are not,
strictly speaking, part of the algorithm), because there are only degrees of abstraction.
For example, you could have a really good interpreter that would take an algorithm
and execute it directly (this is the holy grail of programming: to talk to a computer in
English and have it understand you without your having to program it to understand
you).
But, contrary to his own definition, he points out that algorithms need only be
specifications that determine the behavior of a system, not necessarily sequences of
steps. Programs in Prolog or Lisp, for instance, are sets, not sequences, of statements.
So, you cant identify whether something is an algorithm just by looking at it.
13.6. ALLEN NEWELLS ANALYSIS 519
The bottom line is that there are two intellectual tasks: A philosophical and
computer-scientific task is to devise good models (ontologies) of algorithms and
other computer-scientific items. This is a task that people like Amnon Eden, Barry
Smith, Brian Cantwell Smith, and Ray Turner are working on. (See [Smith, 1996,
Smith, 2002, Eden, 2007], CITE TURNER, CITE BARRY SMITH)
A legal task is to devise good legal structures to protect these computational items.
This is a task that people like David Koepsell and Pamela Samuelson investigate. (See
[Koepsell, 2000] and Samuelsons writings in the Further Sources of Information at
the end of this chapter.)
I close with a quotation from [Newell, 1986, 1035, my italics]:
I think fixing the models is an important intellectual task. It
will be difficult. The concepts that are being jumbled together
methods, processes, mental steps, abstraction, algorithms, procedures,
determinismramify throughout the social and economic fabric. . . . The
task is to get. . . new models. There is a fertile field to be plowed here, to
understand what models might work for the law. It is a job for lawyers
and, importantly, theoretical computer scientists. It could also use some
philosophers of computation, if we could ever grow some.
Readers of this book, take note!

1. In case I need a definition of virtual machine, heres a possibility: A virtual
machine is a Turing machine that is simulated by a universal Turing machine
(see, for example, [Denning and Martell, 2015, 212]: A virtual machine is a
simulation of one computer by another. The idea comes from the simulation
principle behind Alan Turings Universal Machine. In other words, let t be
a Turing machine. Let u be a universal Turing machine. Encode all Turing
machines using, say, Turings coding scheme. Then encode ts code onto us
tape, along with ts data, and let u simulate t. As [Copeland, 1998, 153] says,
the universal machine will perform evry operation that t will, in the same order
as t (although interspersed with sequences of operations not performed by t).
The virtual t machine is a software version (a software implementation?) of the
hardware t machine. (See also:
\item
Pollock, John L.\
(2008),
What Am I?
Virtual Machines and the Mind/Body Problem,
{\it Philosophy and Phenomenological Research\/}
76(2) (March): 237--309.
for another possible defintion.)

[Samuelson et al., 1994, 1.3] discusses the notion in terms of conceptual
metaphors (such as the metaphor of paper for a word-processing program):
Cutting and pasting (in a virtual sense) is so easily accomplished in a

word processor tht one soon begins to think of the text as a physical
entity that can be picked up and moved around. (p. 2325)
By virtual here, they mean metaphorical. Note, in connection with our

discussion of magic in Ch. 3, that this virtual paper is an illusion: As
Samuelson et al. note, we have the illusion that, when inserting text, the old
text obligingly moves over to make room for the new words (p. 2324).
Thge automated spreadsheet metaphor has so fundamentally changed

the experience of using a computer that users feel they are using a
spreadsheet, not just a computer. (p. 2325)
They go on to argue that, because conceptual metaphors (by which, I take

it, they mean virtual machines) are remote from the program text. . . and partly
because of their abstract character, they woud likely be regarded as unprotectable
by copyright law (p. 2326).
2. [Galbi, 1971] offers a useful history of computer patents. One point that becomes
clear from reading it is that the issue of whether software can or should be
patented or copyrighted is not merely an ontological issue. It is also, perhaps

even more so, an economic issue.
Initially, a computer operating in accordance with a particular program was
patentable if the program is new, useful, and unobvious [Galbi, 1971, 274].
That is, it was the physical machine executing the program that was patentable.
This was later modified to allow the method or algorithm for solving certain
problems to be patentable irrespective of what apparatus is used to solve the
problem [Galbi, 1971, 274]a clear move away from the patentability of a
piece of hardware to the patentability of its software instead. Arguably, though,
this could be seen as a move to say that it is still a machine that is patentable, but
it doesnt have to be a specific machine: It could be any machine that is executing
the program under consideration.
Related to this is the following excerpt from a footnote to a 1969 ruling in a case
called Prater and Wei:
In one sense, a general purpose digital computer may be regarded as a
storeroom of parts and/or electrical components. But once a program
has been introduced, the general purpose digital computer becomes
a special purpose digital computer (i.e. a specific electrical circuit
with or without electromechanical components) which along with the
process by which it operates, may be patented, subject, of course, to
the requirements of novelty, utility and nonobviousness.
But think of a smartphone capable of running multiple apps or a computer
capable of running multiple programsmore generally, of a universal
Turing machine capable of simulating multiple Turing machines. Might a
(physical implementation of) universal Turing machine running one program be
patentable, whereas the very same machine running a different program not be
patentable? Is it one machine, or many?
An aspect of the economic, rather than purely philosophical/ontological, issue
concerns whether copyright protection is narrower or broader than patent
protection; this is subject to some degree of disagreement, but the most
that can clearly and absolutely e said is that the nature of protection granted
by a patent is fundamentally differentfrom thetype of protection granted by
a copyright [Galbi, 1971, 274n1]. And another economic aspect concerns
whether or not these patents are, in fact, enforceable [Galbi, 1971, 277].
3. Samuelson 1990, p. 23 (required readings), points out that computer scientists
differ on what is the proper object of legal protection: code (or program), look
and feel, algorithm. So, these need to be distinguished. What is look and feel?
Is it related (or identical) to the process, that is, the program as it is being
executed? Is it the output of the program?
4. Although programs are texts and their texts can be valuable, the
most important property of programs is their behavior (i.e., the set
of results brought about when program instructions are executed).
[Samuelson et al., 1994, 2314]
[The primary source of value in a program is its behavior, not its text.
Does this mean that its the idea, not its expression that counts? I dont
think so, simply because the behavior of a program is more than merely the
idea. (The idea might be more akin to the specifications, or the input-
output behavior.) But behavior here might be something akin to look and
feel. So, what to the authors mean, exactly, by behavior?
5. A second feature of programs is that the behavior of one program can be

precisely reproduced by a textually entirely different program, so program text
and behavior are independent [Samuelson et al., 1994, 2315]
6. Third, programs are, in fact, machines (entities that bring about useful
results, i.e., behavior) that have been constructed in the medium of
text (source and object code). The engineering designs embodied
in programs could as easily be implemented in hardware as in
software, and the user would be unable to distinguish between the
two. [Samuelson et al., 1994, 2316]
But programs, even if engraved on a CD-ROM rather than merely inked on
paper, cant be executed without some kind of interpreting (and behaving (e.g.,
printing) machine. So, its not really the program by itself that can bring about
useful results, but the combination of program and intepreting/behavingin
short, executingdevice.
The second point in this quote about hardware and software, is, of course, the
point about the relation of a Turing machine to a universal Turing machine.
(Could there be a Turing test imitation game to disguish between these?!) I
think that this is a separate issue from the question about whether a program is a
machine.
This is clarified a bit later:
Not only do consumers not value the text that programmers write
(source code), they do not value any form of program text, not
even object code. Computer science has long observed that software
and hardware are interchangeable: Any behavior that can be
accomplished with one can alsob e accomplished with the other. In
principle, then, it would be possible for a puchaser of, say, a word
processor to get a piece of electronic hardware that plugged into her
computer, instead of a disk containing object code. When using
that word processor, the purchaser would be unable to determine
whether she was using the software or the hardware implementation.
Right! Thats the point about universal Turing machines.

Somewhere, Samuelson et al. remark that no one can read the object code on
a CD-ROM, which marks a difference between software and, say, the script of
a play. But that may be the wrong analogy. What about a moved on a DVD?
We cant read the DVD any more than we can read a CD-ROM (in both cases,
we need a machine to interpret them, to convert them to a humanly perceivable
form). Or even a movie on film: We dont typically look at the individual frames;
we need them projected at a certain speed.
7. On programs as machines, [Samuelson et al., 1994, 2320] go on to say:
Physical machines. . . produce a variety of useful behaviors, and so it

is with programs.
And then they add, in a footnote:
Traditional texts may tell humans how to do a task; they do not,

however, perform the task.
But the same is true for a program, whether written in ink on paper, as source
code entered into a computer, or as object code after compilation: It still takes
a physical machine to do what the program tells it to do. The physical
machine translates or associates parts of the program text (however inscribed)
with switch settings in the computer. Once those switches are set in accordance
with the program, the computerthe physical machinewill perform the task.
8. All this stands in sharp contrast to traditional literary works which

are valued because of their expression. . . . Programs have almost
no value to users as texts. Rather, their value lies in behavior.
Yes, but this is a misleading analogy. A better analogy is not between, say,
a program and a novel, but between a program and musical score or a play
script, and between a programs behavior and a performance of that music or
play. Only musicians and literary or theatrical people would be interested in the
score or the script (just as only programmers would be interested in a program
text). Audiences are more interested in the performance (in all cases).
(There is one major difference between programs/behavior on the one hand and
scores-scripts/performances on the other: The very same play or score will be
performed (interpreted) very differently by different performers. But the very
same program should be executed identically by different computers. That,
however, may have more to do with the detail of the instructions than with the
nature of the interpreters. For more on this, see the discussion about recipes in
10.5.2.)
9. While conceiving of programs as texts is not incorrect, it is seriously

incomplete. A crucially important characteristic of programs is
that they behave; programs exist to make computers perform tasks.
I agree with the first sentence. For reasons cited above, I disagree with the
claim that the program itself behaves. But I agree with the final claim: It is
the computer that behaves, that does something, that performs a task. And it
does so in virtue of its program. But the program by itself is static, not dynamic.
What tasks to programmed computers perform? A Turing machine can print,
erase, and move; and it can do anything else that can be decomposed into those
primitive actions. More sophisticated computers can, of course, do many other
things, though, at bottom, all theyre doing are forms of printing, erasing,
and moving (such as printing certain pixels on a screen)
10. Behavior is not a secondary by-product of a program, but rather an
essential part of what programs are. To put the point starkly: No
one would want to buy a program that did not behave, i.e., that did
nothing, no matter how elegant the source code prose expressing
that nothing. [Samuelson et al., 1994, 2317]
I suspect that Samuelson et al. are using program not to refer to text but to a
product intended to be installed in a computer and executed. For mathematically
speaking, there can certainly be programs that have no output (but see 7.5.
This becomes clearer later, when they point out that:
People pay substantial sums of money for a program not because
they have any intrinsic interest in what its text says, but because
they value what it does. . . (its behavior). The primary proof that
consumers buy behavior, rather than text, is that in acquiring a
program, they almost never get a readable instance of the program
text. [Samuelson et al., 1994, 2318]
11. Hence, every sensible program behaves. This is true even though the
program has neither a user interface nor any other behavior evident to
the user. [Samuelson et al., 1994, 2317]
In a footnote, they define user interface as the means by which the user and the
program interact to produce the behavior that will accomplish the users tasks.
A sensible program (by which I will now understand the program-as-product
that I described above) with no user interface might be akin to a Turing machine
that simply computes with no input other than what is pre-printed on its tape.
Having a user interface would seem to turn a sensible program into an
Wegnerian interactive program or an o-machine.
12. Two programs with different texts [such as Microsofts Excel
and Apples Numbers, to use a contemporary example] can have
completely equivalent behavior. A second comer can develop
a program having idential behavior, but completely different text
through. . . black box testing. This involves having a programmer
run the program through a variety of situations, with different inputs,
making careful motes about its behavior. A second programmer
can use this description to develop a new program that produces
the specified behavior (i.e., functionally identical to the firsomething
program) without having access to the text of the first program, let
alone copying anything from it. A skilled programmer can, in other

words, copy the behavior of a program exactly, without appropriating
any of its text. [Samuelson et al., 1994, 2317-2018]
Lets grant that Excel and Numbers dont have exactly the same input-output
behavior or even look-and-feel. But black-box testing could, indeedin theory,
at leastallow two completely textually different programs to have exactly the
same input-output behavior (and look-and-feel), for the simple, mathematical
reason that there can be more than one algorithm for implementing a function.
Note, however, that, in practice, such black-box testing is really akin to the
kind of inductive inference that underlies trial-and-error machines: Unless the
second programmer has a complete input-output specification, that programmer
is limited to reconstructing potentially infinite behavior on the basis of a finite
number of examples.
13. The independence of text and behavior is one important respect in
which programs differ from other copyrighted works. Imagine trying
to create two pieces of music that have different notes, but that
sound indistinguishable. Imagine trying to create two plays with
different dialogue and characters, but that appear indistinguishable
to the audience. Yet, two programs with different texts can be
indistinguishable to users. [Samuelson et al., 1994, 2318]
This is interesting! If true, it suggests that my analogies between
Abstraction/Implementation, score/performance, script/play, etc. (see 14.4,
might not be as close as I would want. But is it true? First, consider a play
and its script. Suppose that I write a version of Hamlet that incorporates all
of Shakespeares lines but adds material that is specifically designed not to be
uttered during a performance (perhaps extra characters who are always offstage
and silent). That would seem to be a counterexample to Samuelson et al. Second,
consider two different algorithms that both compute GCDs. Is it not conceivable
that, when the operations in each are decomposed to the primitive operations
of a Turing machine or the basic recursive functions, the algorithms would be
identical? Or consider a musical score for, say, a four-part choral work. I
could produce two different written scores simply by re-ordering the parts (or,
by writing them in different musical notations). Yet their performances would
not differ.
14. All of my philosophical quibbles notwithstanding, [Samuelson et al., 1994,
2323] do make an interesting point about the appropriateness of copyright for
legal protection of programs:
Program text is, thus, like steel and plastic, a medium in which other
works can be created. A device built in the medium of steel or plastic,
if sufficiently novel, is patentable; an original sculpture built of steel
or plastic is copyrightable. In these cases, we understand quite well
that the medium in which the work is made does not determine the
character of the creation. The same principle applies to software. The
legal character of a work created in the medium of software should

no more be determined by the medium in which it was created than
would be a work made of steel or plastic. In this respect, it makes
no more sense to talk about copyrighting programs than to talk about
copyrighting plastic or steel; it confuses the medium of creation and
the artifact created. In the case of software, the artifact created is
some form of innovative behavior, whether utilitarian or fanciful.
This suggests that patent is more appropriate than copyright. But what about a
program whose sole purpose is to produce an image on a screen, like a painting
or a film? Wouldnt copyright be more appropriate then? Still, the point remains
that its the behavior of the program (when executed by a machine! that is more
important than the text.
15. We suggest that programs should be viewed as virtual machines

and that this has interesting consequences for the proper form of
protection. [Samuelson et al., 1994, 2324]
This is a newly introduced term that they will have to go on to explain. But
its introduction is evidence that the authors have been ignoring the CPU that
inteprets and executes the program. For our definition of virtual machine and
more comments on this viewpoint, see 1 above.
16. Summary of [Samuelson et al., 1994]:
The predominantly functional nature of program behavior and other

industrial design aspects of programs precludes copyright protextion,
while the incremental nature of innovation in software largely
precludes patent protection. (p. 2333)
The argument about patents is this [Samuelson et al., 1994, 2346]:
(a) Patent law requires an inventive advance ofver the prior art
(b) But the innovations in functional program behavior, user interfaces, and
the industrial design of programs. . . are typically of an incremental sort.
A similar argument applies to chip design; as a result, Congress passed the

Semiconductor Chip Protection Act of 1984. The point is that a form of
legal protection other than patent and copyright was needed, recognized, and
implemented.
Although this is their principal argument, they also offer a separate argument
against patenting software [Samuelson et al., 1994, 2345]:
(a) Patents are given for methods of achieving results,

(b) Patents are not given for results themselves.
(c) It is. . . possible to produce functionally indistinguishable program
behaviors through use of more than one method.
(d) a patent could be given for one method of producing a result, but that
would not prevent the use of another method.
(e) If it is the result, not the method, that is the principal source of value of
a program, then the patent on the one method would not protect the result
produced by the other method
Their argument against the appropriateness of copyright law boils down to this
[Samuelson et al., 1994, 2350]:
(a) computer programs are machines whose medium of construction is text
(b) Copyright law does not protect the behavior of physical machines (nor
their internal construction)
That is, program behavior. . . is unpretectable by copyright law on account of its

functionality (p. 2351).
17. One point to make clear early in this chapter is that Im concerned more with the
ontology of software than I am with legal or economic issues in themselves.
18. On Koepsell 1995:

(a) The issue of what software is and whether it should be copyrightable or
patentable is investigated by Koepsell 1995 under the heading of the
ontology of cyberspace. Recall from 2.8 that ontology (as a branch of
philosophy) deals with issues about the nature and properties of things that
exist (or even things that dont exist). So, the ontology of cyberspace asks
questions like these:
What is cyberspace? Is it or does it have dimension? Are there
things in cyberspace? Are things in cyberspace properly called
objects? Are such objects or is cyberspace itself substance(s) or
process(es)? Is cyberspace or the objects in it real or ideal? What
is the categorical scheme of cyberspace? How should cyberspace
fit into a broader categorical scheme? (p. 2)
(b) INCORPORATE MY COMMENTS (Rapaport 1995 from Koepsell &
Rapaport 1995) INTO THIS CHAPTER IN APPROPRIATE PLACES.
19. Letter to Vicky Traube:

I think you might be in a perfect position to answer what I hope is a
relatively simple question.
During my retirement, Ive been working on a textbook on the
philosophy of computer science, and am currently working on a
chapter on the nature of software (computer programs) and whether
they can or should be copyrighted or patented.
Some people argue that programs are like the scripts of plays;
because plays can be copyrighted (and not patented), so should
programs. Others argue that programs are more like machines;
because machines can be patented (but not copyrighted), so should

programs. Finally, others argue that programs are unlike play scripts
(because its not just the text that mattersits what they can do
and, anyway, most people cant read the text of a program, as
they can for a play) *and* that they are unlike machines (because a
computer without a program cant do anything); hence, there should
be some new kind of legal protection mechanism.
One more piece of background before I get to my question:
Presumably, if I go to a bookstore and buy a copy of the script for,
say, Oklahoma, Im not liable for copyright infringement if I read it,
or even if I act it out with some friends (as long as I dont charge
admission, right?). Similarly, if I buy a DVD of the movie version,
Im not liable for copyright infringement if I watch it, even if I watch
it with some friends (again, as long as I dont charge admission?).
But, if I decide to mount a production of it for my local little theater
group, Id have to get permission (from you?).
Now, my question: What is the legal basis for that permission
requirement? Is it the copyright law? Or is it some other legal
protection mechanism?
Thanks for any clarification you can provide!
And her reply:
Terrific question. Two things. First of all, under the copyright law,
permission is required and royalties may be charged for a public
performance even if no admission is charged. If you read a play
with friends in your living room, that is probably not public, but
if you do it anywhere else and people come to see it, even just
people you know, with no admission charge, I consider that a public
performance, requiring authors permission. The legal basis for the
permission requirement is in the Copyright Act (Section 106), which
gives the Copyright Owner the exclusive right to do and authorize
any of the following: . . . (4) in the case of literary, musical, dramatic
and choreographic works, pantomines, and motion pictures and other
audiovisual works, to perform the copyrighted work publicly.
Section 117 relates to Limitations on exclusive rights; Computer
programs.
I hope this is helpful.
117 is online at http://www.law.cornell.edu/uscode/text/17/117.
20. Abstract ideas are not patentable subject matter. In order to prevent
abstract ideas from being indirectly patented, there are exceptions
in the law that will not allow patenting an invention made of
printed text based on the content of the text. The implication of
Godel. . . [numbering] is that the symbols that represent an abstract
idea need not be printed text. The symbols can be something abstract
like exponents of prime numbers. They can be something physical
like electromagnetic signals or photonic signals. . . . Does the law
exempt abstract ideas from patenting only when they are written in
text or does it exempt all physical representations? [PolR, 2009,
Symbols and Reducibility]
This seems to be another version of the paradox that this chapter opened with.
Heres another:
Now consider the Church-Turing thesis that computer programs
are the same thing as recursive functions. . . . Would this be of
consequence to the patentability of software? Developing a computer
program that fulfills a specific purpose is mathematically the same
thing as finding a number that has some specific properties. Can you
patent the discovery of a number?
Another way to look at it is that there is a list of all the possible
programs. We can make one by listing all possible programs in
alphabetic order. Any program that one may write is sitting in this list
waiting to be written. In such a case, could a program be considered
an invention in the sense of patent law? Or is it a discovery because
the program is a preexisting mathematical abstraction that is merely
discovered?
This idea that programs are enumerable numbers is supported by
the hardware architecture of computers. Programs are bits in memory.
Strings of 0s and 1s are numbers written in binary notations. Every
program reads as a number when the bits are read in this manner.
Programming a computer amounts to discovering a number that suits
the programmers purpose. [PolR, 2009, Enumerability]
21. Related to an issue that [Samuelson et al., 1994] raised (see above, somewhere),
[PolR, 2009, A Universal Algorithm] notes that if
I am sued for infringement on a software patent[,] I may try
defending myself arguing I did not implement the method that is
covered by the patent. I implemented the lambda-calculus algorithm
which is not covered by the patent. And even if it did, lambda-
calculus is old. There is prior art. What happens to software patents
then?
Anyone that argues in favor of method patents on software must
be aware that the current state of technology permits this possibility.
22. On [Samuelson et al., 1994]s suggestion that it is the virtual machine that should
be legally protected (whether by copyright, patent, or something sui generis),
compare this:
For example consider this sentence from In re Alappat:[4 ]
4 See http://digital-law-online.info/cases/31PQ2D1545.htm
We have held that such programming creates a new

machine, because a general purpose computer in effect
becomes a special purpose computer once it is programmed
to perform particular functions pursuant to instructions from
program software.
In a single sentence the court tosses out the fundamental principle
that makes it possible to build and sell digital computers. You dont
need to create a new machine every time you perform a different
computation; a single machine has the capability to perform all
computations. This is what universal Turing machines are doing.
But the new machine mentioned in the ruling is a virtual machine.

23. Its one thing to wonder whether software in the form of a computer program
is legally protected (whether by copyright, patent, or something else). But what
about the programming language that the program is written in? On the one
hand, a programming language can itself be viewed as a form of software (its
surely not hardware). On the other hand, while an author of a novel written in
English can copyright that novel, English itself cannot be copyrighted. (But what
about created languages, like Esperanto or Klingon?)
In a 2012 legal case, Oracle sued Google for Googles use of the Java
programming language, which Oracle claimed to own the copyright on. Googles
defense was that programming languages are not copyrightable. The first
decision was in favor of Google, but, on appeal, Oracle seems to have one;
Google is appealing that decision.
Unfortunately, the issues arent as clear cut as they might be, because its
not so much the Java language that is in question, as its APIs (application
programming interfaces), which may, or may not, be considered to be part of
the language itself (as opposed to software written in the language). And, to
further the complications, Google apparently used an open-source (that is, not
copyright-protected) version of the Java APIs. (The original briefs by Oracle and
by Google can be found at [Oracle America, 2012] and [Google, 2012]. Good
overviews are [Macari, 2012], [McAllister, 2012], and [McSherry, 2014]; see
also the Wikipedia article at
[http://en.wikipedia.org/wiki/Oracle v. Google].)

Bender, David (1985-1986), Protection of Computer Programs: The
Copyright/Trade Secret Interface, University of Pittsburgh Law Review 47:
907958.
Boyle, James (2009), What Intellectual Property Law Should Learn from
Software, Communications of the ACM 52(9) (September): 7176.
Crichton, Michael (2006), This Essay Breaks the Law, New York Times
(19 March): WK13.
Accessed 24 September 2014 from:
http://www.nytimes.com/2006/03/19/opinion/19crichton.html
Written in a sarcastic manner by a well-known novelist, this essay presents A
case for reforming the way we grant patents, on the grounds that natural
laws should not be patentable. For a follow-up article, see LabCorp. v.
Metabolite, accessed 24 September 2014 from:
http://www.oyez.org/cases/2000-2009/2005/2005 04 607
Editorial (2009, 8 November), Quick, Patent It!, New York Times
http://www.nytimes.com/2009/11/08/opinion/08sun3.html
Forester, Tom; & Morrison, Perry (1994), Computer Ethics: Cautionary Tales and
Ethical Dilemmas in Computing; Second Edition (Cambridge, MA: MIT Press).
Three sections on pp. 5768 from Ch. 3 survey the copyright-patent issues in
software.
Johnson, Deborah G. (2001), Property Rights in Computer Software, Ch. 6,
pp. 137167, of her Computer Ethics (Upper Saddle River, NJ: Prentice Hall).
A textbook discussion of copyright, patent, and trade-secret issues
surrounding software.
Klemens, Ben (2006), Private Numbers, Chronicle [of Higher Education] Review
52(22) (3 February): B2
Excerpted from:
Klemens, Ben (2006), Math You Cant Use: Patents, Copyright, and Software
(Brookings Institution Press)
http://chronicle.com/article/Collective-Longing-Private/4987/
The Klemens passage begins about halfway down the page, beginning with
the sentence, An entirely new economic arrangement has appeared in
mathematics and its offspring, computer science.
Menand, Louis (2014), Crooner in Rights Spat: Are Copyrifght Laws Too Strict?,
New Yorker (20 October): 8489.
http://www.newyorker.com/magazine/2014/10/20/crooner-rights-spat
A survey of the history of copyright laws, especially as they apply to works of
art.
Petroski, Henry (2008), Twists, Tags and Ties, American Scientist 96(3) (May-
June): 188192.
On the role of implementation: As with any well-written patent, in the one
granted to McHenry he is careful not to restrict his invention to only one
embodiment. . . (p. 192).
Samuelson, Pamela (1988), Is Copyright Law Steering the Right Course?,
IEEE Software (September): 7886.
An excellent overview of the copyright-vs.-patent dispute, arguing in favor of
patent protection for software.
Samuelson, Pamela (1991), Digital Media and the Law, Communications of the
ACM 34(10) (October): 2328.
Six characteristics of digital media seem likely to bring about significant
changes in the law: ease of replication, ease of transmission and multiple
use, plasticity (related to Moors 12.3.1 notion of softwares changeability),
equivalence of works in digital form (referring to the fact that, when digitized,
different kinds of copyrightable works all get implemented in the same
medium), compactness (digitized works take up little space, hence are more
prone to theft), and nonlinearity (due to the availability of hyperlinks, for
instance).
Samuelson, Pamela (2003), Unsolicited Communications as Trespass?,
http://people.ischool.berkeley.edu/pam/papers/acm vol46 p15.pdf
Abstract: Attempting to stretch existing laws to address previously
unforeseen technological issues.
Samuelson, Pamela (2007), Software Patents and the Metaphysics of
Section 271(f), Communications of the ACM 50(6) (June): 1519.
programs/faculty/facultyPubsPDF.php?facID=346&pubID=187
Microsoft argues that neither the intangible sequence of ones and zeros of
the object code, nor the master disks onto which the ogject code has been
loaded, should be considered a component of a patented invention. . . . Only
when object code has actually been installed on a. . . computer does it become
a physical component of a physica device. . . (pp.1516).
Samuelson, Pamela (2007), Does Copyright Law Need to Be Reformed?,
Samuelson, Pamela (2008), Revisiting Patentable Subject Matter,
Is everything under the sun made by humans patentable subject matter?
Samuelson, Pamela (2013), Is Software Patentable?, Communications of the
ACM 56(11) (November): 2325.
programs/faculty/facultyPubsPDF.php?facID=346&pubID=257
Samuelson, Pamela (2015), Copyrightability of Java APIs Revisited,

Communications of the ACM 58(3) (March): 2224,
http://radar.oreilly.com/2014/11/copyrightability-of-java-apis-revisited.html
Shaw, Alex; Li, Vivian; & Olson, Kristina R. (2012), Children Apply Principles
of Physical Ownership to Ideas, Cognitive Science 36(8) (November-December):
13831403.
https://depts.washington.edu/uwkids/Shaw.Li.Olson.2012.pdf
Adults apply ownership not only to objects but also to ideas. But do people
come to apply principles of ownership to ideas because of being taught about
intellectual property and copyrights? . . . like adults, children as young as 6
years old apply rules from ownership not only to objects but to ideas as well.
Sprigman, Christopher Jon (2015), Oracle v. Google: A High-Stakes Legal Fight
for the Software Industry, Communications of the ACM 58(5) (May): 2729.
Thatcher, Matt E.; & Pingry, David E. (2007), Software Patents: The Good, the
Bad, and the Messy, Communications of the ACM 50(10) (October): 4752.
For commentary, see:
Oldehoeft, Rod; Montague, John; Thatcher, Matt E.; & Pingry, David E.
(2007), Patented Algorithms Are Bad, Copyrighted Software Is Good
(letters to the editor), Communications of the ACM 50(12) (December): 9
10.
2. Websites:
Time to Abolish Software Patents?, Slashdot (29 February 2008),
http://yro.slashdot.org/story/08/02/29/0344258/time-to-abolish-software-patents
A blog, with some useful links.
Edwards, David A., Platonism Is the Law of the Land (earlier title: Why
Shouldnt Laws of Nature Be Patentable?), accessed 8 September 2014 from:
http://www.math.uga.edu/davide/Platonism is the Law of the Land.pdf
PolR (2009), An Explanation of Computation Theory for Lawyers, Groklaw
(11 November),
http://www.groklaw.net/article.php?story=20091111151305785,
http://www.groklaw.net/staticpages/index.php?page=20091110152507492,
and
http://www.groklaw.net/pdf2/ComputationalTheoryforLawyers.pdf (accessed
16 December 2013).
Computers dont work the way some legal documents and court precedents
say they do.
The phrase effective method is a term of art [in mathematics and
philosophy]. This term has nothing to do with the legal meaning of effective
and method. The fact that these two words also have a meaning in patent law
is a coincidence.
Sloman, Aaron (2008), Why Virtual Machines Really Matterfor Several
Disciplines, lecture slides accessed 27 October 2014 from:
http://www.cs.bham.ac.uk/research/projects/cogaff/talks/information.pdf
Views virtual machines as mathematical abstractions that can have causal

relations with the physical world.
Touretzky, David (2008), Gallery of CSS Descramblers,
http://www.cs.cmu.edu/dst/DeCSS/Gallery/
If code that can be directly compiled and executed may be suppressed
under the DMCA [[Digital Millenium Copyright Act], as Judge [Lewis A.]
Kaplan asserts in his preliminary ruling, but a textual description of the
same algorithm may not be suppressed, then where exactly should the line
be drawn? This web site was created to explore this issue, and point out
the absurdity of Judge Kaplans position that source code can be legally
differentiated from other forms of written expression. For example, it used to
offer a copy of otherwise suppressed code on a T-shirt!
FUNCTION (I/O PAIRS)_____
/ | \____
_________________________/ |implemented by \
/ | \
serial alg 1 serial alg 2 parallel alg 3
/ \ | | / \
/ \ | | / \
/ \ | | / \
serial alg 1 serial alg 1 | | parallel alg 3 parallel alg 3
expressed in expressed in ... ... expressed in expressed in
parallel serial serial parallel
proglang 1 proglang 2 proglang 3 proglang 4
____________/ | | \ \ | | /\
/ | | \ \ | | / \
/ / | \ \__________ | | / \
/ / | \ \ | | / \
/ _/ | \ \ | | / \
/ / | \ \ | | / \
/ / | \ \ | | | \
/ / | \ \ | | | |
/ / | \ \ ... ... | |
virtual actual serial alg 1 ser.alg 1 ser.alg 1 parallel parallel
machine: computer in serial in serial in serial alg 3 in alg 3 in
theoretical w/ serial proglang 2 proglang 2 proglang 2 proglang 4 proglang 4
model of proglang 2 compiled compiled compiled compiled compiled
computation as its into serial into into into into
Figure 13.3:
for serial machine mach.lang 1 serial parallel serial parallel
13.8. FURTHER SOURCES OF INFORMATION
proglang 2 lang mach.lang 2 mach.lang 3 mach.lang 5 mach.lang 6
-- -- -- ETC. -- -- --
WHICH (IF ANY) OF THESE SHOULD BE COPYRIGHTABLE?
WHICH (IF ANY) OF THESE SHOULD BE PATENTABLE?
Copyright 2010 by William J. Rapaport (rapaport@buffalo.edu)

http://www.cse.buffalo.edu/~rapaport/584/S10/cvspat.html-20100319
535
Chapter 14
What Is Implementation?

I wish to God these calculations had been executed by steam!
Charles Babbage (1821), quoted in [Swade, 1993, 86].
What. . . [Howard H. Aiken, who built the Harvard Mark I computer] had in
mind. . . was the construction of an electromechanical machine, but the plan he
outlined was not restricted to any specific type of mechanism; it embraced a
broad coordination of components that could be resolved by various constructive
mediums. [Chase, 1980, 199, 226]
[W]hy wasnt Mark I an electronic device? Again, the answer is money. It was
going to take a lot of money. Thousands and thousands of parts! It was very clear
that this thing could be done with electronic parts, too, using the techniques of the
digital counters that had been made with vacuum tubes, just a few years before
I started, for counting cosmic rays. But what it comes down to is this: if [the]
Monroe [Calculating Machine Co.] had decided to pay the bill, this thing would
have been made out of mechanical parts. If RCA had been interested, it might
have been electronic. And it was made out of tabulating machine parts because
IBM was willing to pay the bill. Aiken, quoted in [Chase, 1980, 200]
Figure 14.1: 2015,

c Scott Adams Inc./Universal UClick
537
538 CHAPTER 14. WHAT IS IMPLEMENTATION?

1. Required:
Chalmers, David J. (1993), A Computational Foundation for the Study of
Cognition, Journal of Cognitive Science (South Korea) 12(4) (October-December
2011): 323357.
(a) Accessed 3 November 2014 from:
(b) The 2011 version was accompanied by commentaries (including [Egan, 2012],
[Rescorla, 2012b], [Scheutz, 2012], and [Shagrir, 2012]) and a reply by
Chalmers [Chalmers, 2012].
(c) Read 12.
2 of this paper was published (in slightly different form) as On
Implementing a Computation, Minds and Machines 4 (1994): 391402.
(d) Skim the rest.
Rapaport, William J. (1999), Implementation Is Semantic Interpretation,
The Monist 82(1): 109130.
This article was based on a longer essay (which it alludes to as in
preparation), which has since been published as:
Rapaport, William J. (2005), Implementation Is Semantic Interpretation:
Further Thoughts, Journal of Experimental and Theoretical Artificial
Intelligence 17(4) (December): 385417.
Putnam, Hilary (1988), Appendix, Representation and Reality (Cambridge, MA:
MIT Press): 121125.
(a) Theorem. Every ordinary open system is a realization of every abstract finite
automaton.
(b) Putnams argument for this theorem is similar to [Searle, 1990]s argument
about the wall behind me implementing Wordstar, but it is much more detailed.
Chalmers, David J. (1993), Does a Rock Implement Every Finite-State
Automaton?, Synthese 108 (1996): 309333.
Accessed 9 November 2012 from http://consc.net/papers/rock.html
This is a reply to Putnams argument, above.
Chalmers corrects an error in my arguments in this paper in
[Chalmers, 2012, 236238].
(a) Suber, Peter (1997), Formal Systems and Machines: An Isomorphism
http://www.earlham.edu/peters/courses/logsys/machines.htm
(b) Suber, Peter (2002), Sample Formal System S
http://www.earlham.edu/peters/courses/logsys/sys-xmpl.htm
14.2 Introduction
One concept that has repeatedly cropped up in our discussions is that of
implementation:
In Chapter 3 (3.8), we saw that [Denning et al., 1989, 12] said that computer
science studies (among other things) the implementation of algorithmic
processes.
In Chapter 5 (5.5), we read that [Loui, 1987] said that computer science
is, among other things, the software- and hardware-implementation of
algorithms.
In Chapter 9 (9.3.5), we saw [Searle, 1990] saying that the wall behind
my back is right now implementing the Wordstar program, and we
talked about physical implementations of Turing machines and human
cognition. . . implemented by neuron firings.
In Chapter 10 (10.5.2), we discussed implement[ing] a plan [by] copy[ing] an
abstract design into reality.
In Chapter 12, we talked about programs implementing algorithms and the
idea of there being multiple implementations.
And in Chapter 13, we looked at how copyright only applies to
implementations of ideas in language, and we considered the possibility of
there being implementations of implementations.
In computer science, we talk about algorithms implementing functions, about
computer programs implementing algorithms, about data structures implementing
abstract data types, and about computers implementing Turing machines. So, what
is an implementation?
In this chapter, we will look at two theories about the nature of implementation.
One, due to David Chalmers, was designed to reply to [Searle, 1990]. The other will
illuminate the nature of the relation between syntax and semantics.
14.3 Chalmerss Theory of Implementation

David Chalmerss essay, A Computational Foundation for the Study of Cognition
[Chalmers, 2011] (published in a slightly shorter version as [Chalmers, 1995]; see also
[Chalmers, 1996b]), concerns both implementation and cognition. In this section, we
will focus only on what he has to say about implementation.
One of his claims is that we need a bridge between the abstract theory of
computation and physical systems that implement them. Other phrases that he
mentions as synonyms for implement are: realize (that is, make real) and are
described by.
(In all three cases, the entity that is doing the implementing, doing the realizing,
or being described is, typically, a physical (real) thing. And the entity that is being
implemented, or being realized, or doing the describing is, typically, an abstract thing
(or a linguistic description).)
That is, we need a theory of what implementation is. And one reason we need
such a theory is in order to refute [Searle, 1990]s claim that any given [physical]
system can be seen to implement any computation if interpreted appropriately
[Chalmers, 2011, I, my italics].
On the first version of Chalmerss theory of implementation, a physical system P
implements a computation C when the causal structure of P mirrors the formal
structure of C.
There are many aspects of this that need clarification:
What kind of physical system is P?
(It need not be a computer, according to Chalmers.)

What kind of computation is C?
(Is it merely an abstract algorithm? Is it more specifically a Turing-machine

program? Is it a program being executedwhat I called in Ch. 13 a process?
In any case, it would seem to be something that is more or less abstract, which
raises another question: What does it mean for something to be abstract?)
What does when mean?
(Is this intended to be just a sufficient condition, or a stronger biconditional

(when and only when)?)
What is a causal structure?
What is a formal structure?
And the most important question: What does mirror mean?

So, here is Chalmerss second version, which begins to answer some of these
questions:
P implements C if and only if:
1. the physical states of P can be grouped into state-types, and

2. there is a 11 map from the formal states of C to the physical state-types of P,
and
3. the formal state-types (that is, the ones that are related by an abstract state-
transition relation) map onto the physical state-types (that is, the ones that are
related by a corresponding causal state-transition relation).
Figure 14.2 (at the end of this chapter) might help to make this clear.
14.3. CHALMERSS THEORY OF IMPLEMENTATION 541
In this figure, the physical system P is represented by a set of dots, each of which
represents a physical state of P. These dots are partitioned into subsets of dots, that is,
subsets containing the physical states. Each subset represents a state-type, that is, a set
of states that are all of the same type. That takes care of part 1 of Chalmerss account.
The computation C is also represented by a set of dots. Here, each dot represents
one of Cs formal states. The arrows that point from the dots in C (that is, from Cs
formal states) to the subsets of dots in P (that is, to the state-types in P) represent the
11 map from C to P. To make it a 11 map, each formal state of C must map to a
distinct physical state-type of P.1 That takes care of part 2 of Chalmerss account.
The arrows in set C represent the abstract state-transition relation between Cs
formal states. And the arrows in set P between Ps subsets represent the causal
state-transition relation between Ps state-types. Finally, because Cs formal states are
mapped onto Ps physical state-types, the 11 map is a bijection. Chalmers also says
that Cs abstract state-transition relations correspond to Ps causal state-transition
relations. I take it that this means that the 11 map is a homomorphism (that is, a
structure-preserving map).2 Because the map is also onto, it is an isomorphism.
(An isomorphism is a structure- or shape-preserving bijection.) So, P and C have the
same structure. That takes care of part 3 of Chalmerss account.
The only other clarifications that we need to make are that a state is formal if its
part of the abstractthat is, non-physicalcomputation C, and a state is causal if
its part of the physical system P: Here, formal just means abstract, and causal
just means physical.
We can then say that a physical system (or a physical process?) P implements
an abstract computation C (which might be a Turing machine, or a less-powerful
finite-state automaton,3 or a combinatorial-state automaton4 ) if and only if there
is a reliably causal5 isomorphism f : C P. Such an f is a relation between
an abstract, computational model and something in the real, physical, causal world.
This f is 11 and onto6 a structure-preserving homomorphism such that the abstract,
input-output and processing relations in C correspond to reliably causal processes in P.
It follows from this analysis that:
Every physical system implements some computation. That is, for every physical
system P, there is some computation C such that P implements C.
But not every physical system implements any given computation. That is, it is
1 More precisely, a function from set A to set B is 11 = 0
de f for any two members a, a of A, and for any
member b of B, if f (a) = f (a0 ), then a = a0 . In other words, a function is 11 if, whenever two of its outputs
are the same, then their inputs must have been the same. Yet another way to put it is this: A function is 11
if no two (distinct) objects in its domain have the same image in its range.
2 f is a homomorphism =
de f f (R(c1 , . . . , cn )) = f (R)( f (c1 ), . . . , f (cn )). That is, if the ci are related by
some relation R, and if that relationship is mapped by f , then the image of (R(c1 , . . . , cn ) will be the image
of R applied to the images of the ci . Thats what it means to preserve structure.
3 EXPLAIN WHAT THIS IS AND WHY ITS LESS POWERFUL
4 EXPLAIN
5 EXPLAIN
6 Really? Even if C is abstract? Typically, if C is abstract and P is physical, P will have more parts (will
be more detailed) than C. But if f is a bijection, then every part of P must map to some part of C, which
seems to violate the abstract-physical distinction. Perhaps, however, the parts of P that are due merely to its
being physical (that is, the implementation-dependent details) can be ignored.
not the case that, for every P and for every C, P implements C. That is, there
is some P and there is some C such that P does not implement C. For example,
it is highly unlikely that the wall behind me implements Wordstar, because the
computation is too complex.
A given physical system can implement more than one computation. That is, for
any P, there might be two different computations C1 6= C2 such that P implements
C1 and P implements C2 . For example, my computer, right this minute as I type
this, is implementing the VI text-processing program, a clock, Powerpoint, and
several other computer programs.
Consider the first of these consequencesnamely, that every P implements some C.

Does that make the notion of computation vacuous? No, because the fact that some
P implements some C is not necessarily the reason why P is a P. (But, in the case of
cognition, it might be!)
14.4 Implementation as Semantic Interpretation

According to Chalmers, implementation is an isomorphism between an abstract
specification and a concrete, physical process. But computer scientists also use the term
implementation to refer to the relation between one abstract data type (for example, a
stack) and its implementation or representation by another abstract data type (for
example, a linked list).
Let me a bit more precise. Some programming languages, such as Lisp, do not have
stacks as a built-in data structure. So, a programmer who wants to write a program that
requires the use of stacks must find a substitute. In Lisp, whose only built-in data
structure is a linked list, the stack would have to be built out of a linked list: Stacks in
Lisp can be implemented by linked lists. Heres how:
First, a stack is a particular kind of data structure, often thought of as consisting
of a set of items structured like a stack of cafeteria trays: New items are added to the
stack by pushing them on top, and items can be removed from the stack only by
popping them from the top. Thus, to define a stack, one needs (i) a way of referring
to its top and (ii) operations for pushing new items onto the top and for popping items
off the top. That, more or less (mostly less, since this is informal), is a stack defined as
an ADT.
Second, a linked list (list, for short) is itself an abstract data type. It is a sequence
of items whose three basic operations are:
1. first(l), which returns the first element on the list l,
2. rest(l), which returns a list consisting of all the original items except the first,
and
3. make-list(i, l) (or cons(i, l)), which recursively constructs a list by putting item i
at the beginning of list l.
14.4. IMPLEMENTATION AS SEMANTIC INTERPRETATION 543
Finally, a stack s can be implemented as a list l, where top(s) := first(l),

push(s, i) := make-list(l, i), and pop(s) returns top(s) and redefines the list to be
rest(l).
The main point is that not all examples of implementation concern the
implementation of something abstract by something concrete. As we have just seen,
sometimes one abstract thing can implement another abstract thing. What we need is a
more general notion, namely, semantic interpretation.
So, what is semantic interpretation? Consider a formal system (also sometimes
called a symbol system, or a theoryunderstood as a set of sentencesor a
formal language. Such a system consists of:
1. primitive (or atomic) symbols (sometimes called tokens or markers, to be

thought of as the playing pieces in a board game like Monopoly)
(a) These are assumed to have no interpretation, no meaning, hence the term
marker. The racecar token in Monopoly isnt interpreted as a racecar
in the game; its just a token that happens to be racecar shaped, so as to
distinguish it from the token that is top-hat shaped. (And the top-hat token
isnt interpreted as a top hat in the game: Even if you think that it makes
sense for a racecar to travel around the Monopoly board, it makes no sense
for a top hat to do so!)
(b) Examples of such atomic symbols include the letters of an alphabet, or
(some of) the vocabulary of a language, or (possibly) neuron firings, or
even states of a computation.
2. (recursive) rules for forming new (complex, or molecular) symbols, sometimes

called well-formed formulas (wffs)that is, grammatically correct formulas
from old symbols (that is, from previously formed symbols), beginning with
the atomic symbols as the basic building blocks.
(a) These rules might be spelling rules (if the atomic symbols are alphabet
letters), or grammar rules (if the atomic symbols are words), or bundles
of synchronous neuron firings (if the atomic symbols are single neuron
firings).
(b) If the molecular symbols are linear (1-dimensional), then they can be
thought of as strings (that is, sequences of atomic symbols), or words
(if the atomic symbols are letters), or sentences (if the atomic symbols
are words). If the molecular symbols are 2-dimensional, then they are
something called complexes.
3. a distinguished (that is, singled-out) subset of wffs
(a) These are usually called axioms.

(b) But having axioms is optional. If English is considered as a formal system
[Montague, 1970], it doesnt need axioms. But if geometry is considered
as a formal system, it usually has axioms.
4. recursive rules (called rules of inference or transformation rules) for forming

(proving) new wffs (called theorems) from old ones (usually, but not always,
beginning with the axioms).
The syntax of such a system is the study of the properties of the symbols of
the system and the relations among the symbols. Among these relations are the
grammatical relations, which specify which strings of symbols are well formed
(according to the rules of grammar), and the proof-theoretic (or logical) relations,
which specify which sequences of wffs are proofs of theorems (that is, which wffs are
derivable by means of the rules of inference).
Here is an analogy: Consider a new kind of monster/building-block toy system,
consisting of Lego blocks that can be used to construct Transformer monsters. (This
wouldnt be a very practical real toy, because things made out of Legos tend to fall
apart. Thats why this is a thought experiment, not a real one!) The basic Lego blocks
are the primitive symbols of this system. Transformer monsters that are built from
Legos are the wffs of the system. And the sequences of moves that transform the
monsters into trucks are the proofs of theorems.
Real examples of formal systems include propositional logic, first-order logic,
Douglas Hofstadters Mu system [Hofstadter, 1979], Subers S,7 and a system
discussed at http://www.cse.buffalo.edu/rapaport/Papers/book.ch2.pdf. (Parts of that
document will be included here as an appendix to this chapter.)
The important fact about a formal system and its syntax is that there is no mention
of truth, meaning, reference, or any other semantic notion.
The semantics of a formal system is the study of the relations between the
symbols of the system (on the one hand) and something else (on the other hand). The
something else might be what the symbols represent, or what they mean, or
what they refer to, or what they name, or what they describe. Or it might be
the world. If the formal system is a language, then semantics studies the relations
between, on the one hand, the words and sentences of the language and, on the other
hand, their meanings. If the formal system is a theory, then semantics studies the
relations between the symbols of the theory and the worldthe world that the theory
is a theory of .
Semantics, in general, requires three things:
1. a syntactic domain; call it SYNtypically, but not necessarily, a formal

system,
2. a semantic domain; call it SEMcharacterized by an ontology
An ontology is, roughly, a theory or description of the things in the

semantic domain. It can be understood as a (syntactic!) theory of the
semantic domain, in the sense that it specifies (a) the parts of the semantic
domain (its members, categories, etc.) and (b) their properties and
relationships (structural as well as inferential or logical). Such an ontology
can be called a model theory.
7 http://www.earlham.edu/peters/courses/logsys/sys-xmpl.htm
3. a semantic interpretation mapping from SYN to SEM. SEM is a model or

interpretation of SYN; SYN is a theory of SEM.
Here are several examples of semantic domains that are implementations of
syntactic domains:
algorithms are implemented by computer programs

(in language L)
computer programs are implemented by computational processes

(in language L) (on machine m)
abstract data types are implemented by data structures

(in language L)
musical scores are implemented by performances

(by musicians) So,
play scripts are implemented by performances

(by actors)
blueprints are implemented by buildings

(made of wood, bricks, etc.;
cf. The Three Little Pigs)
formal theories are implemented by (set-theoretic) models
it looks as if we can say that semantic domains implement syntactic domains. That
is, implementation is the relation of semantic interpretation.
Typically, implementations are physical realizations of abstractions. But
physical realizations are a special case, as shown by the example of a linked list that
implements a stack, both of which are abstract. So, what we really have is the following
relationship:
I is an implementation of abstraction A in medium M.
Implementation I could be either an abstraction itself or something concrete. By an

abstraction A, I have in mind a generalization of the notion of an abstract data type,
as long as it is kept in mind that one abstraction can implement another abstraction.
And the medium M could be either an abstraction or something concrete.
In fact, there could be a sequence of implementations (or a continuum; see
[Smith, 1987]): a stack can be implemented by a linked list, which, in turn, could be
implemented in Pascal, which, in turn, could be implemented (that is, compiled into)
some machine language L, which, in turn, could be implemented on my Mac computer.
But it could also be more than a mere sequence of implementations, because there could
be a tree of implementations: The very same linked list could be implemented in Java,
instead of Pascal, and the Java or Pascal program could be implemented in some other
machine language on some other kind of computer. ([Sloman, 1998, 2, p. 2] makes

the same point.)
We see this same phenomenon in another, though related, situation: For an
introductory course I once taught, I wrote a very simple Pascal program that added two
integers. This program was compiled (that is, implemented) in a very simple assembly
language designed for instructional purposes. That assembly language program was
written (that is, implemented) in a dialect of Pascal called MacPascal, which was
implemented in MacOS assembly language, which was implemented in the machine
language that was implemented on a Mac II computer. (A similar situation is described
in [Sloman, 1998, 2, p. 2].) An interesting question to think about is this: When two
integers are input to my original Pascal program, and their sum is output, where does
the actual addition take place? Is it my Pascal program that adds the two integers? Or is
it really the Mac II computer that adds them? Or is it one (or all?) of the intermediate
implementations?
Heres something else to think about: Consider a program written in a high-level
programming language. Suppose that the program has a data structure called a student
record, containing information about (that is, a representation of) a student (something
like the person record in the Bloom County cartoon at the end of this chapter.) For
instance, Lucys student record might look something like this:
(student-record:
(name Lucy)
(student-number 12345)
(major computer-science)
(gpa 3.99)
)
This is merely a piece of syntax: a sequence of symbols. You and I reading it might
think that it represents a student named Lucy whose student number is 12345, whose
major is computer science, and whose GPA is 3.99. But as far as the computer
(program) is concerned, this record might just as well have looked like this:
(SR:
(g123 n456)
(g124 12345)
(g125 c569)
(g126 3.99)
)
As long as the program knows how to input new data, how to compute with these
data, and how to output the results in a humanly readable form, it really doesnt matter
what the data look like to us. That is, as long as the relationships among the symbols
are well specified, it doesnt matteras far as computing is concernedhow those
symbols are related to other symbols that might be meaningful to us. That is why it is
syntax.
Now, there are at least two ways in which this piece of syntax could be
implemented. One implementation, of course, is Lucy herself in the real world: A
person named Lucy, with student number 12345, who is majoring in computer science,
and who has a GPA of 3.99. This Lucy is an implementation of that data structure; she
is also a semantic interpretation of it.
Another implementation is the way in which that data structure is actually
represented in the computers machine language. That is, when the program containing
that data structure is compiled into a particular computers machine language, that data
structure will be represented in some other data structure expressed in that machine
language. That will actually be another piece of syntax. And that machine-language
syntax will be an implementation of our original data structure.
But when that machine-language program is being executed by the computer,
some region of the computers memory will be allocated to that data structure (to the
computers representation of Lucy, if you will), which will probably be an array of 0s
and 1s, that is, of bits in memory. These bits will be yet another implementation of
the original data structure, as well as an implementation of the machine-language data
structure.
Now, here is a question to think about: What is the relation between the human
(Lucy herself) and this region of the computers memory? Does the memory location
simulate Lucy? (Do bits simulate students?) Does the memory location implement
Lucy? (Do bits implement students?) Also: The 0s and 1s in memory can be thought
of as the 0s and 1s on a Turing-machine tape, and Lucy can be thought of as an
interpretation of that Turing-machine tape. Now, recall from 10.5.1 what Cleland said
about the difference between Turing-machine programs and mundane procedures: The
former can be understood independently of any real-world interpretation (that is, they
can be understood purely syntactically, to use the current terminology)understood
in the sense that we can describe the computation in purely 0/1 terms. (Can we?
Dont we at least have to interpret the 0s and 1s in terms of a machine-language
data structure, interpretable in turn in terms of a high-level programming-language data
structure, which is interpretable, in turn, in terms of the real-world Lucy?) Mundane
procedures, on the other hand, must be understood in terms of the real world (that is,
the causal world) that they are manipulating. (This is a good topic for class discussion!)
So, why do I think that implementation is semantic interpretation? The word
implement comes from a Latin word meaning to fill up, to complete, as in filling in
the details of an abstraction. And I have suggested the following:
Let I be something abstract or concrete.

Let A be an abstraction, that is, a generalization of the notion of an
abstract data type (as long as it is kept in mind that one abstraction can
implement another abstraction).
And let M be any abstract or concrete medium.
Then:
I is an implementation of A in medium M
means:
I is a semantic interpretation of syntactic domain A in semantic
domain M.
As another example of an abstract implementation, consider a top-down-design,

stepwise-refinement of a computer program (see 6.4.3): Each level (each refinement)
is an abstract implementation of the previous, higher-level one. A concrete
implementation would be an implementation in a physical medium.
Is implementation sui generis? That is, is it something that is not like anything
else? Or is it, in fact, something else, such as one of the following:
individuation:
the relation between the lowest level of a genus-species tree (such as dog
or human) and individual dogs or humans (for example, my cat Bella
individuates the species Felis catus).
instantiation:
the relation between a specific instance of something and the kind of thing that it
is (for example, the specific redness of my notebook cover is an instance of the
color red).
exemplification:
the relation between a (physical) object and a property that it has (for example,
Bertrand Russell exemplifies the property of being a philosopher).
reduction:
the relation between a higher-level object and the lower-level objects that it is
made of (for example, water is reducible to a molecule consisting of two atoms of
hydrogen and one atom of oxygen, or, perhaps, the emotion of anger is reducible
to a certain combination of neuron firings).
In [Rapaport, 1999] and [Rapaport, 2005b], I argue that each of these may be
implementations, but not vice versa, and that all of them are semantic interpretations.

1. On implementation as reduction:
[Goodman, 1987, 480] argues that the kind of reduction involved in reducing
ordered pairs to Wiener-Kuratowski doubletonsand, presumably, in reducing
natural numbers to sets of one kind or anotheris different from the kind of
reduction involved in reducing heat to mean molecular kinetic energy. There are
many ways of reducing ordered pairs or numbers to sets; there is only one way to
reduce heat to something physical. The former is multiply realizable; the latter
is not. Presumably, the reduction of algorithm to Turing machines is of the
former kind, given the equivalent reductions to recursive functions, etc.
Moreover, he argues, because knowledge of the non-reduced term is independent
of knowledge of the reducing term, such reductions are intensional.
2. On Lego-like objects, see the definition of constructive object in
http://www.encyclopediaofmath.org/index.php/Constructive object
3. On multiple realization:
The search for a single causative factorFactor Xis likely to be
successful if the event being studied has a fixed form and fixed
determinants. But the essence of migraine. . . lies in the variety of
forms it may take and the variety of circumstances in which it may
occur. Therefore, though one type of migraine may be associated
with Factor X, and another with Factor Y, it seems impossible, on
prima facie grounds, that all attacks of migraine could have the same
aetiology. [Sacks, 1999, 175176]
Sacks concludes that, to explain migraine, or any other phenomenon that exhibits
such multiple realizability, more than one universe of discourse is needed. At
the very least, there is, first, what I have called the Abstraction, or the functional
causal-network description; second, what I have called the Implementation, or
the anatomical-physiological description; and, third, there might also be what
Sacks calls the experiential (p. 176).
Also, there is evidence that nervous systems are multiply realized even among
animals; see [Pennisi, 2013].
Also, not only can all 16 truth-functional connectives be implemented
as (electronic) logic gates, they can also be implemented in bacteria!
[Benenson, 2013]
4. The Abstraction-Implementation distinction is mirrored in the occupant-role
distinction made in functionalist theories of the mind ([Lycan, 1990, 77]; on
functionalism, see 20.4). Hamlet is a role; Richard Burton (in the 1964
Broadway production of Hamlet) occupied that role. Alternatively, we could say
that Burton implemented Hamlet in the medium of human being (and a drawing
of Hamlet implements Hamlet in the medium of an animated cartoon version of
the play).
5. The ideas that Abstractions can implement other Abstractions and that there can
be continua of implementations is a consequence of what [Lycan, 1990, 78]
refers to as the relativity of implementation: He wants
to attack the dichotomies of software/hardware. . . . . . . my
objection is that software/hardware talk encourages the idea of
a bipartite Nature. . . as against reality, which is a multiple hierarchy
of levels of nature. . . . See Nature as hierarchically organized in
this way, and the function/structure distinction goes relative:
something is a role as opposed to an occupant, a functional state [that
is, an Abstraction, in my terms] as opposed to a realizer [that is, an
Implementation], or vice versa, only modulo a designated level of
nature.
6. The relationships among a chess game, a Civil War battle, and a computer
program whose input-output relationships can be interpreted as either of those
can be viewed as an implementation (chess game) that is abstracted (into a
computer program), which is then re-implemented as a Civil War battle. One
consequence of. . . [this] approach is that isomorphic models are equivalent. If
there is a one-to-one function from the domain of one model onto the domain of
another that preserves the relations of the model, then any sentence of the formal
language that is true in one model is true in the other. . . . [T]he best a formal
theory can do is to fix its interprettion up to isomorphism. Any model of the
theory can be changed into another model just by substituting elements of the
domain and renaming [Shapiro, 1996, 158159]. Here, models are just
7. Its surely worth mentioning somewhere that the problem of what
implementation is is closely related to the mind-body (or the mind-brain)
problem (very roughly, (how) is the mind implemented in the brain?). Also,
there is a wide literature in philosophy on realization (especially as related to
the mind-brain problem), which is only briefly touched on here and in the Further
Sources bibliography. For a nice discussion of this, see [Scheutz, 2001, 1]
8. Perhaps another great insight (see Ch. 7, 7.6)and one that Peter Denning
would probably applaud (Ch. 7, 35)is that the first three insights can be
implemented in matter. Alan Kay divides this insight into a triple whammy:
The core of computer science (Guzdials phrase). . .
. . . is all about the triple whammy of computing.
1 Matter can be made to remember, discriminate, decide and do
2 Matter can remember descriptions and interpret and act on them
3 Matte can hold and interpret and act on descriptions that describe
anything that matter can do.
He later suggests that the third item is the most powerful, followed
by the first, and then the second. He also suggests that issues about
the limits of computability and multiple realizability are implicit in these.
[Guzdial and Kay, 2010]
Note that the triple whammy is closely related to Pat Hayess notion of magic
paper (9.4).
9. [Rescorla, 2013, 1, p. 682] usefully dubs Chalmerss view of implementation
structuralism about computational implementation. It is the fact that the
structure of the physical system matches (mirrors, in Chalmerss terms; more
precisely, is isomorphic to) the structure of the computational system that
matters. Although Rescorla agrees that such structural identity is necessary for a
physical system to implement a computation, he denies that it is sufficient. That
is, although any physical system that implements a computation must have the
same structure as the computation, there are (according to Rescorla) physical
systems that have the same structure as certain computations but that are not
implementations of them (1, p. 683). This is because semantic relations to the
social environment sometimes help determine whether a physical system realizes
a computation (Abstract, p. 681). The key word here is sometimes: On
my position, the implementation conditions for some but not all computational
models essentially involve semantic properties (2, p. 684). Thus, Rescorla
disagrees both with those such as [Ladyman, 2009] who hold to the semantic
view of computational implementation (the view that if a physical system
implements a computation. . . [then] the system has representational properties)
and with those such as [Chalmers, 1995] and [Piccinini, 2006a] who hold to the
non-semantic view (the view that semantics [n]ever informs computation
(2, p. 683).
To argue for this, Rescorla must provide a counterexample to the non-semantic
view (1, p. 684). That is, he must provide an example of a physical
implementation that requires a representational feature (that requires semantics).
It is not enough to find an implementation that merely has a semantics; there are
plenty of those, because a semantic interpretation can always be given to one. So
here is one example that he gives, a Scheme program for Euclids algorithm for
computing GCDs (4, p. 686):
(define (gcd a b)
(if (= b 0)
a
(gcd b (remainder a b))))
This is a recursive algorithm that we can paraphrase in English as follows:

To compute the GCD of integers a and b, do the following:
If b = 0, then output a
else compute the GCD of b and the remainder of dividing a by b.8
8 Or,
if you prefer:
To compute the GCD of a and b, do:
If b = 0, then output a
else do:
Divide a by b; let r be the remainder;
Compute the GCD of b and r.
Rescorla points out that, To do that, the machine must represent numbers. Thus,
the Scheme program contains content-involving instructions. . . (4, p. 687).
A content-involving instruction is [one that] is specified, at least partly, in
semantic or representational terms (3, p. 685; he borrows this notion from
[Peacocke, 1995]). So, the Scheme program is specified in semantic terms
(specifically, it is specified in terms of integers). Therefore, if a physical system
is going to implement the program, that physical system must represent integers;
that is, it requires semantics. Hence, The Scheme program is a counter-example
to the non-semantic view of computational implementation (4, p. 687).
I can see that the machine does represent numbers (or can be interpreted as
representing them). But why does he say that it must represent them? I can even
see that for an agent to use such a physical computer to compute GCDs, the agent
must interpret the computer as representing numbers. But surely an agent could
use this computer, implementing this program, to print out interesting patterns
of uninterpreted symbols.
To respond to this kind of worry, Rescorla asks us to consider two copies of this
machine, calling them M10 and M13 . The former uses base-10 notation; the latter
uses base-13. When each is given the input pair (115, 20), each outputs 5.
But only the former computes the GCD of 115 and 20. (The latter was given
the integers 187 and 26 as inputs; but their GCD is 1.) So M10 implements the
program, but M13 does not; yet they are identical physical computers.
One possible response to this is that the semantics lies in the users interpretation
of the inputs and outputs, not in the physical machine. Thus, one could say
that the machines do both implement the program, but that it is the users
interpretation of the inputs, outputs, and that programs symbols that makes all
the difference. After all, consider the following Scheme program:
(define (MYSTERY a b)
(if (= b 0)
a
(MYSTERY b (remainder a b))))
If we are using base-10 notation, then we can interpret MYSTERY as GCD;

if we are using base-13 notation, then we might either be able to interpret
MYSTERY as some other numerical function or else not be able to interpret
it at all. In either case, our two computers both implement the MYSTERY
program.
One possible conclusion to draw from this is that any role that semantics has to
play is at the level of the abstract computation, not at the level of the physical
implementation.
Rescorlas response to this might be incorporated in these remarks:
The programs formal structure does not even begin to fix a unique
semantic interpretation. Implementing the program requires more
than instantiating a causal structure that mirrors relevant formal

structure. (4, p. 688).
I agree with the first sentence: We can interpret the MYSTERY program in
many ways. I disagree with the term requires in the second sentence: I would
say that implementing the program only requires instantiating the mirroring
causal structure. But I would go on to say that if one wanted to use the physical
implementation to compute GCDs, then one would, indeed, be required to do
something extra, namely, to provide a base-10 interpretation of the inputs and
outputs (and an interpretation of MYSTERY as GCD).
In fact, Rescorla agrees that the semantic interpretation of MYSTERY as GCD
is required: there is more to a program than meaningless signs. The signs have
an intended interpretation. . . (4, p. 689). But it is notoriously hard (some
would say impossible) to pin down what the intended interpretation of any
formal system is.
Further discussion of Rescorlas arguments would take us too far afield, but here
are some questions to consider: Are the inputs to the Euclidean GCD algorithm
numerals (like 10) or numbers (like 10 or 13)?9 What about the inputs to a
computer program written in Scheme that implements the Euclidean algorithm:
Are its inputs numerals or numbers? (It may help to consider this analogous
question: Is the input to a word-processing program the letter a or an electronic
signal or ASCII-code representing a?)
10. [Rescorla, 2012a, 12; italics in original] gives another example of semantic
computation, in the sense of a computation that requires numbers, not (merely)
numerals:
A register machine contains a set of memory locations, called

registers. A program governs the evolution of register states. The
program may individuate register states syntactically. For instance,
it may describe the machine as storing numerals in registers, and it
may dictate how to manipulate those syntactic items. Alternatively,
the program may individuate register states representationally.
Indeed, the first register machine in the published literature models
computation over natural numbers [[Shepherdson and Sturgis, 1963,
219]]. A program for this numerical register machine contains
instructions to execute elementary arithmetical operations, such
as add 1 or subtract 1. A physical system implements the
program only if can execute the relevant arithmetical operations.
A physical system executes arithmetical operations only if it bears
appropriate representational relations to numbers. Thus, a physical
system implements a numerical register machine program only if
it bears appropriate representational relations to numbers. Notably,
a numerical register machine program ignores how the physical
9 Note that the base-10 numeral 10 represents the number 10, but the base-13 numeral 10 represents
the number 13.

system represent numbers. It applies whether the systems numerical

notation is unary, binary, decimal, etc. The program characterizes
internal states representationally (e.g. a numeral that represents the
number 20 is stored in a certain memory location) rather than
syntactically (e.g. decimal numeral 20 is stored in a certain
memory location). It individuates computational states through
denotational relations to natural numbers. It contains mechanical
rules (e.g. add 1) that characterize computational states through their
numerical denotations.
I agree that this is a semantic computation. Note that it is not a Turing machine
(which would be a purley syntactic computation). And note that there cannot be
a physical numerical register machine, only a syntactic one. This is not because
there are no numbers, but because (if numbers do exist) they are not physical!
11. Re: Chalmers 1993:

Note that Chalmerss implementing function (2, p. 327) maps from the physical
system to the FSA (or CSA). Thus, it is a description function: the inverse of a
semantic interpretation function.
12. Rescorla seems to agree with me that implementation is semantic interpretation:
Physical system P realizes/implements computational model M just

in case [c]omputational model M accurately describes physical
system P. (2, p. 1278)
Concerning what I call implementation details, [Rescorla, 2014b, 2, p. 1280]

says: A physical system usually has many notable properties besides those
encoded by our computational model: colors, shapes, sizes, and so on.
However, a significant difference is that Rescorlas version is somewhat more
narrowly focused than mine. Where I view any form of semantic interpretation
as a form of implementation, Rescorla makes the notion more limited by
elaborating on the following idea (3, p. 1280): M accurately describes P just in
case P reliably moves through state space according to mechanical instructions
encoded by M. Putting these two statements together, we get this formulation:
Physical system P realizes/implements computational model M

just in case P reliably moves through state space according to
mechanical instructions encoded by M.
The difference in focus is due in part to the fact that Rescorla is more interested
in the special case of implementation of a computation, whereas I am more
interested in implementation more generally.
13. One place where Chalmers very clearly states that (what I call) an Abstraction is
a description of (what I call) an implementation is here:
Certainly, I think that when a physical system implements an a-

computation [that is, a computation abstractly conceived], the a-
computation can be seen as a description of it. [Chalmers, 2012, 215]
14. Somewhere in this chapter, must emphasize that abstraction is the opposite of
implementation. (For a backward reference in 17.4.) (For some discussion of
this, see Rapaport 1996, Ch.2, discussion of Rosenblueth & Wiener.)

Brown, Curtis (2004), Implementation and Indeterminacy, in J. Weckert & Y. Al-Saggaf
(eds.), Computing and Philosophy Conference, Canberra: Conferences in Research and
Practice in Information Technology, Vol. 37 (Australian Computer Society): 2731.
Accessed 23 October 2014 from:

http://digitalcommons.trinity.edu/cgi/viewcontent.cgi?article=1000&context=phil faculty
A critique of Chalmerss theory of implementation.
Dresner, Eli (2010), Measurement-Theoretic Representation and Computation-Theoretic

Realization, Journal of Philosophy 108(6) (June): 275292.
Examines the association between numbers and the physical world that is made
in measurement and argues that implementation and (measurement-theoretic)
representation are a single relation (or concept) viewed from different angles
(p. 276).
Eden, Amnon H.; & Turner, Raymond (2007), Problems in the Ontology of Computer
Programs, Applied Ontology 2(1): 1336.
Preprint accessed 23 October 2014 from:

http://www.eden-study.org/articles/2007/problems-ontology-programs ao.pdf
4 discusses concretization. . . a process during which an entity or entities of
one category are synthesized (come into being) from entities of a more abstract
category.
Kramer, Jeff (2007), Is Abstraction the Key to Computing?, Communications of the

ACM 50(4) (April): 3642.

http://www.ics.uci.edu/andre/informatics223s2007/kramer.pdf
The role of abstraction (an inverse of implementation) is considered for its role in
computer education.
Ladyman, James (2009), What Does It Mean to Say that a Physical System Implements
a Computation?, Theoretical Computer Science 410(45) (17 February): 376383.
1. Accessed 28 October 2014 from:

main.pdf? tid=04248730-5f6c-11e4-8b44-
00000aab0f26&acdnat=1414588006 ff4ead0d50f1774e6d0779cf705e19e6
2. A logical transformation is a map from a set of logical states to a set of logical
states, whereas a physical process is a change in a physical system whereby it
goes from a particular physical state to a particular physical state. Hence, strictly
speaking a physical process cannot be said to implement a logical transformation
because all it could ever do is implement the part of the map that takes one of
the logical input states to another logical input state. . . . For a physical system to
implement a logical transformation there must be a family of processes and each
of the physical states that represent the logical input states must be taken by one
member of the family to the appropriate physical state, that is the one that represents
the right logical output state. (p 379)
Lycan, William G. (1987), The Continuity of Levels of Nature, in William G. Lycan

(ed.), Mind and Cognition: A Reader (Basil Blackwell): 7796; excerpted from Chs. 45
of William G. Lycan, Consciousness (Cambridge, MA: MIT Press, 1987).
. . . software/hardware talk encourages the idea of a bipartite Nature, divided
into two levels, roughly the physiochemical and the (supervenient) functional
or higher-organizationalas against reality, which is a multiple hierarchy of
levels of nature. . . . See Nature as hierarchically organized in this way, and the
function/structure distinction goes relative: something is a role as opposed to an
occupant, a functional state as opposed to a realizer, or vice versa, only modulo a
designated level of nature.
McMullin, Ernan (1985), Galilean Idealization, Studies in History and Philosophy of
Science 16(3): 247273.
Primarily an examination of the nature of idealization in the development
of scientific theories (and how or whether such idealizations falsify or
misrepresent reality), but also contains useful discussions of the differences
between scientific theories and scientific models (esp. pp. 257ff).
Popper, Karl (1978), Three Worlds,
http://tannerlectures.utah.edu/ documents/a-to-z/p/popper80.pdf
Poppers three worlds are the world of physical objects, the world of thinking
and subjective experiences, and the world of the products of the human mind,
including languages, scientific theories, and physical artifacts. Many world-3
objects are embodied or physically realized, in. . . world-1 physical objects.
Rescorla, Michael (2012), Are Computational Transitions Sensitive to Semantics?,
Australian Journal of Philosophy 90(4) (December): 703721.
1. Preprint accessed 30 October 2014 from:

http://www.philosophy.ucsb.edu/docs/faculty/papers/formal.pdf
2. Mathematical models of computation, such as the Turing machine, are abstract
entities. They do not exist in space or time, and they do not participate in causal
relations. Under suitable circumstances, a physical system implements or physically
realizes an abstract computational model. Some philosophers hold that a physical
system implements a computational model only if the system has semantic or
representational properties [. . . [Ladyman, 2009]]. Call this the semantic view of
computational implementation. In contrast, [[Chalmers, 1995], [Piccinini, 2006a]],
and others deny any essential tie between semantics and physical computation. I
agree with Chalmers and Piccinini. (2.1, p. 705)
Scherlis, William L.; & Scott, Dana S. (1983), First Steps towards Inferential
Programming, in R.E.A. Mason (ed.), Information Processing 83: 199212,
http://repository.cmu.edu/cgi/viewcontent.cgi?article=3542&context=compsci
2 discusses the relation between a specification the program that implements it.
Scheutz, Matthias (1998), Implementation: Computationalisms Weak Spot, Conceptus
Journal of Philosophy 31(79): 229239.
1. Accessed 21 October 2014 from:

http://hrilab.tufts.edu/publications/scheutz98conceptus.pdf
2. Analyzes Searles and Putnams arguments, concluding that a better notion of
impelmentation is. . . [needed] that avoids state-to-state correspondences between
physical systems and abstrat objects
Scheutz, Matthias (1999), When Physical Systems Realize Functions, Minds and
Machines 9: 161196.
1. Preprint accessed 22 October 2014 from:

http://hrilab.tufts.edu/publications/scheutz99mm.pdf
2. 3 is an especially good discussion of Putnams argument.
3. Scheutz approaches implementation by way of its inverse, abstraction (from a
physical system); see 7.
4. To refute Putnam 1988, Scheutz replaces the notion of implementation of a
computation with realization of a function. His view is not inconsistent with
the semantic theory of Rapaport 1999, at least when Scheutz says (7, p. 174):
what is the same [in the case of two realizations] is the syntactic structure of. . . the
function; that is, they are different semantic interpretations of the same syntactic
structure.
Scheutz, Matthias (2001), Computational versus Causal Complexity, Minds and

Machines 11: 543566.
Preprint accessed 23 October 2014 from:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.21.8293&represent=rep1
&type=pdf
2 critiques Chalmerss theory of implementation. The rest of the paper introduces
a new theory of implementation based on a notion of bisimulation.
Scheutz, Matthias (2012), What It Is Not to Implement a Computation: A Critical

Analysis of Chalmers Notion of Implementation, Journal of Cognitive Science
(Republic of Korea) 13(1) (January-March): 75106.

Argues that Chalmerss definition of implementation still allows for unwanted
implementations (p. 75)
Shagrir, Oron (2012) Computation, Implementation, Cognition, Minds and Machines

22(2) (Summer): 137148.
A critique of Putnam 1988, [Searle, 1990], [Chalmers, 1996b], [Scheutz, 2001],

[Piccinini, 2006a], arguing that there can be systems that simultaneously
implement different complex automata (p. 137).
Sloman, Aaron (2008), Why Virtual Machines Really Matterfor Several Disciplines,
lecture slides accessed 27 October 2014 from:
http://www.cs.bham.ac.uk/research/projects/cogaff/talks/information.pdf
Explores this definition of implementation (and its relationship to the concept of a

virtual machine): Phenomena of type X. . . are fully implemented in, or realised in,
or grounded in phenomena of type Y. . . if and only if:
(a) type X phenomena cannot exist without some entities and processes of type Y.
...
(b) certain entities and processes of type Y are sufficient for the phenomena of
type X to existthey constitute the implementation.
(slide 15)
Wilson, Robert A. (2001), Two Views of Realization, Philosophical Studies 104: 131.
Argues that realization is context sensitive.
Also cite Rescorla.

Figure 14.2: P is a physical system whose physical states (dots) are grouped into state
types (partitions of P) related by causal state-transtion relations; C is a computation
whose formal states (dots), which are related by abstract state-transition relations, are
mapped to Ps physical-state types by abstract state-transition relations.
Figure 14.3: 2010,

c Berkeley Breathed
Chapter 15

III. Are Computer Programs
Theories?
I havent formalized my theory of belief revision, but I have an algorithm

that does it.
Frances L. Johnson (personal communication, February 2004).
. . . within ten years most theories in psychology will take the form of
computer programs, or of qualitative statements about the characteristics
of computer programs.
[Simon and Newell, 1958, 78]
Figure 15.1: 2014,

c Universal Uclick
563
564 CHAPTER 15. ARE PROGRAMS THEORIES?

1. Required:
Simon, Herbert A. (1996), Understanding the Natural and Artificial Worlds,
Ch. 1, pp. 124 of Simons The Sciences of the Artificial, Third Edition (Cambridge,
MA: MIT Press).
Originally written in 1969; updated in 1981.
In this chapter, Simon also has interesting things to say about whether
computer science is a science.
Thagard, Paul (1984), Computer Programs as Psychological Theories, in
O. Neumaier (ed.), Mind, Language, and Society (Vienna: Conceptus-Studien):
7784.
A critique of the whole idea.
Weizenbaum, Joseph (1976), Computer Power and Human Reason: From

Judgment to Calculation (New York: W.H. Freeman).
Ch. 5 (Theories and Models), pp. 132153.
Ch. 6 (Computer Models in Psychology), pp. 154181.
Ch. 6 discusses computer programs as theories and the potential evils of
AI, as well as presenting Weizenbaums objections to Simon.
Moor, James H. (1978), Models and Theories, 4 of his Three Myths
of Computer Science, British Journal for the Philosophy of Science 29(3)
(September): 213222.
4. Highly Recommended:
Simon, Herbert A.; & Newell, Allen (1956), Models: Their Uses and Limitations,
in Leonard D. White (ed.), The State of the Social Sciences (Chicago: University of
Chicago Press): 6683.
Makes many of the same points as Simon 1996, but goes into a bit more detail
on the nature of theories and models.
5. Recommended:
Wilks, Yorick (1990?), One Small Head: Models and Theories, pp. 121134,
in 4 (Programs and Theories) of Derek Partridge & Yorick Wilks (eds.), The
Foundations of Artificial Intelligence: A Sourcebook (Cambridge, UK: Cambridge
University Press).
Published in slightly different form as:
Wilks, Yorick (1974), One Small HeadModels and Theories in
Linguistics, Foundations of Language 11(1) (January): 7795.
Has a useful, if sometimes confusing, overview of the many meanings of
theory and model.
Daubert v. Merrell Dow Pharmaceuticals (92-102), 509 U.S. 579 (1993)
http://openjurist.org/509/us/579 (accessed 13 November 2012).
A Supreme Court case concerning what counts as generally accepted

reliability by the scientific community. Has interesting observations on the
nature of scientific theories and expertise. My colleague Sargur N. Srihari
recommended this to me, after his experience being called as an expert witness
on handwriting analysis, on the grounds that his computer programs that could
recognize handwriting were scientific theories of handwriting analysis.
Green, Christopher D. (2004), Philosophical Approaches to Explanation and
Scientific Models, and Their Relations to Connectionist Cognitive Science, Ch. 3
of his (How) Do Connectionist Networks Model Cognition? (Ph.D. dissertation,
Department of Philosophy, University of Toronto).
Contains a useful survey of different views of scientific explanation and
scientific models, embedded in a discussion of connectionism.
15.2 Introduction
The issue raised in the quote that opens this chapter, from a former graduate student
at my university, is whether an algorithm, which is a pretty formal, precise thing,
is different from a formal theory. Some might say that her algorithm is her theory.
Does it really make sense to say that you dont have a formal theory of something if
you do have a formal algorithm that implements your theory? Roger Schank, an AI
researcher famous for taking a scruffythat is, non-formalapproach to AI used
formal algorithms to express his non-formal theories. That sounds paradoxical.
As part of our investigation into the nature of computer programs, we have
seen that algorithms are said to be implemented in computer programs. But, if
implementation is semantic interpretation, then computer programs are semantic
interpretations of algorithms, in the medium of some programming language. However,
some philosophers have argued that computer programs are theories; but theories are
more like abstractions than they are like implementations. And others have argued that
computer programs are simulations or models of real-world situations, which sounds
more like an implementation than an abstraction. In this chapter, we will look into the
relation of a program to that which it models or simulates, the nature of simulation, and
whether programs are (scientific) theories.
15.3 Three Distinctions

Lets begin by considering three distinctions, between simulations and emulations,
between mere simulations and the real thing, and between theories and
models.
15.3.1 Simulation vs. Emulation

There is no standard, agreed-upon definition of either simulation or emulation.
(This unfortunate situation occurs all too frequently. Therefore, it is always important
for you to try to find out how a person is using such terms before deciding whether to
agree with what they say about them.)
Following [Roth, 1983], lets say that
x simulates y means (roughly): y is a real or imagined system, and x is a

model of y, and we experiment with x in order to understand y.
This is only a rough definition, because I have not said what is meant by system,
model, or understand, not to mention real, imagined, or experiment! Typically,
a computer program (x) is said to simulate some real-world situation y when program
x is a model of situation y. If we want to understand the situation, we can do so
by experimenting with the program; perhaps the program is easier to deal with or to
manipulate than the real-world situation. In an extreme case, x simulates y if and only
if x and y have the same input-output behavior, but they might differ greatly in some of
the details of how they work.
And, following [Habib, 1983], lets say that
15.3. THREE DISTINCTIONS 567
x emulates y means (roughly)
either: x and y are computer systems, and x interprets and

executes ys instruction set by implementing ys operation
codes in xs hardwarethat is, hardware y is implemented as
a virtual machine on x
or: x is some software feature, and y is some hardware feature,

and x simulates y, doing what y does exactly as y does it.
In general, x emulates y if and only if x and y not only have the same input-output
behavior (x not only simulates y) but x also uses the same algorithms and data structures
as y.
It is unlikely that being a simulation and being an emulation are completely distinct
notions. More likely, they are the ends of a spectrum, in the middle of which are xs
and ys that differ in the level of detail of the algorithms and data structures that x
uses to do ys job. At the pure simulation end of the spectrum, only xs and ys
external, input-output behavior agree; at the pure emulation end, all of their internal
behavior also agree. Perhaps, then, the only pure example of emulation would be
y itself! Perhaps, even, there is no real distinction between simulation and emulation
except for the degree of faithfulness to what is being simulated or emulated (and except
for cases in which y is only an imagined situation, whereas x will always be something
real).
15.3.2 Simulation vs. the Real Thing

On the other hand, to say that x is merely a simulation of y is to suggest that y is real
but that x is not. The word simulation has a connotation of imitation or unreal.
For example, it is often argued that a simulation of a hurricane is not a real hurricane,
or that a simulation of digestion is not real digestion.
But there are cases where a simulation is the real thing. (Or would such simulations
be better called emulations?) For example, although a scale model of the Statue of
Liberty is not the real Statue of Liberty, a scale model of a scale model (of the Statue
of Liberty) is itself a scale model (of the Statue of Liberty). A Xerox or PDF or faxed
copy of a document is that document, even for legal purposes (although perhaps not for
historical purposes; see [Korsmeyer, 2012]). Some philosophers and computational
cognitive scientists have argued that a computational simulation of cognition really
is cognition [Edelman, 2008], [Rapaport, 2012b]. And, in general, a simulation of
information is that information.
15.3.3 Theories vs. Models

When people say that something is a theory in ordinary language, they often mean
that it isnt necessarily true, that it is mere speculation. But scientists and philosophers
use the word theory in a more technical sense. (For a humorous illustration of this,
see Fig. 15.1.)
This, by the way, is one reason that people who believe in the theory of evolution
and those who dont are often talking at cross purposes, with the former saying that
evolution is a true, scientific theory:
Referring to biological evolution as a theory for the purpose of contesting
it would be counterproductive, since scientists only grant the status
of theory to well-tested ideas. (Terry Holliday, Kentucky education
commissioner, 2011; cited in Science 337 (24 August 2012): 897.)
and the latter saying that, if it is only a theoryif, that is, it is a mere conjecturethen
it might not be true:
The theory of evolution is a theory, and essentially the theory of evolution
is not scienceDarwin made it up. (Kentucky state Representative Ben
Waide, 2011; cited in Science 337 (24 August 2012): 897.)
They are using the word in very different senses.
Further complicating the issue, there are at least two views within the philosophy
of science about what scientific theories are:
On the syntactic approach to theories (due to the Logical Positivists
[Uebel, 2012]), a theory is an abstract description of some situation (which usually
is, but need not be, a real-world situation) expressed in a formal language with an
axiomatic structure; that is, a theory is a formal system (see 14.4). Such a theory
is typically considered to be a set of sentences (linguistic expressions, well-formed
formulas) that describe a situation or that codify (scientific) laws about a situation.
(This is the main sense in which the theory of evolution is a theory.) Such a
description, of course, must be expressed in some language. Typically, the theory
is expressed in a formal, axiomatic language that is semantically interpreted by rules
linking the sentences to observable phenomena. These phenomena are either directly
observableeither by unaided vision or with the help of microscopes and telescopes
or are theoretical terms (such as electron) that are definable in terms of directly
observable phenomena (such as a vapor trail in a cloud chamber). (Well say more
about this in 15.4 CHECK THIS!.)
On the semantic approach to theories (due largely to the philosopher
Patrick Suppes [Frigg and Hartmann, 2012]), theories are the set-theoretic models
of an axiomatic formal system. Such models are isomorphic to the real-
world situation being modeled. (Weaker semantic views of theories see
them as state spaces (http://en.wikipedia.org/wiki/State space) or prototypes
(http://en.wikipedia.org/wiki/Prototype), which are merely similar to the real-world
situation.)
15.4 Computer Programs as Theories

Computational cognitive scientists such as P.N. Johnson-Laird, Allen Newell, Zenon
Pylyshyn, and Herbert Simon have all claimed that computer programs are theories, in
the sense that the programming languages in which they are written are languages for
theories, that the programs are ways to express theoriest This informal way of putting
15.4. COMPUTER PROGRAMS AS THEORIES 569
the idea sounds circular, so lets see if we can be a bit more precise. First, consider
these passages from their writings (my italics throughout):
[Simon and Newell, 1962, 97]:

1. Computers are quite general symbol-manipulating devices that can
be programmed to perform nonnumerical as well as numerical
symbol manipulation.
2. Computer programs can be written that use nonnumerical symbol
manipulating processes to perform tasks which, in humans, require
thinking and learning.
3. These programs can be regarded as theories, in a completely literal
sense, of the corresponding human processes. These theories are
testable in a number of ways: among them, by comparing the
symbolic behavior of a computer so programmed with the symbolic
behavior of a human subject when both are performing the same
problem-solving or thinking tasks.
[Johnson-Laird, 1981, 185186]:

Computer programming is too useful to cognitive science to be left
solely in the hands of the artificial intelligenzia [sic]. There is a
well established list of advantages that programs bring to a theorist:
they concentrate the mind marvelously; they transform mysticism into
information processing, forcing the theorist to make intuitions explicit and
to translate vague terminology into concrete proposals; they provide a
secure test of the consistency of a theory and thereby allow complicated
interactive components to be safely assembled; they are working models
whose behavior can be directly compared with human performance. Yet,
many research workers look on the idea of developing their theories
in the form of computer programs with considerable suspicion. The
reason. . . [i]n part. . . derives from the fact that any large-scale program
intended to model cognition inevitably incorporates components that lack
psychological plausibility. . . . The remedy. . . is not to abandon computer
programs, but to make a clear distinction between a program and the
theory that it is intended to model. For a cognitive scientist, the single
most important virtue of programming should come. . . from the business
of developing [the program]. Indeed, the aim should be neither to simulate
human behavior. . . nor to exercise artificial intelligence, but to force the
theorist to think again.
[Pylyshyn, 1984, 76]:

[T]he. . . requirementthat we be able to implement [a cognitive] process
in terms of an actual, running program that exhibits tokens of the
behaviors in question, under the appropriate circumstanceshas far-
reaching consequences. One of the clearest advantages of expressing a
cognitive-process model in the form of a computer program is, it provides

a remarkable intellectual prosthetic for dealing with complexity and for
exploring both the entailments of a large set of proposed principles and
their interactions.
[Johnson-Laird, 1988, 52]:
[T]heories of mind should be expressed in a form that can be modelled
in a computer program. A theory may fail to satisfy this criterion for
several reasons: it may be radically incomplete; it may rely on a process
that is not computable; it may be inconsistent, incoherent, or, like a
mystical doctrine, take so much for granted that it is understood only by it
adherents. These flaws are not always so obvious. Students of the mind do
not always know that they do not know what they are talking about. The
surest way to find out is to try to devise a computer program that models
the theory.
The basic idea is that a theory must be expressed in some language. If you dont
express it in a language, how do you know what it is? As E.M. Forster is alleged to
have said, How can I know what I think till I see what I say?. And if you dont write
your theory down in some language, no one can evaluate it.
Scientific theories, on this view, are sets of sentences. And the sentences have to
be in some language: Some theories are expressed as a program written in English,
some in the language of mathematics, some in the language of formal logic, some in
the language of statistics and probability. The claim here is that some theories can be
expressed in a programming language.
One advantage of expressing a theory as a computer program is that all details must
be filled in. That is, a computer program must be a full implementation of the theory.
Of course, there will be implementation-dependent details: For example, if the theory
is expressed in Java, there will be details of Java that are irrelevant to the theory itself.
So, one must try to ensure that such details are indeed irrelevant. One way to do so is to
make sure that two computer programs expressing the same theory but that are written
in two different programming languageswith different implementation-dependent
detailshave the same input-output, algorithmic, and data-structure behavior (that is,
that they fully emulate each other).
Another advantage of expressing a theory as a computer program is that you can
run the program to see how it behaves and what predictions it makes. So, in a sense,
the theory becomes its own model and can be used to test itself.
15.5 Models
The notion of model is associated with the Puzzle of the Model in the Middle, or the
Model Muddle [Wartofsky, 1966], [Wartofsky, 1979], [Rapaport, 1995].
There are different uses of the term model: It can be used to refer to a syntactic
domain, as in the phrase mathematical model of a real-world situation. And it can
be used to refer to a semantic domain, as in the phrase set-theoretic model of a
mathematical theory.
15.6. COMPUTER PROGRAMS ARENT THEORIES 571
This dual, or Janus-faced,1 nature of models leads to what [Smith, 1987] called
a correspondence continuum: Scientists typically begin with data that they then
interpret or model using a formal theory; so, the data are the syntactic domain, and
the formal theory is its semantic domain. The formal theory can then be modeled set-
theoretically or mathematically; so, the formal theory is now the syntactic domain, and
the set-theoretic or mathematical model is the semantic domain. But that set-theoretic
or mathematical model can be interpreted by some real-world phenomenon; so, the
model is now the syntactic domain, and the real-world is the semantic domain. To
close the circle, that real-world phenomenon is just the some of the kind of data that
we started with! (Compare the example of student records and students at the end of
our discussion of implementation, in 14.4.)
15.6 Computer Programs Arent Theories

NEED TO SAY MORE ABOUT THEIR ARGUMENTS!
However, [Moor, 1978, 4] and [Thagard, 1984] argue that computer programs are
not theories, on the grounds that they are neither sets of (declarative) sentences nor
set-theoretic models of axiom systems.
Here are some questions to ask yourself about this position:
1. Must a syntactic theory be expressed in declarative sentences? (A declaritive

sentence is a sentence that is either true or else false.)
2. Must a computer program be expressed in imperative language? (An imperative
sentence is a sentence that says Do this!; it has no truth value.)
3. Must a semantic theory be a set-theoretic model of a real-world situation?

4. Could a computer processthat is, a program being executedbe a model of a
real-world sitation?
1 http://en.wikipedia.org/wiki/Janus

1. Notes on Simon & Newell 1956 (see Highly Recommended readings, above):
(a) Here are one or two (or three?) arguments to the effect that programs are
theories:
i. They equate model and theory (p. 66), but they admit that some
people limit the notion of model to mathematical theory. They also
state that mathematics is a language (p. 66), one of three main kinds
of scientific languages or theories: the mathematical, the verbal, and
the analogical (p. 67).
Presumably, they would classify programming languages as being of

the mathematical kind, from which it would follow that computer
programs are theories expressed in that language.
ii. It should also be noted that, at the end of the article, they equate
theory with analogy (p. 82; see also [Hofstadter and Sander, 2013].
Insofar as a computer program (better?: a computer process) is an
analogy to some real-world phenomenon, it is thereby a theory of that
phenomenon.
iii. More precisely, they define a theory simply as a set of statements
or sentences. (They may, of course, be mathematical statements, or
equations, instead of verbal statements.) (p. 67).
Presumably, the statements of a theory could be statements of a

computer program, expressed in a programming language. Then,
clearly, a computer program, considered as a set of statements, would
be a theory.
(b) A theory, for them, is a syntactic object: a set of sentences. Its semantic
counterpart is the content of the theory, desribed thus:
. . . the content of a theory is comprised of all the assertions about
the world, whether true or not, that are explicitly stated by the
theory or that can logically be inferred from the statements of the
theory. (p. 68, my italics)
First, they seem to be considering three items here: (syntactic) sentences of
some language, the assertions made by those sentences, and the world.
I think that what they are calling assertions are what others might call
propositions. (We discussed the relation of a sentence to a proposition in
Ch. 12; briefly, the two sentences It is snowing (in English) and Il neige
(in French) both express a single proposition.) So, the English sentence
It is snowing expresses the proposition that it is snowing (or, to use their
terminology, that sentence asserts that it is snowing), that that proposition
(or assertion), in turn is true of the world just in case that it is (really)
snowing.
Similarly, the computer programming sentence x := x + 1 might express

the situation (lets call it that) that x is to be incremented by 1, which will be
(or become) true if the contents of the register named x are incremented
by 1. (We will discuss this kind of semantics for programming languages a
bit more in Ch. 16.)
Second, normally, one can only logically infer one statement from another
in a declarative language. If a programming language is not declarative,
then a logic for it might need to be developed in order to be able to
talk about logical inference of one statement from another. There are, in
fact, logics for imperative languages, so this need not be a serious issue.
Moreover, as we will see in Ch. 16, declarative comments can be added
to programs, and one can reason about the statements in a program by
reasoning about those comments.
(c) Another important point that they make is that the language that a theory
is expressed in can make the content of that theory easier or harder to
understand and analyze: It is easier to do arithmetic using Arabic numerals
than Roman numerals (p. 69).
(d) They do explicitly discuss computers in this context, first on p. 71, where
they talk about Computing machines that have been programed [sic] to
represent a particular theory. . . . Note that here it is the physical computer
that represents a theory. This is slightly different from talking about the
computers program as expressing a theory. On their way of talking the
computer is programed to carry out the arithmetic computations called for
in the mathematical theory [of some phenomena]. Thus, the computer is
an analogue for the arithmetic process. I find this rather unclear, but, in
any case, they go on to emphasize (italics in the original) that This is
not, however, the only way of employing digital computing machines as
theories.
(e) There is certainly a difference between a theory and the world that it is a
theory of. One such difference is this:
Why should theories of all kinds make irrelevant statements
possess properties not shared by the situations they model? The
reason is clearest in the case of electromechanical analogues. T
operate at all, they have to obey electromechanical lawsthey
have to be made of somethingand at a sufficiently microscopic
level these laws will not mirror anything in the reality being
pictured. If such analogies serve at all as theories of the
phenomena, it is only at a sufficiently high level of aggregation.
(p. 74).
This is the problem of implementation-dependent details that we discussed
in Ch. 14. It is an unavoidable problem arising whenever an Abstraction
is implemented. It is only at the more abstract levels (sufficiently high
level[s] of aggregation) that we can say that an implementation and a
corresponding abstract theory are the same.
(f) Here is a reason why theories expressed as computer programs may be

better than theories expressed in mathematics or in English (verbally): It
has to do with the idea that such computational theories are analogies:
. . . what is the particular value of the computer analogy? Why not
work directly toward a mathematical (or verbal) theory of human
problem-solving processes without troubing about electronic
computers? ldotsit is at least possible, and perhaps even plausible,
that we are dealing here with systems of such complexity that we
have a greater chance of building a theory by way of the computer
program than by a direct attempt at mathematical formulation.
(p. 81)
Note first that they seem to consider computer programs as analogy
theories, not mathematical theories! Second, computation is perhaps the
best way of managing complexity (as we saw in Ch. 3).
(g) There are two advantages of expressing a theory in a programming
language. First, we would experiment with various modifications of
the. . . program to see how closely we could simulate in detail the observable
phenomena (pp. 8182). In other words, we can run the program to see
how it behavesto see how good a theory it isand we can then modify
the program (and then run the modified version) in order to make it a better
theory.
Second, the program can be written in such a way that it explains what it
is doing: The computer, however complex its over-all program, could be
programed to report, in accurate detail, a description of any part of its own
computing processes in which we might be interested (p. 82). This, of
course, can make it easier to debug and improve the program, or, to use
different terminology, to correct and improve the theory.
2. Where Simon and Newell 1956 consider models and theories to be the same
thing, others distinguish them. [Apostel, 1961, 12] suggests that a computer
can be a model of the central nervous system, and that that model might be
easier to study than the system itself. If it is the computer that is the model, then
it makes sense to say that the computers program expresses the theory that the
model is a model of. (Keep in mind, however, that there is an ambiguity over
whether the model is a model of a theory or of some other phenomenon.)
3. For [Weizenbaum, 1976, Ch. 5, pp. 140143], A theory is first of all a text,
hence a concatenation of the symbols of some alphabet. But A theory
is. . . not merely any grammatically correct text that uses a set of terms somehow
symbolically related to reality. It is a systematic aggregate of statements of laws.
Its content. . . lies at least as much in the structure of the interconnections that
relate its laws to one another, as in the laws themselves. And we have to be abel
to draw. . . consequences from it. So, a theory, for Weizenbaum, is what we
have elsewhere (WHERE?) called a formal theory: a set of axioms (laws) and
rules of inference expressed in some formal language. With a few modifications
(such as to convert this kind of declarative system into an imperative one), this
can be seen to describe a computer program expressed in some programming
language.
Indeed, [Weizenbaum, 1976, 144145] says just this:
. . . theories are texts. Texts are written in a language. Computer

languages are languages too, and theories may be written in them.
. . . Theories written in the form of computer programs are ordinary
theories as seen from one point of view. . . . But the computer
program has the advantage [over a set of mathematical equations or
even a theory written in English] not only that it may be understood
by anyone suitable trained in its language, . . . but that it may also be
run on a computer. . . . A theory written in the form of a computer
program is thus both a theory and, when placed ona computer and
run, a model to which the theory applies.
He cites Newell and Simon (1958?) in support of this.
I would make a slight distinction between the theory (the program) and the model
(the computer or process).
For [Weizenbaum, 1976, 143144], a model satisfies a theory; that is, a model
obeys those laws of behavior that a corresponding theory explicitly states or
which may be derived from it. We may say, given a theory of a system B, that A
is a model of B if that theory of B is a theory of A as well. So, a model (all by
itself) is a (semantic) implementation of a theory. And a model of some system
is a distinct implementation of a theory of that system. We begin with some
system, we develop a theory about it (so the original system is a model of that
theory), and then we can find or create another system that is also a model of
that theory. If the second system is easier to understand than the original system,
then we can use both the theory and the new model to understand the original
system.
If a computer program is such a theory, then what would count as a model of it?
One possibility, consistent with what others have said, is that a computer that is
executing the program is the model. (Another possibility is that the process that
is being executed is the model.)
4. Not sure exactly where this comment should go. Possibly much earlier, in Ch. 3,
where I talk about computer science as the study of what is computable. Or
possibly later, in Ch. 19, when I discuss computationalism in cognitive science.
In any case, [Weizenbaum, 1976, Ch. 6, esp. p. 157] makes am important point
about computer programs as theories (and computers as models). He notes that,
when one studies a model B of some theory of a system X, because B is easier

to study than the system X that is really of interest, one is essentially using a
metaphor. But such uses have their limitations:
To those fully in the grip of the computer metaphor, to understand X

is to be able to write a computer program that realizes X.
The problem that Weizenbaum is pointing to is that, while B might be a good way
to understand some aspects of X, it is unlikely to give us a full understanding of
X or to be the best or only way to understand X. There are limitations to any
theory and to any model.
This does not mean, however, that one shouldnt try to understand X via B.
5. On [Thagard, 1984]:
(a) [Thagard, 1984, 77] argues that, on two conceptions of what a theory is
the syntactic and the semanticcomputer programs are not theories.
He also argues that programs are not models. Rather, a program is a
simulation of a model which approximates to a theory.
On the syntactic theory of theories, a theory is a set of sentences (perhaps

expressed as a formal system with axioms GIVE CITATION TO SECTION
OF THIS BOOK WHERE THIS IS DISCUSSED). On the semantic theory
of theories, a theory is a definition of a kind of system. Presumably, he
will argue that a program is neither a set of sentences nor a definition (of a
kind of system). He has not yet said what a model is.
(b) Indeed, on pp. 7778, he argues that, because programs are sets of
instructions, which do not have truth values and hence are not sentences,
programs cannot be theories in the syntactic sense. One question that
can be raised about this is whether programs written in a programming
language such as Prolog, whose statements can be interpreted as declarative
sentences with truth values, could be considered to be theories. If so, then
why couldnt any program that was equivalent in some sense to such a
Prolog program also be considered a theory?
Another question that can be raised is this: Suppose that we are looking for
a buried treasure. I might say, I have a theory about where the treasure is
buried: Walk three paces north, turn left, walk 5 paces, and then dig. Ill
be that you find the treasure. Is this not a theory? I suppose that Thagard
might say that it isnt. But isnt there a sentential theory that is associated
with itperhaps something like The treasure is buried at a location that
is three paces north and 5 paces west of our current location. Doesnt my
original algorithm for finding the treasure carry the same information as
this theory?
(c) [Thagard, 1984, 78] considers a model to be a set-theoretic interpretation

of the sentences in a [syntactic] theory. . . . . . . a system of things. . . which
provide an interpretation of the sentences. But a program is not. . . a
system of things, nor does itldotsprovide an interpretation for anything.
Hence, a program is not a model.
But a program being executeda processcan be considered to be a

system of (virtual) things that are interpretations of data structures in the
program. If a process might be a model, then why couldnt the program be
a theory?
(d) On the semantic (or structuralist [Thagard, 1984, 78]) view of theories,
a theory is a definition of a kind of natural system (p. 79). Given
some scientific laws (which, presumably, are declarative, truth-functional
sentences, perhaps expressed in the language of mathematics), we would
say that something is a certain kind of natural system if and only if it is a
system of objects satisfying those laws. (The system is defined as being
something that satisfies those laws.)
But this seems very close to what a model is. In fact, [Thagard, 1984, 79]
says that a real system R is a system of the kind defined by the theory T .
But how is that different from saying that R is an implementation of (that
is, a model of) T ?
Thagards response is that, first, a program. . . can not be said to define a

system and that, second, programs contain a host of characteristics which
we know to be extraneous to the real system that they are supposed to be
like (p. 79). (We have discussed the notion of such implementation side-
effects in Ch. 14. HAVE WE?)
(e) First, we can at best say that a program simulates a system: it does
not define a system. . . [Thagard, 1984, 79]
He says this because simulations arent definitions: A simulation of the
solar system, to use his example (see 9.7.2), doesnt define the solar
system. This seems reasonable, but it also seems to support the idea that a
process (not necessarily a program) is a model (and hence that a program
would be a theory).
(f) As for the problem of implementation details, [Thagard, 1984, 80] says
that if our program [for some aspect of human cognition] is written in
LISP, it consists of a series of definitions of functions. The purpose of
writing those functions is not to suggest that the brain actually uses them,,
but to simulate at a higher level the operation of more complex processes
such as image. . . processing. In other words, the real system that is
being simulated (modeled?) by the program (process?) need not have LISP
functions. But, as he notes, it will have complex processes that do the
same thing as the LISP functions. But isnt this also true of any theory
compared to the real system that it is a theory of? A theory of cognitive

behavior expressed in declarative sentences will have, say, English words
in it, but the brain doesnt.
(g) [Thagard, 1984, 80] does admit that he shall take models to be like
theories (on the semantic conception) as being definitions of kinds of
systems. And he notes that a model contains specifictions which are
know to be false of the target real systemthat is, implementation details!
The problem with this, according to Thagard, is that if you try to make a
prediction about the real system based on the model, you might erroneously
make it based on one of these implementation details (pp. 8081). But
that seems to be a problem endemic to any model (or any theory, for that
matter).
(A similar point is made in [Humphreys, 1990, 501]: Inasmuch as the
simulation has abstracted from the material content of the system being
simulated, has employed various simplifications in the model, and uses
only the mathematical form, it obviously and trivially differs from the real
thing, but in the respect, there is no difference between simulations and
any other kind of mathematical model. . . .)
If you make a prediction that turns out to be false, you may have to
change your theory or your model. Perhaps you have to eliminate that
implementation detail. (But others will always crop up; otherwise, your
theory or model will not merely describe or simulate the real system; it
will be the real system. But there are well-known reasons why a life-sized
map of a country is not a very good map! CITE CARROLL, BORGES, ET
AL.)
(h) Thagards final summary is not really the wholesale rejection of programs
as theories that it might first appear to be. It is more subtle:
a program P, when executed on a computer, provides a simulation
of a system of a kind defined by a model M, where M defines
systems which are crude versions of the systems defined by a
theory T , and the set of systems defined by T is intended to
include the real system R. [Thagard, 1984, 82]
I can live with this: The process and R are both implementations of T .
6. On Daubert vs. Merrell-Dow:
(a) Questions to consider:
i. Is a computational theory (of X) a theory?
ii. Is a computational theory (of X) a scientific theory?
iii. What is a computational theory (of X)?
(b) Presumably, a computer scientist is an expert on computer science. But
is a computer scientist who writes a computer program about (or whom
develops a computational theory of) something else (X 6= computer
science) thereby an expert on X? (Or must that computer scientist become,
or work with, an expert on X?)
(c) Two good quotations about the nature of science (perhaps they should go
in Ch. 4?):
. . . scientists do not assert that they know what is immutably
truethey are committed to searching for new, temporary
theories to explain, as best they can, phenomena (Brief for
Nicolaas Bloembergen et al. as Amici Curiae 9, cited in Daubert
at II.B.24 in the online version, my italics)
and
Science is not an encyclopedic body of knowledge about the
universe. Instead, it represents a process for porposing and
refining theoretical explanations about the world that are subject
to further testing and refinement. (Brief for American Association
for the Advancement of Science and the National Academy of
Sciences as Amici Curiae 78, cited in Daubert at II.B.24)
(d) Justice Blackmun, writing in Daubert at II.B.24, citing these two quotes,
states that in order to qualify as scientific knowledge, an inference or
assertion must be derived by the scientific method. So, if a computer
program that can, say, identify handwriting is a good scientific theory of
handwriting, then its creator is a scientific expert on handwriting?
There are two concerns with this: First, a computer program that can
identify handwriting need not be a good scientific theory of handwriting. It
might be a lucky guess not based on any scientific theory, or it might not
even work very well outside carefully selected samples. Second, even if it
is based on a scientific theory of handwriting and works well on arbitrary
samples, the programmer need only be a good interpreter of the theory, not
necessarily a good handwriting scientist.
However, if a computer scientist studies the nature of handwriting and
develops a scientific theory of it that is then expressed in a computer
program capable of, say, identifying handwriting, then it would seem to
be the case that that computer scientist is (also) a scientific expert in
handwriting.
(e) Blackmun, writing in Daubert at II.C.28, suggests four texts of whether
a theory or technique [NB: which could include a computer program,
whether or not such programs are (scientific) theories] is scientific
knowledge:
i. Testability (and falsifiability) (II.C.28): computer programs would
seem to be scientific on these grounds, because they can be tested and
possibly falsified, by simply running the program on a wide variety of
data to see if it behaves as expected.
ii. Peer review (II.C.29): Surely, a computer program can (and should!)
be peer reviewed.
iii. Error rate (II.C.30): Its not immediately clear what Blackmun might
have in mind here, but perhaps its something like this: A scientific
theorys predictions should be within a reasonable margin of error. To

take a perhaps overly simplistic example, a polling error of 54 points
is not a very accurate (scientific) measurement, nor is a measurement
error of 5.00000 0.00001 inches if made with an ordinary wooden
ruler. In any case, surely a computer programs errors should be
reasonable.
iv. General acceptance (II.C.31): A computer program that is not based
on a generally accepted scientific theory or on generally accepted
scientific principles would not be considered scientific.
Whether or not Blackmuns four criteria are complete or adequate is not
the point here. The more general point is that, whatever criteria are held to
be essential to a theorys being considered scientific should also apply to
computer programs that are under consideration.
7. Another good quote on the nature of science:

. . . there are important differences between the quest for truth in
the courtroom and the quet for truth in the laboratory. Scientific
conclusions are subject to perpetual revision. Law, on the other hand,
must resolve disputes finally and quickly. (Daubert, at III.35)
Note that this applies not only to the law, but also to everyday life, as Herbert
Simon was well aware in his discussions of the limits of human rationality.
8. Comments on [Simon, 1996a]:
(a) In the late 1950s, the hypothesis was advanced that human
thinking is information processing, alias symbol manipulation.
. . . [T]hese ideas [were translated] into symbolic (nonnumerical)
computer programs that simulated human mental activity at the
symbolic level. The traces of these programs could be compared
in some detail with data that tracked the actual paths of human
thought (especially verbal protocols) in a variety of intellectual
tasks, and the programs veracity as theories of human thinking
could thereby be tested. [Simon, 1996a, 160, my italics]
Note the slide from simulation (something that a model does) to theory
(ordinarily considered as a set of statements). This is more a statement that
(at least some) computer programs are scientific theories than an argument
for that conclusion.
(b) An argument for it can, perhaps, be constructed from this passage:
These programs, which predict each successive step in behavior
as a function of the current state of the memories together with
the current inputs, are theories, quite analogous to the differential
equation systems of the physical sciences. [Simon, 1996a, 161
162]
Heres the reconstructed argument:
i. Differential equation systems of the physical sciences predict

successive steps in physical processes as a function of the current state
together with the current inputs.
ii. Anything that allows prediction (of successive steps in some process
as a function of the current state together with the current inputs) is a
theory.
iii. Therefore, differential equation systems are theories.
iv. Cognitive computer programs predict successive steps in human
cognitive behavior as a function of the current state of the memories
together with the current inputs.
v. Therefore, they are (psychological) theories.
The point is that the reason that we consider differential equation systems to
be theories is the same reason that we should consider computer programs
(cognitive ones in particular, but programs in general) to be theories.
Well, maybe not all computer programs. Arguably, a computer program

for adding two numbers or for computing income tax is not a theory. (But
maybe they should be considered to be theories expressed computationally:
a theory of addition in the first case, a theory of taxation in the second!)
(c) The slide from simulation to theory is intentional; Simon believes that
computer programs are simultaneously both:
Thus the digital computer provided both a means (program) for
stating precise theories of cognition and a means (simulation,
using these programs) for testing the degree of correspondence
between the predictions of theory and actual human behavior.
[Simon, 1996a, 160]
Thus, computer programs are a very special kind of theory. Not only are
they statements, but they are simultaneously modelsinstances of the very
thing that they describe. Well, perhaps not quite: They only become such
instances when they are being executed.
The argument that Thagard makes that programs cant be theories because
they are not sets of declarative sentences just seems parochial. They are
surely sets of (imperative) statements that have the additional benefit that
they can become an instance of what they describe (alternatively: that they
can control a device that becomes an instance of what they describe).
This duality gives them the ability to be self-testing theories. And their
precision gives them the ability to pay attention to details in a way
that theories expressed in English (and perhaps theories expressed in
mathematics) lack.
(d) Simon hedges a bit, however:
. . . a program was analogous to a system of differential (or

difference) equations, hence could express a dynamic theory.
[Simon, 1996a, 161]
So, is it the case that a program is a theory? Or is merely the case that a
program expresses a theory? Perhaps this distinction is unimportant. After
all, it hardly seems to matter whether a system of equations is a theory or
merely expresses a theory. (The distinction is roughly akin to that between
a sentence and the proposition that it expresses.)
9. ...computational models are better able to describe many aspects of

the umiverse better than any other models we know. All scientific
theories can, for example, be modeled by programs. [Knuth, 2001,
168]
Presumably, they are better because some of them can solve (by simulation and
approximation) equations for which there are not yet any analytic solutions.
But note that Knuth is here assuming that such computer programs are models
(of scientific theories).
10. Theres a nice definition of simulation in [Peschl and Scheutz, 2001b, 1],
ascribed to Shannon (but not in the cited work):
Simulation understood as the process of designing a model of

a real system and conducting experiments with this model for the
purpose either of understanding the behavior of the system or of
evaluating various strategies (within the limits imposed by a criterion
or set of criteria) for the operation of the system
In other words, roughly, A simulates B if A is a model of B and if you experiment

with A in order to understand B.
11. According to [Winsberg, 2001, S443], computational simulations are models.

They are created from antecedently known scientific laws concerning some
physical system (p. S444). However, this is not always the case. In cognitive
science, for example, a computer program may be the best or only expression
of those laws; that is, they are not necessarily antecedently known. On the other
hand, even in cognitive science, it may be the case that we know the laws, at least
in a statistical sense, and then we write a program to simulate them.
Interestingly, [Winsberg, 2001, S449] considers a theory to be a family of

models. A model, in this sense, is an abstract mathematical entity. This is
the semantic view of theories. It would seem, on this view, that a theory is an
abstract mathematical structure common to a family of models. The text of a
computer program, then, could be considered an expression of that structure (in
much the same way that both Arabic and Roman numerals are expressions of
numbers considered as structures. When the program is executed, the process is
simulates (or, in some cases, is) the system that it models.
12. [Humphreys, 2002] suggests a view of computational models that can account
for the Civil War-chess example discussed in earlier chapters: one of the
characteristic features of mathematical [including computational] models is that
the same model. . . can occur in, and be successfully employed in, fields with quite
different subject matters (p. S2, my italics). He goes on to say, Let the analog
computer solve one and you automatically have a solution to the other (p. S5)
but there seems no reason not to include digital computers in this observation.
13. Further discussion of simulation can be found in [Denning and Martell, 2015,
212213].

1. Bostrum, Nick (2003), Are You Living in a Computer Simulation?, Philosophical
Quarterly 53(211): 243255,
http://www.simulation-argument.com/
Argues that if the human species is [not] likely to go extinct before reaching a
[technologically advanced] posthuman stage and if any posthuman civilization
is. . . likely to run a significant number of simulations of their evolutionary history,
then we are almost certainly living in a computer simulation. We will discuss this
a bit more in Ch. 20.
See also:
(a) Papakonstantinou, Yannis
(2015), Created Computed Universe, Communications of the ACM 58(6)
(June): 3638.
(b) Hoffman, Donald D. (2009), The Interface Theory of Perception: Natural
Selection Drives True Perception to Swift Extinction, in S. Dickinson,
M. Tarr, A. Leonardis, & B. Schiele (eds.), Object Categorization: Computer
and Human Vision Perspectives (Cambridge, UK: Cambridge University
Press): 148165,
http://cogsci.uci.edu/ddhoff/interface.pdf
Argues that our internal mental image of the external world need not
bear any resemblance to the actual external world, any more than the
graphical user interface for an operation system need bear a resemblance
to the diodes, resistors, voltages and magnetice fields in the computer,
on the grounds that what is important from the standpoint of evolution is
not accuracy but fitness. The very evolutionary processes that endowed
us with our interfaces might also have saddled us with the penchant to
mistake their contents for objective reality (1.6). This seems to be
consistent with Bostrums theory.
2. Coward, L. Andrew; & Sun, Ron (2001??), Some Criteria for an Effective Scientific
Theory of Consciousness and Examples of Preliminary Attempts at Such a Theory,
http://www.cogsci.rpi.edu/rsun/coward-sun-cc2003.pdf
3. Downes, Stephen (1990), Herbert Simons Computational Models of Scientific
Discovery, PSA: Proceedings of the [1990] Biennial Meeting of the Philosophy of
Science Association, Vol. 1, pp. 97108.
A critique of Simons views on the philosophy of science in general, and of
programs as theories in particular.
4. Frigg, Roman; Hartman, Stephan; & Imbert, Cyrille (eds.) (2009), Special Issue on
Models and Simulations, Synthese 169(3): 425626.
http://link.springer.com/journal/11229/169/3 (accessed 17 February 2014)
5. Green, Christopher D. (2001), Scientific Models, Connectionist Networks, and
Cognitive Science, Theory and Psychology 11: 97117,
http://www.yorku.ca/christo/papers/models-TP2.htm
Analyzes the use of connectionist (or neural-network) computer programs as

models of cognition, and argues that Just because two things share some properties
in common does not mean that one models the other. Indeed, if it did, it would mean
that everything models everything else. There must be at least a plausible claim of
some similarity in the ways in which such properties are realized in the model and
the thing being modeled (IV, final paragraph)
6. Humphreys, Paul (1990), Computer Simulations, PSA: Proceedings of the [1990]
Biennial Meeting of the Philosophy of Science Association, Vol. 2, pp. 497506.
7. Humphreys, Paul (2002), Computational Models, Philosophy of Science 69
(September): S1S11,
http://ist.uap.asia/ jpira/compthink/17232634.pdf
8. Lane, Peter C.R.; & Gobet, Fernand (2003), Developing Reproducible and
Comprehensible Computational Models, Artificial Intelligence 144: 251263.
9. Neuman, Peter G. (1993), Modeling and Simulation, Communications of the ACM
36(6) (June): 124.
(a) Contains useful real-life examples of ways in which simulations (and theories) can
fail to be precise models of reality.
(b) Discusses the illusion that the virtual is real (quoting Rebecca Mercuri).
10. Partridge, Derek; & Wilks, Yorick (eds.) (1990), The Foundations of Artificial
Intelligence: A Sourcebook (Cambridge, UK: Cambridge University Press).
Especially the following two sections, containing the articles listed (many of which
are available elsewhere, some online):
3 (Levels of Theory, pp. 95118), containing:
Marr, David (1977), Artificial Intelligence: A Personal View, pp. 97
107; reprinted from Artificial Intelligence 9: 3748.
Boden, Margaret A., Has AI Helped Psychology?, pp. 108111.
Partridge, Derek, Whats in an AI Program?, pp. 112118.
4 (Programs and Theories, pp. 119164), containing:
Wilks 1990, cited above in Recommended Readings
Bundy, Alan; & Ohlsson, Stellan (198?), The Nature of AI Principles,
pp. 135-154; reprinted from AISB Quarterly ##4750.
Simon, Thomas W., Artificial Methodology Meets Philosophy,
pp. 155164.
11. Peschl, Markus F.; & Scheutz, Matthias (2001), Explicating the Epistemological Role
of Simulation in the Development of Theories of Cognition, in Proceedings of the 7th
Colloquium on Cognitive Science (ICCS-01): 274280,
http://www.academia.edu/719102/Explicating the epistemological role of simulation in the development of theories of cognition
12. Scheutz, Matthias; & Peschl, Markus (2000), Some Thoughts on Computation and
Simulation in Cognitive Science, in Proceedings of the 6th Congress of the Austrian
Philosophical Society: 534540, http://tinyurl.com/ScheutzPeschl2000
(first accessed 13 November 2012);
http://hrilab.tufts.edu/publications/scheutzpeschl00linz.pdf
Argues that computer programs are good simulations (and even implementations)
of cognition, but only as long as they respect the temporal metric imposed by
physics.
13. Shieh, David (interviewer) (2009), The Trouble with Computer Simulations: Linked in
with Sherry Turkle, Chronicle of Higher Education (March 27): A14,
http://chronicle.com/article/The-Trouble-With-Computer/5386
Computer simulations have introduced some strange problems into reality.
14. Simonite, Tom (2009, August 29), Soap Bubbles to Take the Drag out of Future Cars,
New Scientist,
http://www.newscientist.com/article/dn17706-soap-bubbles-to-take-the-drag-out-of-
future-cars.html
Despite the increasing sophistication of computer simulations, finding ways to

show complex air flows visually is critical to understanding aerodynamics. . . and
new ways to do that in large wind tunnels are valuable. You cannot solve
everything completely in space and time on a computer, [Alex] Liberzon [of Tel
Aviv University] told New Scientist. Simulations do not capture the full complexity
of wakes and other features, which can exhibit large changes in behaviour caused
by very small changes.
15. Turner, Ray (2010), Programming Languages as Mathematical Theories, in Jordi

Vallverdu (ed.), Thinking Machines and the Philosophy of Computer Science: Concepts
and Principles (IGI Global): 6682.
That computer science is somehow a mathematical activity was a view held by

many of the pioneers of the subject, especially those who were concerned with its
foundations. At face value it might mean that the actual activity of programming
is a mathematical one. Indeed, at least in some form, this has been held. But
here we explore a different gloss on it. We explore the claim that programming
languages are (semantically) mathematical theories. This will force us to discuss
the normative nature of semantics, the nature of mathematical theories, the role
of theoretical computer science and the relationship between semantic theory and
language design.
16. Tymoczko, Thomas (1979), The Four-Color Problem and Its Philosophical Significance,
Journal of Philosophy 76(2) (February): 5783,
http://www.thatmarcusfamily.org/philosophy/Course Websites/Math S08/Readings/tymoczko.pdf
Can a computer program be (part of) a proof of a mathematical theorem?

For a survey of critiques of Tymoczkos arguments, see [Scherlis and Scott, 1983,
3].
17. Winsberg, Eric (2001), Simulations, Models, and Theories: Complex Physical Systems
and Their Representations, Proceedings of the 2000 Biennial Meeting of the Philosophy
of Science Association, Part I: Contributed Papers, Philosophy of Science 68(3)
(September) Supplement: S442S454.
Chapter 16
Can Computer Programs Be

Verified?
Version of 14 June 2015
Mechanical computers should, Babbage thought, offer a means to

eliminate at a stroke all the sources of mistakes in mathematical tables.
. . . A printed record could. . . be generated. . . , thereby eliminating every
opportunity for the genesis of errors. . . . Babbage boasted that his
machines would produce the correct result or would jam but that they
would never deceive. [Swade, 1993, 8687].
Figure 16.1: 2014,

c Jorge Cham
587
588 CHAPTER 16. CAN COMPUTER PROGRAMS BE VERIFIED?

1. Required:
Fetzer, James H. (1988), Program Verification: The Very Idea, Communications
of the ACM 31(9) (September): 1048-1063.
Reprinted in [Colburn et al., 1993, 321358].
A highly controversial essay!
(a) De Millo, Richard A.; Lipton, Richard J.; & Perlis, Alan J. (1979), Social
Processes and Proofs of Theorems and Programs, Communications of the ACM
22(5): 271280.
The prequel to Fetzer 1988, itself highly controversial.
(b) Ardis, Mark; Basili, Victor; Gerhart, Susan; Good, Donald; Gries, David;
Kemmerer, Richard; Leveson, Nancy; Musser, David; Neumann, Peter; & von
Henke, Friedrich (1989), Editorial Process Verification (letter to the editor, with
replies by James H. Fetzer and Peter J. Denning), ACM Forum, Communications
of the ACM 32(3) (March): 287290.
The first of many sequels to Fetzer 1988.
This one includes a strongly worded letter to the editor of CACM, signed by
10 computer scientists, protesting the publication of Fetzer 1988, a reply by
Fetzer, and a self-defense by the editor.
(a) Any of the following 4 essays that started the field of program verification.
Hoares is the most important.
i. McCarthy, John (1963), Towards a Mathematical Science of Computation,
in C.M. Popplewell (ed.), Information Processing 1962: Proceedings of DFIP
Congress 62 (Amsterdam: NorthHolland): 2128.
ii. Naur, Peter (1966), Proof of Algorithms by General Snapshots, BIT 6: 310
316.
iii. Floyd, Robert W. (1967), Assigning Meanings to Programs, in
Mathematical Aspects of Computer Science: Proceedings of Symposia in
Applied Mathematics, Vol. 19 (American Mathematical Society): 1932.
iv. Hoare, C.A.R. (1969), An Axiomatic Basis for Computer Programming,
Communications of the ACM 12: 576580, 583.
(b) Dijkstra, Edsgar W. (1975), Guarded Commands, Nondeterminacy and Formal
Derivation of Programs, Communications of the ACM 18(8): 453457
An example of program verification in action.
[Continued on next page.]
(c) Gries, David (1981), The Science of Programming (New York: Springer-Verlag).
A classic textbook. Contrast its title with that of [Knuth, 1973].
4. Recommended:
Any of the following articles, which are representative of the battle that resulted
from Fetzer 1988 and Ardis et al. 1989.
(a) Pleasant, James C.; AND Paulson, Lawrence; Cohen, Avra; & Gordon,
Michael; AND Bevier, William R.; Smith, Michael K.; & Young, William D.;
AND Clune, Thomas R.; AND Savitzky, Stephen (1989), The Very Idea
(5 letters to the editor), Technical Correspondence, Communications of the
ACM 32(3) (March): 374377.
(b) Fetzer, James H. (1989), Program Verification Reprise:
The Authors Response (to the above 5 letters), Technical Correspondence,
(c) Dobson, John; & Randell, Brian (1989), Program Verification: Public Image
and Private Reality, Communications of the ACM 32(4) (April): 420422.
(d) Muller, Harald M.; AND Holt, Christopher M.; AND Watters, Aaron (1989),
More on the Very Idea (3 letters to the editor, with reply by James H. Fetzer),
Technical Correspondence, Communications of the ACM 32(4) (April): 506
512.
(e) Hill, Richard; AND Conte, Paul T.; AND Parsons, Thomas W.; AND Nelson,
David A. (1989), More on Verification (4 letters to the editor), ACM Forum,
(f) Tompkins, Howard E. (1989), Verifying Feature-Bugs (letter to the editor),
Technical Correspondence, Communications of the ACM 32: 11301131.
Barwise, Jon (1989), Mathematical Proofs of Computer System Correctness,
Notices of the American Mathematical Society 36: 844851.
A cooler head prevails. Besides being an admirably clear and calm summary
of the Fetzer debate, it discusses:
(a)
(b) the relation between algorithms and programs
(c) the possibility of finding fault with an argument yet believing its
conclusion (see A of this book)
(d) the nature of philosophy of X (see 2.8 of this book)
(e) the difference between the truth of a premise and agreeing with it (see A
of this book)
(f) the relation of math to the world
(g) the nature of models, and the difference between models and the real
thing
(h) [Smith, 1985], to be discussed in Ch. 17.
See also:
Dudley, Richard (1990), Program Verification (letter to Jon Barwise (ed.),
Computers and Mathematics column, with a reply by Barwise), Notices of the
American Mathematical Society 37: 123124.
Fetzer, James H. (1998), Philosophy and Computer Science: Reflections on

the Program Verification Debate, in Terrell Ward Bynum & James H. Moor
(eds.), The Digital Phoenix: How Computers Are Changing Philosophy, Revised
Edition (Oxford: Blackwell, 2000): 253273.
The last word?
16.2 Introduction
Is computer programming like mathematics? Or is it more like engineering? After
all, many people identify computer programming with software engineering. Yet
many others think of a program as being like a mathematical proof: a formal structure,
expressed in a formal language. For example, [Suber, 1997a] compares programs to
proofs this way: A programs input is analogous to the axioms used in a proof; the
programs output is analogous to the theorem being proved; and the program itself
is like the rules of inference that transform axioms into theorems, with the program
transforming the input into the output.1
Or perhaps a program is more like the endpoint of a proof, namely, a mathematical
theorem. In that case, just as theorems can be proved (and, indeed, must be proved
before they are accepted), perhaps programs can be proved (and, perhaps, should be
proved before they are used). Can we prove things about programs? What kinds of
things might be provable about them?
Two answers have been given to the first of these questions: yes and no. (Did you
expect anything else?) One of the most influential proponents of the view that programs
can be proved is Tony Hoare:
Computer programming is an exact science in that all the properties of a
program and all the consequences of executing it in any given environment
can, in principle, be found out from the text of the program itself by means
of purely deductive reasoning. [Hoare, 1969, 576, my emphasis]
When the correctness of a program, its compiler, and the hardware of the
computer have all been established with mathematical certainty, it will be
possible to place great reliance on the results of the program, and predict
their properties with a confidence limited only by the reliability of the
electronics. [Hoare, 1969, 579, my emphasis]
I hold the opinion that the construction of computer programs is a

mathematical activity like the solution of differential equations, that
programs can be derived from their specifications through mathematical
insight, calculation, and proof, using algebraic laws as simple and elegant
as those of elementary arithmetic. [Hoare, 1986, my emphasis]
And among those arguing that programs are not like mathematical proofs are the
computer scientists Richard De Millo, Richard Lipton, and Alan Perlis:
. . . formal verifications of programs, no matter how obtained, will not play
the same key role in the development of computer science and software
engineering as proofs do in mathematics. [De Millo et al., 1979, 271]
And the philosopher James Fetzer argues that the things that we can prove about
programs are not what we think they are:
1 Suber makes a stronger claim, that computers are physical implementations of formal systems. This
would require the rules of inference to satisfy the constraints of being an algorithm, but not all formal
systems require that.
. . . there are reasons for doubting whether program verification can

succeed as a generally applicable and completely reliable method for
guaranteeing the performance of a program. [Fetzer, 1988, 1049]
16.3 Program Verification

Program verification is a subdiscipline of computer science. It can be thought of as
theoretical software engineering, or the study of the logic of software.
The idea behind it is to augment, or annotate, each statement S of a program with:
1. a proposition P expressing a pre-condition of executing S, and
2. a proposition Q expressing a post-condition of executing S.
By proposition is meant what computer scientists call a Boolean statement, that
is, a statement that is either true or else false. Note that a mathematical proof consists
of propositions (premises and conclusions) that are Boolean statements, whereas the
proofs themselves (the sequences of propositions beginning with axioms and ending
with a theorem) are not Boolean-valued: They are neither true nor false; rather,
they are more like programs, in that they are either correct (the technical terms are
valid and sound) or incorrect (technically, invalid or unsound). A proof (or
a logical argument) is valid iff it is truth preserving, that is, iff its conclusion must
be true if its premises are true. But, of course, the premises of an argument might
not be true. If they are true, and if the argument is valid, then its conclusion must be
true. Such an argument is said to be sound; it is unsound iff either one or more of its
premises is false or it is invalid. Roughly, a valid argument that is unsound because of
a false premise is like a correct program whose input is garbage; the output of such a
program is also garbage (this is the famous saying: garbage in, garbage out). (For
more on this, see Appendix A.)
A pre-condition of a statement S is a situation (either in the world in which the
program is being executed or in the computer that is executing the program) that must
be true in order for S to be able to be executed; that is, the pre-condition must be true
before S can be executed.
A post-condition of S is a situation that will necessarily be true after S is executed.
Such annotations are typically written as comments preceding and following S in
the program; if comments are signaled by braces, then the annotation would be written
as follows:
{P} S {Q}
Such an annnotation is semantically interpreted as saying:
If P correctly describes the state of the computer (or the state of the world)
before S is executed,
and if S is executed,
then Q correctly describes the state of the computer (or the state of the
world) after S is executed.
16.3. PROGRAM VERIFICATION 593
The state of the computer includes such things as the values of all registers (that is,
the values of all variables).
So, if we think of a program as being expressed by a sequence of executable
statements:
begin S1 , S2 , . . . , Sn end.
then the program annotated for program verification will look like this:
begin {I & P1 }S1 {Q1 }, {P2 }S2 {Q2 }, . . . , {Pn }Sn {Qn }, {O} end.
where:
I is a proposition describing the input,
P1 is a proposition describing the initial state of the computer (or the world),
Qn is a proposition describing the final state of the computer (or the world),
O is a proposition describing the output, and
for each i, Qi logically implies Pi+1 and Qn logically implies O.

(Often, Qi = Pi+1 .)
The claim of those who believe in the possibility of program verification is that we
can then logically prove whether the program does what its supposed to do without
having to run the program. We would construct a proof of the program as follows:
premise: The input of the program is I.

premise: The initial state of the computer is P1 .
premise: If the input is I and the initial state is P1 ,
and if S1 is executed,
then the subsequent state will be Q1 .
premise: S1 is executed.
conclusion: The subsequent state is Q1 .
premise: If Q1 , then P2 .
conclusion: P2 .
premise: If the current state is P2 ,

and if S2 is executed,
then the subsequent state will be Q2 .
conclusion: The subsequent state is Q2 .
... ...
conclusion: The final state is Qn .
premise: If Qn , then O.
conclusion: O.
(If the program isnt a straight-line program such as this, but is a structured
program with separate modules, then it can be recursively verified by verifying each
module.)
If debugging a program by running it, and then finding and fixing the bugs, is part
of practical software engineering, then you can see why program verification can be
thought of as theoretical software engineering.2
Moreover, this annotation technique can also be used to help develop programs that
would thereby be guaranteed to be correct, another reason why program verification is
argued to be an important part of software engineeing [Gries, 1981].
16.4 The Fetzer Controversy

Nonsense!, said [Fetzer, 1988], thus initiating a lengthy controversy in the pages of
the Communications of the ACM and elsewhere, with strongly worded letters to the
editor chastising the editor for publishing Fetzers paper, supportive letters to the editor
praising the decision to publish, and articles in other journals attempting to referee
between the publish-or-not-to-publish controversy as well as the more substantive
controversy over whether programs can, or should, be verified.
In Fetzers terminology, let
hA1 , . . . , An , P1 , . . . , Pm , S1 , . . . , Sl i
be a sequence of propositions, where:
the Ai are axioms, that is, propositions that are necessarily true by virtue of
their meanings (or assumed to be true for the time being),
the Pj are premises, that is, propositions that are contingently or empirically
assumed to be true (but that would normally require some justification), and
the Sk are other propositions.
Then to say that this sequence is a (valid) demonstration (or proof) of Sl from
the axioms and premises means (by definition) that each Sk follows from previous
propositions in the sequence by a (valid) rule of inference.
This is a fairly standard definition of a logical proof of a theorem Sl . Note that if
n = 0 (that is, if there are no axioms, and, especially, if m = 0 alsothat is, if there
are no premises), then there will typically have to be lots of rules of inference, and
the demonstration is said to be done by natural deduction (because it is the way that
logicians naturally prove things [Pelletier, 1999]).
Now, again following Fetzers terminology,
2 On the history of the term bug, see [Hopper, 1981], [Krebs and Thomas, 1981]. However, the idea that
the first computer bug was really a bug (actually, a moth), is an urban legend, because the term was used in
the non-entomological sense as early as 1889; see [Shapiro, 1985] and the OED online at:
http://www.oed.com/view/Entry/24352?rskey=qWlUWg&result=2#eid (accessed 7 November 2013).)
For a photo of the allegedly first bug, see http://www.catb.org/jargon/html/B/bug.html
16.4. THE FETZER CONTROVERSY 595
Sl is abolutely verifiable =de f m = 0

Sl is relatively verifiable =de f m > 0
In either case, Sl is verifiable relative to the premises.3 Note that what Fetzer calls
absolute verifiability is still verifiability except that the verifiability is relative to the
axioms (not to the premises).
Given all of this terminology, Fetzer phrases the fundamental question of program
verification this way: Are programs absolutely verifiable? (One question that you
should keep in mind as you read the papers involved in this controversy is this:
Do the pro-verificationists claim that programs are absolutely verifiable, in Fetzers
terminology?)
To be absolutely verifiable requires there to be program rules of inference that
are themselves justified or truth-preserving, or it requires there to be program axioms
that are necessarily true about the performance that a machine will display when such
a program is executed [Fetzer, 1988, 1052].
But note that this is different from program verification in the Hoare-Gries tradition,
because it conflates logical relations with causal relations. The former are abstract; the
latter are part of the real world. Consider these analogies:
A1 A proof is to a theorem as verification is to a program: You prove theorems; you
verify programs.
A2 Axioms, premises, and rules of inference are to theorems as input and programs
are to outputs. Therefore, perhaps, verifying a program is like proving a theorem.
(This can be called the Suber-Fetzer analogy.) But on this analogy, verifying a
program is more like proving the inference rules! So, analogy (A1) seems better.
A3 Axioms, premises, and intermediate conclusions are to theorems as input and
program are to output. Both intermediate conclusions and program are a
sequence (or at least a set) of expressions that begin with something given and
end with a desired result. But you are trying to prove the theorem; you are not
trying to prove the intermediate conclusions. And you verify an entire program,
not just its output.
Moreover, in favor of [A1], most theorems are of the form if P, then Q; most
programs can be put in the form if you do S1 , . . . , Sn , then youll accomplish goal
G.
Fetzer claims that theorems are not like programs, because only programs have
causal and physical interpretations as their intended interpretation. And this is where
the main problem arises, in Fetzers point of view: What about program statements
that specify output behaviors, such as ringing a bell (some programing languages
have a command BEEP whose intended behavior is to ring a bell) or drawing a
circle (suppose you have a programming language one of whose legal instructions is
DRAW CIRCLE(x, y, r), whose intended behavior is to draw a circle at point (x, y)
with radius r). How can you prove or verify that the program will do this? How can
3 In fact, the conclusion of a formal proofa theoremis only true relative to the axioms of the formal
theory. For discussion, see [Rapaport, 1984, 613].

you mathematically or logically prove that the (physical) bell will (actually) ring or
that a (physical) circle will (actually) be drawn? How can you logically prove that the
(physical bell works or that the pen has ink in it? Fetzers point is that you cant.
(And the controversy largely focused on whether thats whats meant by program
verification.) This should remind you of the discussion in Ch.10 about Clelands
interpretation of the Church-Turing Computability Thesis: Is preparing Hollandaise
sauce, or physically ringing a bell, or physically drawing a circle a computable task?
But, according to Fetzer, its not just ringing bells, drawing circles, or, for that
matter, cooking thats at issue. What about the PRINT command that is required
of all Turing-machine programs? And, according to Fetzer, its not just a matter of
causal output, because you can replace every PRINT(x) command with an assignment
statement: p := x. Even this is a causal statement, because it instructs the computer
to change the values of bits in a physical register p, and so Fetzers argument goes
through: How can can you logically prove that the physical computer will actually
work? Indeed, the history of early modern computers was largely concerned with
ensuring that the vacuum tubes would be reliable [Dyson, 2012b].
The point, according to Fetzer, is that we must distinguish between the program
and the algorithm that it implements. The former implements the latter: A program is
a causal model of a logical structure, and, while algorithms might be capable of being
absolutely verified, programs cannot.
It might be replied, on behalf of the pro-verificationists, that we can still do relative
verification: verification relative to causal axioms that relate these commands to
causal behaviors. So, we can say that, if the computer executing program P is in
good working order, and if the world (the environment surrounding the computer) is
normal, then P is verified to behave in accordance with its specifications.
No, says Fetzer: Algorithms and programs that are only intended for abstract
machines can be absolutely verified (because there is nothing physical about such
machines; they are purely formal). But programs that can be compiled and executed
can only be relatively verified.
16.5 Summary
The bottom line is that programs of the latter type need causal rules of inference of
the form: input I causes output O. Perhaps the BEEP command would have to be
annotated something like this:
{The bell is in working order.} BEEP {A sound is emitted.}
If such causal rules are part of the definition of an abstract machine, then we can have
absolute verification of the program. But if they are merely empirical claims, then we
can only have relative verification of the program.
Even so, absolute verification is often thought to be too tedious to perform and can
lure us into overconfidence. The former problem seems to me not to be overly serious;
its tedious to prove theorems in mathematics, too; in any case, techniques are being
devised to automate program verification. The latter problem is more important, for
16.5. SUMMARY 597
precisely the reasons that Fetzer adduces. Just because youve proved that a program
is correct is no reason to expect that the computer executing it will not break down.
The general tone of the responses to Fetzer included these objections:
How dare you publish this! [Ardis et al., 1989]
So what else is new? We program verificiationists never claimed that you could
logically prove that a physical computer would not break down.
Verification techniques can find logical faults; it is logically possible to match a
program or algorithm to its specifications.
You can minimize the number of IcO (??) rules such that they only apply to
descriptions of logic gates and the physics of silicon.
Many programs are incorrect because, for example of the limits of accuracy in
using real numbers.
Verifiably incorrect programs can be better than verifiably correct programs if
they have better average performance (cf. [Moor, 1979]).

1. Each proposition of Euclids Elements consisted of an algorithm (for
constructing a geometric figure using only compass and straightedge) and
a proof of correctness. See, e.g., the statement of Proposition 1 at
http://tinyurl.com/kta4aqh (accessed 17 February 2014).4
2. Heres another way to express Fetzers point: Even if program verification

techniques can prove that a program is correct, there may still be performance
limitations. So, Fetzers point may be along the lines of Chomskys competence-
performance distinction.
3. A Turing machine is like an actual digital computing machine, except

that (1) it is error free (i.e., it always does what its table says it should
do), and (2) by its access to an unlimited tape it is unhampered by
any bound on the quantity of its storage of information or memory.
[Kleene, 1995, 27]
The former limitation of real Turing machines does not obviate the need
for program verification. Even an ideal Turing machine could be poorly
programmed.
4. The remarks. . . on the desired automatic functioning of the

device [that is, von Neumanns definition of a computer, as
quoted in Ch. 9, 19] must, of course, assume that it functions
faultlessly. Malfunctioning of any device has, however, always a
finite probabilityand for a complicated device and a long sequence
of operations it may not be possible to keep this probability negligible.
Any error may vitiate the entire output of the device. For the
recognition and correction of such malfunctions intelligent human
intervention will in general be necessary.
However, it may be possible to avoid even these phenomena
to some extent. The device may recognize the most frequent
malfunctions automatically, indicate their presence and location by
externally visible signs, and then stop. Under certain conditions
it might even carry out the necessary correction automatically and
continue. [von Neumann, 1945, 1.4, p. 1].
One way to read this is as a recognition/anticipation of Fetzers point. Given
this inevitability, the focus presumably has to be on the elimination of logical
errors, so that program verification still has a role to play. The second paragraph
suggests that some machine verification might be automated, but that just
leads to an endless regress: Even if the logical structure of that automation is
guaranteed, the physcial device that carries it out will itself be subject to some
residual malfunction possiblities.
4 http://www.perseus.tufts.edu/hopper/text?doc=Perseus%3Atext%3A1999.01.0086%3Abook%3D1%3Atype%3DProp%3Anumber%3D1
Of course, another way to read this (as well as the entire program verification
debate) is to recognize that no one, and no thing, is perfect. Theres always
the chance of error or malfunction, and the point is, at least, to minimize it (if
complete elimination is physically impossible).
5. The entire issue of program verification might be considered as a subset of

the more general engineering issue of reliability. ([Newell, 1980, 159], for
example, assumes that a symbol system should be totally reliable, nothing in
its organization reflecting that its operators, control or memory could be erroful
[sic!]. He goes on to say that universality is always relative to physical limits, of
which reliability is one (p. 160), where universality is defined as the ability to
produce an arbitrary input-output function (p. 147). This suggest that, even
if a program could be proved mathematically to be correct, the process that
executes it would still be limited by physical correctness, so to speak, and that,
presumably, cannot be mathematically proved. (Note that this point was already
made in 1980.)
Reliability is also related to what [Staples, 2014, Staples, 2015] considers to

be the heart of engineering knowledge, namely, how well artifacts (including
computer programs) meet their specifications (see the discussion in Ch. 5, 5.7
at 22).) [Carhart, 1956, 149] argues that it is not sufficient just to look at
the physical components of a complex system, but that the interactions of the
components are also important. He calls for a total systems approach that
would include not only the physical components (hardware) but also the people
who operate it (which he calls softwarerecall our discussion in 12.4 at 10).
Adapting this to the Fetzer controversy, we might want to say that it is not
sufficient merely to verify the static computer program (and its components),
but also the way that program is dynamically executed.
6. Put this quote at the beginning of this chapter? Or of Ch. 10?
I should like to ask the same question that Descartes asked. You
are proposing to give a precise definition of logical correctness
which is to be the same as my vague intuitive feeling for logical
correctness. How do you intend to show that they are the same?
. . . [T]he average mathematician. . . should not forget that intuition is
the final authority. . . . [Rosser, 1978, 4,11], slightly incorrectly cited
in [De Millo et al., 1979, 271]
[De Millo et al., 1979] cite this presumably to support their argument that formal
proofs dont always yield acceptance or belief. (My college friend Alan Arkawy
always advised that you had to believe a mathematical proposition before you
could try to prove it.)
Note the similarity to the CTCT; this is not accidental. The Descartes cited
in the quote is not the real Descartes, but a fictional version visited by a time-
traveling mathematician who tries to convince him that the modern and formally
precise, - definition of a continuous curve is equivalent to Descartess
intuitive definition in terms of being able to be drawn without lifting pencil from
paper. They agree that, despite the informality of Descartess definition and the
formality of the mathematicians, they can be verifiedbut not provedto
be equivalent by seeing that they agree on a wide variety of cases. The cited
quotation is the mathematicians response to a logician who tries to convince the
mathematician that the mathematicians intuitive notion of proof bears the same
relation to Descartess intuitive notion of continuity that the logicians formal
and logical definition of proof bears to the mathematicians intuitive notion of
proof.
7. Comments on [Scherlis and Scott, 1983]:
(a) They argue that program verification and development should go hand-in-
hand, rather than verification coming after a program is complete. Others
have made the same point (e.g., Dijkstra, Gries: CITE), but Scherlis
& Scotts notioninferential programmingdiffers from program
derivation. Whereas program derivations [are] highly structured
justifications for programs[,] inferential programming [is] the process of
building, manipulating,and reasoning about program derivations (p. 200,
my italics).
(b) A correctness proof [shows] that a program is consistent with its
specifications (p. 201), where Specifications differ from programs in
that they describe aspects or restrictions on the functionality of a desired
algorithm without imposing constraints on how that functionality is to be
achieved. (p. 202)
(c) They take issue with [De Millo et al., 1979]. First, they observe that
the claim that Mathematicians do not really build formal proofs in
practice; why should programmers? (p. 204) is fallacious, because
formalization plays an even more important role in computer science
than in mathematics, and this, in turn, because computers do not run
informal programs (p. 204). Moreover, formalization in mathematics
has made possible much advancement independent of whether there is
any sense in looking at a complete formalizatio of a whole proof. Often
there is not. (p. 204)
(d) They advocate, not for a complete proof of correctness of a completed
program, but for proofs of correctness of stages of development, together
with a justification that derivation steps preserve correctness. This is
exactly the way in which proofs of theorems are justified: If the axioms
and premises are true, and if the rules of inference are truth-preserving,
then the conclusions (theorems) will be true (relative to the truth of the
axioms and premises).
(e) Where [De Millo et al., 1979] claim that some view theorems and proofs as
analogous to programs and verifications, respectively, but that theorems are
more like specifications, proofs like programs, and formal demonstrations
(were anyone to actually sit down and provide one) like verifications,
[Scherlis and Scott, 1983] say that a better analogy is this: problems are
like specifications, theorems are like programs, and proofs are like program
derivations. (p. 207). They add that all such analogies must be taken with
a big grain of salt, since all thee words can mean many things.
8. FETZER ON LOGIC, REASONING, AND ARGUMENT ANALYSIS
Fetzers paper makes some comments about the nature of logical reasoning and
about knowledge and belief that are relevant to what youve been doing in your
argument analyses.
(a) On p. 1050, column 1, he says:

[W]hat makes . . . a proof a proof is its validity rather than its
acceptance (by us) as valid, just as what makes a sentence true is
[that] what it asserts to be the case is the case, no[t] merely that it
is believed (by us) and therefore referred to as true.
Note that Ive been allowing you to evaluate the truth-value of the premises
of an argument, not by trying to demonstrate whether they are true, but
by trying to say whether and why you believe them. According to Fetzers
quote, it would follow that you are not really Jevaluating the premises.
Hes correct! Whether a statement (or premise) is true or not does not
depend on whether you (or anyone) believes it to be true. It is (or isnt) true
iff what it states corresponds to reality.
Nevertheless, thats very hard (if not impossible) to prove. And thats
why Im allowing you to do something that is a bit easier, and a bit more
realistic, andfor our purposesjust as useful, namely, to try to explain
whether and why you believe the statement.
(b) In the same location, at the beginning of the next section, Fetzer says:
Confidence in the truth of a theorem (or in the validity of an
argument) . . . appears to be a psychological property of a person-
at-a-time. . . .
Its that confidence that Ive been asking you to examine, explain, and
defend. Because its a psychological property of you now, I only grade
you on how well you explain and defend it, not on what it is.
(c) Finally, in column 2 on the same page, he says:
[A]n individual z who is in a state of belief with respect to a
certain formula f . . . cannot be properly qualified as possessing
knowledge that f is a theorem unless his belief can be supported
by means of reasons, evidence, or warrants. . . .
This is a very complicated way of making the following important point: If

you believe a statement f , that belief doesnt count as knowing that f is the
case unless you have a reason for your belief. In other words, knowledge
is belief plus (at least) a reason. (Actually, most philosophers agree that
knowledge requires a third thing: knowledge is belief, plus a reason, plus
f being true: You cant know something thats false.)
This need to justify your beliefs is what turns a mere opinion, or an

expression of feeling, into a claim that is worthy of holding and of
convincing others of. Its why we have arguments to justify conclusions,
and why we have to also justify all the premises of the arguments. (And
its why we have to justify, recursively, all the justifications, until we reach
some starting point that is a self-justifying belief. But its not clear that
there really are any, which means that our investigations may never end!)
9. At the 40th Anniversary celebration of the founding of the UB Department

of Computer Science & Engineering, our 2nd Ph.D., Bruce Shriver, former
president of the IEEE Computer Society, said in his address that hardware does
not have flaws, only software does. Note that this is almost exactly the converse
of Fetzers point!
10. Comments on Barwise 1989:
(a) Perhaps incorporate here some my observations in the Readings section,

above.
(b) Barwise sees the issue between Fetzer and his opponents as being a special
case of the more general question of how mathematics can be applied to the
real world, given that the former is abstract and purely logical whereas the
latter is concrete and empirical (p. 846, col. 2). (See also [Wigner, 1960]
and 4.9, 7.10, esp. 37.) This fits with other comments above about the
ways in which mathematical expressions such as programs can be formally
proved, whereas physical objects such as computers are subject to physical
breakdown.
(c) But there is another aspect to that issue in the philosophy of math, namely,
the relation between the syntax of a formal mathematical expression and
its semantic interpretation in the real world:
The axiomatic method says that our theorems are true if our
axioms are. The modeling method says that our theorems model
facts in the domain modeled if there is a close enough fit between
the model and the domain modeled. The sad fact of the matter is
that there is usually no way to proveat least in the sense of
mathematical proofthe antecedent of a conditional of either
of these types. (p. 847, col. 2, italics in original, my boldface)
And this fits in nicely with [Smith, 1985], which well look at in the next
chapter. (He cites [Smith, 1985], noting that Computer systems are not
just phyusical objects that compute abstract algorithms. They are also
embedded in the physical world and they interact with users. . . . Thus, . . .
our mathematical models need to include not just a reliable model of the
computer, but also a reliable model of the environment in which it is to be
placed, inluding the user. (p. 850, col. 1).
(d) Barwise notes that Fetzer is only willing to talk about the causal (that
is, physical) role of computers, which is not susceptible of mathematical
verification, whereas the field of program verification only concerns
abstract programs. (p. 848, col. 2). So it really seems that both sides are
not only talking past each other, but are actually consistent with each other.
(e) As for the issues about the nature and value of proofs as raised by
[De Millo et al., 1979], Barwise observes that mathematical theories of
formal proofs arent what mathematicians since the time of the ancient
Greeks were constructing (p. 849, col. 1). Although he doesnt say so
explicitly, he very strongly suggests that the relation between such theories
of formal proofs (and there are more than one) and the informal proofs
since the time of the ancient Greeks is akin to the CTCT.
11. Somewhere note that the program verification issue is also a subissue of the
question concerning the relation of software to hardware.
12. . . . mathematical reasoning alone can

never establish the correctness of a program or hardware design
in an absolute sense, but only relative to some formal specification of
its desired behavior [MacKenzie, 1992, 1066, col. 2].
Similarly, a formal proof of a theorem only establishes its truth relative to the
truth of its axioms, not its absolute truth.
MacKenzie continues:
Mathematical argument can establish that a program or design

is a correct implementation of that specification, but not that
implementation of the specification means a computer system that
is safe, secure, or whatever. [MacKenzie, 1992, 1066, col. 2].
There are two points to notice here. First, a mathematical argument can establish
the correctness of a program relative to its specification, that is, whether the
program satisfies the specification. In part, this is the first point, above. But,
second, not only does this not necessarily mean that the computer system is safe
(or whatever), it also does not mean that the specification is itself correct.
Presumably, a specification is a relatively abstract outline of the solution to
a problem. Proving that a computer program is correct relative tothat is,
satisfiesthe specification does not guarantee that the specification actually
solves the problem!
. . . what can be proven correct is not a physical piece of hardware,

or program running on a physical machine, but only a mathematical
model of that hardware or program. [MacKenzie, 1992, 1066, col. 2].
This is the case for reasons that [Smith, 1985] discusses and that we will look
at in the next chapter. It is not unrelated to the CTCT: You cant show that two
systems are the same, in some sense, unless you can talk about both systems
in the same language. In the case of the CTCT, the problem concerns the
informality of the language of algorithms versus the formality of the language
of Turing machines or recursive functions. In the present case, the problem
concerns the mathematical language of programs versus the non-linguistic,
physical nature of the hardware. Only by describing the hardware in a (formal,
mathematical) language, can a proof of equivalence be attempted. But then we
also need a proof that that formal description is correct; and that cant be had.
13. (These comments might go better in Ch. 17, after discussing [Smith, 1985].):
For Fetzer, computing is about the world; it is external and contextual (recall
our discussion of Cleland, in Ch. 10 and elsewhere WHERE?). Thus, computer
programs cant be verified, because the world may not be conducive to correct
behavior: A physical part might break; the environment might prevent an
otherwise perfectly running, correct program from accomplishing its task
(such as making hollandaise sauce on the Moon using an Earth recipe); etc.
For Smith, computing is about a model of the world; it is internal and narrow.
Thus, computer programs cant be verified, but for a different reason, namely,
the model might not match the world.
Perhaps a better way of looking at things is to say that there are two different
notions of verification: an internal and an external one. (For related
distinctions, see [Tedre and Sutinen, 2008, 163164].)
14. Software is sometimes taken to be an artefact. . . . However, is
software mathematical, or a physical thing? . . . [T]he key point is
as follows. Like the design of an artefact, it is the objective content
of a computer program that is important to its characterisation and
identity. The objective content of a program remains unchanged,
regardless of whether it is represented as magnetic regions on a
hard disk, or as ink on a page. . . . Nevertheless, the controversy
and lessons. . . from Fetzers (1988) paper on program verification
remind us that the execution of a program on a physical computer
is categorically different to the content of the program, and different
to a characterisation of the programs possible behaviour using
formal theories. The execution of a program on a computer
system is a phenomena [sic]5 in the real world, and it is this
5 It should be phenomenon.
phenomena [sic] which might change the world in a way that

satisfies requirements.6 Theories of program analysis or formal
verification are not purely mathematicalthey are ultimately also
empirical engineering theories when they support claims that the
behavior of real programmed computer systems meet requirements.
[Staples, 2014, 2, my italics]
Note how the issue that we discussed in Ch.12 about the nature of software vs.
hardware is relevant to the issue of program verification. Does a formal proof of
a programs correctness apply to the program as software or to the program as
hardware (perhaps to the process that comes into existence when the program is
executed)? This is also relevant to a discussion we will have in the next chapter
about the relation between computers and the world.
15. Related to Smiths Gap:
Human fallibility means some of the more subtle, dangerous bugs
turn out to be errors in design; the code faithfully implements the
intended design, but the design fails to correctly handle a particular
rare scenario. [Newcombe et al., 2015, 67]
In other words, the abstraction that produced the design (the specification) was
in error, and those are precisely the kinds of errors that Smith warns about.
6 That is, the way that the program satisfies the requirements of its specification.

1. Arnow, David (1994), Teaching Programming to Liberal Arts Students: Using Loop
Invariants, SIGCSE 94, Phoenix, AZ: 141144.
How to use program-verification techniques to develop programs. Compare to Tam
1992, below.
2. Avigad, Jeremy; & Harrison, John (2014), Formally Verified Mathematics,
Comminications of the ACM 57(4) (April): 6675,
http://cacm.acm.org/magazines/2014/4/173219-formally-verified-mathematics/fulltext
With the help of computational proof assistants, formal verification could become
the new standard for rigor in mathematics (from the introductory blurb, p. 66).
Discusses the history and nature of formal proofs in math.
3. Blum, Bruce I. (1989), Formalism and Prototyping in the Software Process, Information
and Decision Technologies 15: 327341; reprinted in [Colburn et al., 1993, 213238].
4. Bowen, J.P.; & Hinchey, M.G. (1995), Ten Commandments of Formal Methods, IEEE
Computer 28(4): 5663.
5. Burkholder, Leslie (1999), Are AI and Mechanics Empirical Disciplines?, Journal of
Experimental and Theoretical Artifical Intelligence 11: 497500.
Uses Fetzer 1988 and Barwise 1989 to argue that AI is an empirical science.
6. Clarke, Edmund M.; Emerson, E. Allen; & Sifakis, Joseph (2010), Model Checking:
Algorithmic Verification and Debugging, Communications of the ACM 52(11)
(November): 7484.
7. Cohen, Avra (1989), The Notion of Proof in Hardware Verification, Journal of
Automated Reasoning 5: 127139; reprinted in [Colburn et al., 1993, 359374].
8. Colburn, Timothy R. (1991), Program Verification, Defeasible Reasoning, and
Two Views of Computer Science, Minds and Machines 1: 97116; reprinted in
[Colburn et al., 1993, 375399].
9. Colburn, Timothy R. (1993), Computer Science and Philosophy, in
[Colburn et al., 1993, 331].
10. Colburn, Timothy R.; Fetzer, James H.; & Rankin, Terry L. (eds.) (1993), Program
Verification: Fundamental Issues in Computer Science (Dordrecht, Holland: Kluwer
Academic Publishers).
An anthology containing many of the papers discussed or listed in this chapter, as

well as a bibliography that is even longer than this one!
11. Davis, Phillip J.; & Hersh, Reuben (1998), The Ideal Mathematician, in their The
Mathematical Experience (Boston: Houghton Mifflin): 3444,
http://babel.ls.fi.upm.es/pablo/Jumble/ideal.pdf
A partly facetious description of the behavior of (some) mathematicians, including
a discussion of the nature of proof as carried out by mathematicians.
For an antidote to their characterization of mathematicians, read [Frenkel, 2013].
12. Devlin, Keith (1992), Computers and Mathematics (column), Notices of the American
Mathematical Society 39(9) (November): 10651066.
Contains a
dialogue between a math professor who defends [De Millo et al., 1979]s notion
of social proof and a philosophy student who defends the more formal logical
notion of proof. Serves as an introduction to [MacKenzie, 1992].
13. Dewar, Robert (2009), CS Education in the U.S.: Heading in the Wrong Direction?
Point, Communications of the ACM 52(7) (July): 4143.
Argues in favor of teaching formal reasoning in computer science, including
program-verification techniques.
For an opposing viewpoint, see:
Astrachan, Owen (2009), CS Education in the U.S.: Heading in the Wrong
Direction? Counterpoint, Communications of the ACM 52(7) (July): 4345.
14. Dijkstra, Edsgar W. (1974), Programming as a Discipline of Mathematical Nature,
American Mathematical Monthly (June-July): 608612.
Argues that the correctness of programs could and should be established by
proof, that structured programs are simpler to prove than unstructured ones
[Dijkstra, 1968], that theorems about programs make program proofs easier, and
that to prove the correctness of a given program was. . . putting the cart before the
horse. A much ore promising approach turned out to be letting correctness proof
and program grow hand in hand (pp. 609610).
15. Dijkstra, Edsger W. (1983), Fruits of Misunderstanding (EWD-854), reprinted in
Datamation (15 February 1985): 8687,
http://www.cs.utexas.edu/users/EWD/transcriptions/EWD08xx/EWD854.html
16. Fetzer, James H. (1991), Philosophical Aspects of Program Verification, Minds and
Machines 1: 197216; reprinted in [Colburn et al., 1993, 403427].
A summary, reply to objections, and further discussion of the relation of software
to hardware.
17. Fetzer, James H. (1993), Program Verification, in Allen Kent & James G. Williams
(eds.), Encyclopedia of Computer Science and Technology, Vol. 28, Supp. 13 (New
York: Marcel Dekker): 237254; reprinted in Allen Kent & James G. Williams (eds.),
Encyclopedia of Microcomputers, Vol. 14: Productivity and Software Maintenance: A
Managerial Perspective to Relative Addressing (New York: Marcel Dekker): 4764.
An encyclopedia-style article on program verification, but written from Fetzers
perspective.
18. Fetzer, James H. (1996), Computer Reliability and Public Policy: Limits of Knowledge
of Computer-Based Systems, Social Philosophy and Policy 13(2) (Summer): 229266.
. . . in the real world, the operation of computer systems is inherently uncertain,
which raises profound problems for public policy. (from the Abstract)
19. Fetzer, James H. (1998), Computer Systems: The Uncertainty of their Reliability,
Bridges 5(3/4): 197217.
The reliability of computer systems. . . depends on the. . . interaction of hardware

and software. . . and the accuracy and completeness of the knowledge upon which
they are based. . . . The reliability of computer-based systems is necessarily
uncertain. . . [and] must not be taken for granted. (from the Abstract)
20. Floyd, Christiane (1987), Outline of a Paradigm Change in Software Engineering,
in G. Bjerknes et al. (eds.), Computers and Democracy: A Scandinavian Challenge
(Brookfield, VT: Gower Publishing): 191210; reprinted in [Colburn et al., 1993, 239
259].
21. Floyd, R.W. (1967), Assigning Meaning to Programs, Mathematical Aspects of
Computer Science (Proceedings of Symposia in Applied Mathematics 19) (American
Mathematical Society): 1932; reprinted in [Colburn et al., 1993, 6581].
22. Frenkel, Karen (1993), An Interview with Robin Milner, Communications of the ACM
36(1) (January): 9097,
http://delivery.acm.org/10.1145/160000/151241/a1991-frenkel.pdf
A companion piece to [Milner, 1993], with interesting observations on AI, the

semantics of programming languages, program verification, and the nature of
computer science.
23. Glass, Robert L. (2002), The Proof of Correctness Wars, Communications of the ACM
45(8) (August): 1921.
A historical survey, with some useful references, arguing that the controversies over
program verification are extremely healthy for the field of computing (roughly
for the same reasons that, it was argued in Ch. 2, philosophy is important: the
challenging of assumptions).
24. Glanz, James (1995), Mathematical Logic Flushes Out the Bugs in Chip Designs,
Science 267: 332333.
25. Henzinger, Thomas (1996), Some Myths about Formal Verification, ACM Computing
Surveys 28(4es) (December), article no. 119.
A brief position statement in favor of formal verification.
26. Hinchey, Mike; Jackson, Michael; Cousot, Patrick; Cook, Byron; Bowen, Jonathan P.; &
Margaria, Tiziana (2008), Software Engineering and Formal Methods, Communications
of the ACM 51(9) (September): 5459.
A statement of the argument in favor of program verification.
27. Hoare, C.A.R. (1986), Mathematics of Programming, Byte (August); reprinted in

[Colburn et al., 1993, 135154].
A summary of Hoares view that the construction of computer programs is a
mathematical activity. . . , that programs can be derived from their specifications
through mathematical insight, calculation, and proof. . . because Computers are
mathematical machines. . . . Computer progras are mathematical expressions. . . . A
programming language is a mathematical theory. . . even though Nothing is as I
have described it.
28. Hoare, C.A.R. (2009), Retrospective: An Axiomatic Basis for Computer Programming,
29. Humelsine, Jim (2010), Software Still as Much an Art as Science, Letters to the Editor,
30. Lamport, Leslie (2015), Who Builds a House without Drawing Blueprints?,
Communications of the ACM 58(4): 3841,
http://cacm.acm.org/magazines/2015/4/184705-who-builds-a-house-without-drawing-
blueprints/fulltext
Argues that the main reason for writing a formal spec[ification] is to apply tools to
check it, that the math needed for most specifications is quite simple: predicate
logic and elementary set theory, that a specification can and should be embedded
as a comment within the code it is specifying, andperhaps most importantly
that thinking does not guarantee that you will not make mistakes. But not thinking
guarantees that you will.
31. Leroy, Xavier (2009), Formal Verification of a Realistic Compiler, Communications of
the ACM 52(7) (July): 107115, preprint at
http://gallium.inria.fr/xleroy/publi/compcert-CACM.pdf
See also the introductory editorial:
Morrisett, Greg (2009), A Compilers Story, Communications of the ACM 52(7)
(July): 106,
https://www.deepdyve.com/lp/association-for-computing-machinery/technical-
perspective-a-compiler-s-story-30zQuKanaE
(accessed 25 February 2015), containing philosophical remarks on the value of
program verification.
32. MacKenzie, Donald (1992), Computers, Formal Proofs, and the Law Courts, Notices of
the American Mathematical Society 39(9) (November): 10661069.
Discusses a legal challenge to a claim that a certain computer program had been
verified. The claim was that the verification was of the relatively informal,
[De Millo et al., 1979]-style variety of proof; the challenge was that a formal,
mathematical proof was necessary..
33. MacKenzie, Donald (2001), Mechanizing Proof: Computing, Risk, and Trust (Cambridge,
MA: MIT Press).
Ch. 6
(Social Processes and Category Mistakes) concerns the [De Millo et al., 1979]-
[Fetzer, 1988] controversy.
For a review, see:
Hayes, Brian (2002), The Search for Rigor, American Scientist 90 (July-August):
382384,
http://www.americanscientist.org/bookshelf/pub/the-search-for-rigor
34. Meyer, Bertrand (1985), On Formalism in Specifications, IEEE Software 2(1)
(January): 626; reprinted in [Colburn et al., 1993, 155189].
35. Mili, Ali; Desharnais, Jules; & Gagne, Jean Raymond (1986), Formal Models of
Stepwise Refinement of Programs, ACM Computing Surveys 18(3) (September): 231
276.
1 contains a formal presentation of program correctness.

36. Myers, Ware (ed.) (1986), Can Software for the Strategic Defense Initiative Ever Be
Error-Free?, IEEE Computer (November): 6167.
37. Naur, Peter (1982), Formalization in Program Development, BIT 22: 437453;
reprinted in [Colburn et al., 1993, 191210].
38. Naur, Peter (1992), The Place of Strictly Defined Notation in Human Insight, in
Computing: A Human Activity (Addison-Wesley); reprinted in [Colburn et al., 1993, 261
274].
39. Nelson, David A. (1992), Deductive Program Verification (A Practitioners
Commentary), Minds and Machines 2: 283307.
40. Nelson, David A. (1993), Review of Boyer & Moores A Computational Logic Handbook
and Moores Special Issue on System Verification (Journal of Automated Reasoning), in
Minds and Machines 4: 93101.
41. Neumann, Peter (1996, July), Using Formal Methods to Reduce Risks,
42. Olson, Steve (1984, January/February), Sage of Software, Science: 7580.
43. Pollack, Andrew (1999, May 3), For Coders, a Code of Conduct: 2000 Problem Tests
Professionalism of Programmers, The New York Times: C1, C12;
http://tinyurl.com/Pollack1999.7 (accessed 14 November 2012).
44. Scherlis, William L.; & Scott, Dana S. (1983), First Steps towards Inferential
Programming, in R.E.A. Mason (ed.), Information Processing 83 (JFIP and Elsevier
Science Publishers): 199212; reprinted in [Colburn et al., 1993, 99133].
45. Shustek, Len (ed.), An Interview with C.A.R. Hoare, Communications of the ACM
52(3) (March): 3841.
46. Stewart, Ian (2007), The Future of Proof, Prospect (22 February).
On the nature of mathematical proof, written for a general audience.
47. Tam, Wing C. (1992), Teaching Loop Invariants to Beginners by Examples, SIGCSE
Bulletin 24(1) (March): 9296.
How to use program-verification techniques to develop programs. Compare to
Arnow 1994, above.
48. Tedre, Matti (2007), Know Your Discipline: Teaching the Philosophy of Computer
Science, Journal of Information Technology Educaiton 6:105122.
. . . one cannot formally prove either that an engineered product has the intended
qualities or that an engineered product will not fail. . . (p. 108)
49. Turing, Alan M. (1949), Checking a Large Routine, in Report of a Conference
on High Speed Automatic Calculating Machines (Cambridge, UK: University
Mathematics Lab): 6769; reprinted in Morris, F.L.; & Jones, C.B. (1984), An Early
Program Proof by Alan Turing, Annals of the History of Computing 6(2) (April): 139
143.
7 http://www.nytimes.com/1999/05/03/business/for-coders-a-code-of-conduct-2000-problem-tests-
professionalism-of-programmers.html
This paper. . . is remarkable in many respects. The three. . . pages of text contain
an excellent motivation by analogy, a proof of a program with two nested loops,
and an indication of a general proof method very like hat of Floyd [1967] (Morris
& Jones 1984: 139).
50. Verity, John W. (1985), Bridging the Software Gap, Datamation (15 February): 8488.
Contains interviews with Dijkstra, Hoare, and Gries, and includes a copy of
Dijkstras essay, Fruits of Misunderstanding (EWD-854, also online at
https://www.cs.utexas.edu/users/EWD/transcriptions/EWD08xx/EWD854.html,
accessed 16 February 2015).
51. For interesting contrasting views of the relationship between computer programs and
mathematical proofs, see:
(a) Suber, Peter (1997), Formal Systems and Machines: An Isomorphism,

http://www.earlham.edu/peters/courses/logsys/machines.htm
(b) Thurston, William P. (1994), On Proof and Progress in Mathematics, Bulletin of
the American Mathematical Society 30(2) (April): 161177.
Chapter 17
How Do Programs Relate to the

World?
Version of 31 August 2015
Today, computing scientists face their own version of the mind-

body problem: how can vitual software interact with the real world?
[Wadler, 1997, 240]
Figure 17.1: 1985,

c Universal Press Syndicate
613
614 CHAPTER 17. HOW DO PROGRAMS RELATE TO THE WORLD?
17.1 Required Reading:

Smith, Brian Cantwell (1985), Limits of Correctness in Computers, ACM SIGCAS
Computers and Society 14-15 (1-4) (January): 1826.
Also published as:
Technical Report CSLI-85-36 (Stanford, CA: Center for the Study of Language and
Information)
Reprinted in:
Dunlop, Charles; & Kling, Rob (eds.), Computerization and Controversy (San
Diego: Academic Press, 1991): 632646.
Colburn, Timothy R.; Fetzer, James H.; & Rankin, Terry L. (eds.) (1993),
Program Verification: Fundamental Issues in Computer Science (Dordrecht,
Holland: Kluwer Academic Publishers): 275293.
Preprints online at:
http://eliza.newhaven.edu/ethics/Resources/14.Reliability-
Responsibility/LimitsOfCorrectness.pdf (accessed 3 March 2015)
http://sdg.csail.mit.edu/6.894/dnjPapers/p18-smith.pdf (accessed 4 March
2015).
17.2 Introduction
In this chapter, we will examine the relationship between a program and the world
in which the program is executed. We have touched on this topic before, primarily
in 10.5.1 where we looked at the question whether a program designed to create
Hollandaise sauce could succeed in a world in which the environmental conditions
were not conducive to creating Hollandaise sauce (yet the program might be provably
correct and even execute flawlessly), and in other places (3.10, 3.13, 7.10, 8.15,
16.6). Because a program is a piece of syntax (roughly, a collection of formal marks or
symbols, grammatically organized), and because semantics is the study of the relation
between syntax and the world, we are interested in the semantics of a program.
Our focus in this chapter will be an important paper, written in 1985, by Brian
Cantwell Smith: Limits of Correctness in Computers [Smith, 1985]. Indeed, I
consider this to be one of the three central papers in the philosophy of computer science,
the others being [Turing, 1936] and [Turing, 1950].
17.3 Background
As we saw in the previous chapter, program verification is the subdiscipline of
computer science that applies logic to a computer program in order to prove that it
will do what it is supposed to do. One way of accomplishing this is to annotate each
statement S of a program with pre- and post-conditions:
{P}S{Q}
where:
1. pre-condition P is a declarative proposition describing the input and the state of
the computer before executing statement or instruction S,
2. post-condition Q is a declarative proposition describing the output and the state
of the computer after executing S, and
3. and the braces are comment delimiters.
It is to be understood as follows:
If P is true, and if S is executed, then Q is (or will be, or should be) true.
One objection to program verification is that a program can be proven correct yet
not do what you intend. One reason, as we saw in the last chapter, might be that
the computer on which the program is run might fail physically. That is, the computer
system might fail at the hardware level (this was one of the points of [Fetzer, 1988]).
A second reason, as we saw in 10.5.1, might be that the world is inhospitable.
There are actually two ways in which this latter problem might arise: There might
be a physical problem, not with the computer or with the environment, but with the
connection between them: At a simple level, the cables connecting the computer to the
world (say, to a printer) might be flawed. Or the world itself might not provide the
correct conditions for the intended outcome (it might be too hot, or too cold, and so
on).
A third reason is related to the possible hyper-computability of interactive
programs, as we discussed in 11.3: Such programs cannot be verified, because they
might depend on the non-verifiable behavior of an oracle or human user.1
What does correct mean in this context? Does it mean doing what was
intended? If so, whose intent counts? Here is Smith on this question:
What does correct mean, anyway? Suppose the people want peace, and
the President thinks that means having a strong defense, and the Defense
department thinks that means having nuclear weapons systems, and the
weapons designers request control systems to monitor radar signals, and
the computer companies are asked to respond to six particular kinds of
radar pattern, and the engineers are told to build signal amplifiers with
certain circuit characteristics, and the technician is told to write a program
to respond to the difference between a two-volt and a four-volt signal on a
particular incoming wire. If being correct means doing what was intended,
whose intent matters? The technicians? Or what, with twenty years of
historical detachment, we would say should have been intended?
According to Smith, the cause of these problems lies not in the relation of programs
to the world, but in the relation of models to the world.
17.4 Models
What is a model? According to Smith, to design a computer system to solve a real-
world problem, we must do two things:
1. Create a model of the real-world problem.
2. Represent the model in the computer.
The model that we create has no choice but to be delimited, that is, it must be
abstractit must omit some details of the real-world situation. Abstraction, as we
saw in Ch. 14, is the opposite of implementation. It is the removal of irrelevant
implementation details.
Why must any real-world information be removed? There are many reasons, one
of which is that it is methodologically easier to study a phenomenon by simplifying it,
coming to understand the simplified version, and then adding some complexities back
in, little by little. Another reason is related to the nature of maps. You cant use, or even
have, a map of Florida that is the size of Florida. Such a map might be thought to be
more useful than a smaller, more manageable one, in that it would be able to show all
the detail of Florida itself. But its lack of manageability is precisely the problem.2 What
1 However, a possible way to avoid this latter problem is due to Amir Pneulis use of temporal logic to
analyze reactive programs [Hoffmann, 2010].

2 [Carroll, 1893, Ch. 11], online at
http://etc.usf.edu/lit2go/211/sylvie-and-bruno-concluded/4652/chapter-11-the-man-in-the-moon/
17.4. MODELS 617
about eating your cake and having it, too? Suppose we omit an incomplete and separate
model of the world altogether and use the world as its own model, as [Brooks, 1991]
suggests. Perhaps we can use the real world as a representation of itself: Robots dont
need a map showing where there is a wall; if it bumps into one, it will know that its
there.
In any case, the first step in solving a problem is to create an abstract, that is,
simplified, model of it. For example, Figure 17.2 is a picture showing (in a 2-
dimensional way, of course!) a 3-dimensional, real-world house and a 2-dimensional
model of it.
Figure 17.2: Right: 3D, real-world house; Left: 2D model of it
The second step is to use logical wffs or object-oriented data structures to represent,
not the real-world situation, but the model. Figure 17.3 adds to the house and the
houses model a computer representation of the model.
Figure 17.3: Right: 3D real-world house; Middle: 2D model of 3D real-world house;

Left: computer model of 2D model
(Note to readers of the draft of this chapter: I am not an expert on illustrations.

The intention of these figures is to show a 3D picture of a house; a 2D drawing of
(accessed 4 March 2015), [Royce, 1900, Vol. 1, esp. pp. 507ff], [Rosenblueth and Wiener, 1945, 320],
[Rapaport, 1978, 5], [Borges, 1981, 234], [Eco, 1982].
it, preferably in the same colors!; and a computer representation of the 2D drawing.
Please use your imagination to create a better set of figures. Smiths original is in
Figure 17.4.)
Figure 17.4: From [Colburn et al., 1993, 283]
His point is that computers only deal with their representations of these abstract
models of the real world. They are twice removed from reality.
Is that necessarily the case? Cant we skip the intermediate, abstract model, and
directly represent the real-world situation in the computer? Perhaps, but this wont help
us avoid the problem of partiality (or abstraction, or idealization, or simplification)..
All models are necessarily partial. If they werent, there would be too much
complexity, and we would be unable to act. The only rational way to deal with
(real-world) complexity is to analyze it, that is, to simplify it, that is, to deal with
a partial (abstract) representation or model of it. (We have already discussed the
role of computer science in managing complexity; see Chs. 3, 4, and 15. See also
[Simon, 1962, Simon, 1996b].) We are condemned to do this whenever we humans
must act or make decisions: If we were to hold off on acting or making a decision
until we had complete and fully accurate information about whatever situation we were
in, we would either be paralyzed into inaction or else the real world would change
before we had a chance to complete our reasoning. (As Alan Saunders (and, later, John
Lennon) said, Life is what happens to us while we are making other plans.)3 This
is the problem that [Simon, 1996b] recognized when said that we must always reason
3 http://quoteinvestigator.com/2012/05/06/other-plans/
17.5. SUMMARY 619
with uncertain and incomplete (even noisy) information: We must satisfice. GET
EXACT CITATION. And this holds for computation as well as thinking.
But action is not abstract: You and the computer must act in the complex, real
world. Therefore, such real-world action must be based on partial models of the real
world, that is, on incomplete and noisy information. Moreover, there is no guarantee
that the models are correct.
Action can help: It can provide feedback to the computer system, so that the system
wont be isolated from the real world. Recall the blocks-world program that didnt
know that it had dropped a block, but blindly continued executing its program to
put the block on another (10.5.1). If it had had some sensory device that would have
let it know that it no longer was holding the block that it was supposed to move, and
if the program had had some kind of error-handling procedure in it, then it might have
worked much better (it might have worked as intended).
The problem, as Smith sees it, is that mathematical model theory only discusses
the relation between the model and a description of the model. It does not discuss the
relation between the model and the world. A model is like eyeglasses for the computer,
through which it sees the world. The model is the world as the computer sees it.
(The philosopher Immanual Kant said that the same thing is true about us: Our
concepts are like eyeglasses that distort reality; our only knowledge of reality is filtered
through our concepts, and we have no way of knowing how things really are in
themselves, unfiltered through our concepts (as illustrated in Figure 17.5).
Figure 17.4 is Smiths version of this picture. The problem is that we have to act in
W on the basis of M.
Similarly, to prove a program correct, we need both (a) a specification (a model of
the real-world problem) that says (declaratively) what the computer systems should
do and (b) a program (a computer model of the specification model) that says
(procedurally) how to accomplish this. A correctness proof, then, is a proof that any
system that obeys the program will satisfy the specification. But this is a proof that two
descriptions are compatible. The program is proved correct relative to the specification.
So, a better term might be proof of relative consistency. Suppose the proof fails to
show correctness; what does this mean? It means either that the program is wrong,
or that the specification is wrong (or both). And, indeed, often we need to adjust both
specification and program.
The real problems lie in the model-world relation, which correctness does not
address. This is one of the morals of Clelands and Fetzers claims. That is, programs
can fail because the models can fail to correspond to the real world in appropriate
ways. But that italicized clause is crucial, because all models abstract from the real
world, but each of them do so in different ways.
17.5 Summary
To sum up: Real-world tasks are complex. Models abstract from this complexity,
so they can never match the rich complexity of the world. Computers see the world
through models of these models (but so do people!). Would an automated system
designed to decide quickly how to respond to an emergency make you feel uneasy?
Figure 17.5: A cognitive agent looking at a real-world object that the agent categorizes
as a house. Light reflecting off the house (the thing-in-itself) enters the agents eyes
and the resulting neural signals are filtered through the agents mental concepts,
producing a mental image of the house. The mental image may, or may not, be a
perfect representation of the house, but the agent has no way to directly compare the
mental image with the real house, independent of the agents concepts.
But so should a human who has to make that same decision. And they should both
make you uneasy for the same reason: They have to reason and act on the basis
of partial (incomplete) information. (Well discuss this in the next chapter!) Such
reasoning cannot be proved correct (and simulation only tests the computer-model
relation, not the model-world relation). So, empirical reliability must supplement
program verification. Therefore, we must embed the computer in the real world.

1. Probably should incorporate my Covey Award paper, online at
http://www.cse.buffalo.edu/rapaport/Papers/covey-short.pdf
(a) On that papers discussion of input-output:
A terminating computation is one in which all the processes

terminate; its output is the values left in the shared memory.
[Denning and Martell, 2015, 155, my emphases]
Note that this output need not be reported to a user; its just left there. By
contrast,
A nonterminating computation is one in which at least some
of the processes execute in repeating cycles; its outputs are
sequences of values emitted from designated proesses in the set.
[Denning and Martell, 2015, 155, my emphases]
Emitted values sounds like values that are not necessarily stored
anywhere, but that are reported to the outside world.
2. The debate between Rescorla-Cleland and Piccinini as to whether computation
concerns the internal, syntactic manipulation of symbols or the external,
semantic interpretation of them is reminiscent of Smiths gap. As
[Rescorla, 2007, 265] observes, we need a computable theory of the semantic
interpretation function, but, as Smith observes, we dont (cant?) have one, for
reasons akin to the CTCT problem.
3. Re the opening motto:
How does a program interact with the world? Via the process (which is a
physical/causal entity (or event?) in the world. So, how does the program interact
with the process? Via the compiler? No; thats just a first step, translating the
program into a the machine language that the machine understands. The loader?
Yes, because thats what copies/transduces? the machine-language program into
the memory; it sets the switches.
4. In 17.4, cite McMullin on idealization, and say more about idealization. Also,
mention that using the world as its own model is how robotic vacuums work; say
more about Brooks.
5. Possibly include a short discussion of Wigner? At least cite in Further Readings
as follows:
Wigner, Eugene (1960), The Unreasonable Effectiveness of Mathematics
in the Natural Sciences,
Communications in Pure and Applied Mathematics 13(1) (February),
https://www.dartmouth.edu/matc/MathDrama/reading/Wigner.html,
http://www.maths.ed.ac.uk/aar/papers/wigner.pdf (both accessed 7 April
2015)
Also see reply by a computer scientist in [Hamming, 1980].
6. Any machine is a prisoner of its input and output domains. SS,

our example [symbol] system, presents an abstract machine-centered
view, so that the external world is pretty much what is seen by the
machine. But this is deceptive. Machines live in the real world
and have only a limited contact with it. Any machine, no matter
how universal, that has no ears (so to speak) will not hear; that
has no wings, will not fly. Thus universality will be relative to the
input and output channels. Such a consideration is alleviated in
theoretical discussions by the simple expedient of considering only
abstract inputs and outputs. It can be alleviagted in the real world
by providing transducers that encode from one input-output channel
to another. Thus, being able to produce any function between two
given domains [that is, being universal in Newells sense] permits4
inducing any function between two other domains if the domains are
hooked up appropriately. [Footnote in original omitted here.] But this
interface limit must always be remembered. [Newell, 1980, 148; my
italics]
I wont describe SS here, but you can think of it as a slightly more elaborate
kind of Turing machine. So Newells first point is that Turing machines
only see the external world from the point of view of the marks on its tape.
The italicized sentence could have been taken straight out of [Smith, 1985]!
Smith, however, might have replaced and with but, to emphasize the
twice-removed-from-reality nature of the limited contact.
The sentence that follows the italicized one should be compared to
Piccininis remarks in [Piccinini, 2011, 741, n. 11] (see Ch. 7, NOTES12).
By universality, Newell means the ability to produce an arbitrary input-
output function (p. 147). His point in this sentence is that, in order for a
Turing machine to deal with the real world, it has to overcome its reliance
on the just the marks on its tape. As the next sentence suggests, we dont
have to worry about that if were not dealing with the real world. If,
for example, we wanted to write a toy computer program that deals with
houses, its enough to let it deal with computer representations of houses,
as in Fig. 17.3.
The transducers do the hard work of encoding the real world into the syntax
of the Turing-machine tapes marks and decoding those marks back into
real-world entities.
The penultimate sentence reminds us that the input-output codings must
themselves be computable. And the last sentence takes us back to
Smith, suggesting even another reason to be cautious about the limits of
correctness: Not only is the computer separated from the real world in the
4I think requires would have been a more appropriate word!
way that Smith notes, but the links that it does have are indirect and subject
to their own translation limitations.
7. [Thagard, 1984, 82], citing [Zeigler, 1976], also notes that computers are twice
removed from reality, because a computer simulates a model which models a
real system.
8. Smiths discussion at the beginning of his paper, about the computer that
confused the rising moon with a Soviet nuclear attack because this lunar
reflection hadnt been predicted by the systems designers (p. 18), is reminiscent
of Clelands hollandaise-sauce on the moon. If Clelands recipe assumes that it is
being executed on Earth, then it cant be faulted for failing on the moon; someone
using it on the moon is the guilty culprit. But if Clelands recipe advertises itself
as being universally applicable, then it is at fault. One can debug the recipe
by adding a precondition that it must be executed on Earth, or by expanding it to
clarify how to modify it if it is used on the moon (as in Denver cake mix recipes).
Similarly, the early-warning system computer must be debugged by clarifying an
unstated assumption (perhaps that any signal of a certain type must be a nuclear
attack).
9. In general, it is impossible, when writing specifications, to include

everything that you want: legal contracts, and other humanly
interpretable specifications, are always stated within a background
of common sense, to cover the myriad unstated and unstatable
assumptions assumed to hold in force. (Current computer programs,
alas, have no common sense. . . .) [Smith, 1985, 23]
This is both a statement of the gap, as well as the insight behind McCarthys
notion of circumscription.
10. . . . as well as modelling the artifact itself, you have to model the
relevant part of the world in which it will be embedded. [Smith, 1985,
24, col. 1]
This is essentially Shapiros point that I discuss in [Rapaport, 2015]. This is one
way to handle Clelands hollandaise-sauce problem.
Indeed, I should probably incorporate some of that Covey-award essay in this
chapter!
11. The systems that land airplanes are hybridscombinations of

computers and peopleexactly because the unforeseeable happens,
and because what happens is in part the result of human action,
requiring human interpretation. [Smith, 1985, 24, col. 2]
First, note that this anticipates the points about extended cognition made by
Hutchins, and Clark & Chalmers (give citations). Second, whether this is a
necessary truth is quite another matter. Smith wrote this long before the advent
of self-driving cars. Now, it remains to be seen, even as of this writing (2015),
whether self-driving cars will continue to need human intervention (remember:
self-driving elevators dont need very much of it!). Nevertheless, such hybrid
or extended systems will probably remain a reality.
12. Reasoning is based on partial models. Which means it cannot be

guaranteed to be correct. [Smith, 1985, 24, col. 2]
This is what lies behind Simons notion of satisficing and bounded rationality.
13. [Wegner, 1997, 82] points out that interactive computer programs are grounded
in. . . external reality rather than being autistic.
14. Note that the issue of hypercomputation, especially interactive computation, is

directly relevant to this chapter. After all, most interesting computer programs
have to interact with the real world.
15. Human fallibility means some of the more subtle, dangerous bugs
turn out to be errors in design; the code faithfully implements the
intended design, but the design fails to correctly handle a particular
rare scenario. [Newcombe et al., 2015, 67]
This is a nice statement of Smiths Gap.
16. A specification is an abstraction. It should describe the important

aspects and omit the unimportant ones. Abstraction is an art that is
learned only through practice. . . . [A] specification of what a piece of
code does should describe everything one needs to know to use the
code. It should never be necessary to read the code to find out what it
does. [Lamport, 2015, 39]
Smiths gap is due, in part, to the fact that specs are abstractions. How does one
know if something that has been omitted from the spec is important or not? This
is why it is an art and why theres no guarantee that the model is correct (matches
reality).
17. Smiths main point is that program verification proves at most that a program
satisfies its specification. But theres no way to formally prove that a
specification is correct, that it matches reality [Smith, 1985, 6]. Both the
specification and the program are models. The former is a model of the world;
the latter is a model of the former (hence its double remove from reality).
18. [Cleland, 2004, 220] also talks about something very much like Smiths gap:
There is a causal gap between the actions specified by a quotidian procedure
and the physical processes that make it reliable. But she finds a positive aspect
to it: the gap between action and process is actually a perk. Because of it, one
can make a great Hollandaise sacue without knowing any chemistry. One need
only know how to perform the action specified by the recipe. Nature takes care
of the rest.
19. Nice question for further discussion (thanks to Albert Goldfain, personal
communication, 3 April 2007):
An artist must paint not what he sees in nature, but what is there.
To do so he must invent symbols, which, if properly used, make his
work seem even more real than what is in front of him. (Charles E.
Burchfield,
https://www.burchfieldpenney.org/collection/charles-e-
burchfield/biography/)
Change artist to programmer and paint to program: Programmers must

program not what they see in nature, but what is there. To do so they must
invent symbols, which, if properly used, make their work seem even more real
than what is in front of them. Is the artists task different from the scientists
or the programmers? Can programs (or paintings, or scientific theories) end up
seeming more real to us than the things that they are models of? Is it easier to
understand the behavior of the process of a program that models a hurricane (for
example) that to understand the real hurricane itself?
20. [Mahoney and Haigh (ed.), 2011, 104105] describes the software life cycle,
which begins with a real world (physical) system, analyzed into (an abstract)
computational model of that system, then implemented in a computer
program, and ultimately resulting in a (once again physical) implementation as a
machine. (Compare Marrs three levels of analysis!). He notes that, despite the
fact that We are on firmest theoretical ground at the end of this cycle with the
machine as operational model of [the original physical] system, the problems
of air traffic control systems, of national weather systems, of airline booking
systems, all lie at the beginning of the cycle, where a real-world system must
be transformed into a computational model. That is where Smiths Gap is found.
That is where software engineering is not about software, indeed where it may
not be about engineering at all.
21. [Mahoney and Haigh (ed.), 2011, 117] goes on to say that modeling the real
world in the computer involves an understanding not of computers but of the
real-world situation in question. . . . That is not what one learns in studying
computer science; that is not what computer science is about. There are two
observations to make about this: First, this makes Smiths Gap even wider. Note
that, to model the real world for a computer, the modeler needs to understand
both the real world and the computer. Second, it makes a strong recommendation
about undergraduate education in computer science!
22. Smiths Gap may be related to at least one view of the nature of computer science
(the issues we looked at in Ch. 3): [Mahoney and Haigh (ed.), 2011, 183184]
observes that even if (or to the extent that) computer science is not natural science
but a science of the artificial (especially given the fact that computers work on
partial models (hence artificial) models of the world, those two worlds [natural
and artificial] meet in the physical computer, which enacts a program in the
world.

1. The New York Times report on the moon-missile mistake:
Canadian Is Praised over Missile Scare, New York Times (23 December 1960).
2. For more examples like Smiths moon-mistaken-as-missile, only with more dire
consequences, see:
Neumann, Peter G. (1993), Modeling and Simulation, Communications of the
ACM 36(4) (June): 124.
3. Cerf, Vinton G. (2013), Whats a Robot?, Communications of the ACM 56(1) (January):
7,
http://www.cs.grinnell.edu/davisjan/csc/105/2013S/articles/CACM-reliability.pdf
Suggests that we should treat as robots any programs that can have real-
world. . . effect. . . . [T]hose of us who live in and participate in the creating of
software-based universes might wisely give thought to the potential impact that
our software might have on the real world, a thought that echoes [Smith, 1985].
4. Dewar, Robert (2009), CS Education in the U.S.: Heading in the Wrong Direction?,
5. Fetzer, James H. (1999), The Role of Models in Computer Science, The Monist 82(1):
2036.
A clearly written summary and critique of [Smith, 1985], arguing that there are
more than merely the two models that Smith considers (a specification as a model
of the world and a program as a model of the specification).
6. Hayes, Brian (2007), Calculating the Weather, American Scientist 95(3) (May-June):
271273.
Another interesting article on computational modeling (and what can go wrong).
7. Jackson, Michael (2003), Why Software Writing Is
Difficult and Will Remain So, Information Processing Letters 88: 1325, preprint at
http://users.mct.open.ac.uk/mj665/turski07.pdf (accessed 18 April 2015).
Discusses what I call Smiths gap: the interplay between the formal world of the
computer and its programming language with the informal world where the problem
to be solved is located (from the abstract, p. 13).
8. Morrisett, Greg (2009), A Compilers Story, Communications of the ACM 52(7) (July):
106.
9. van Fraassen, Bas C. (2006), Representation: The Problem for Structuralism,
Philosophy of Science 73 (December): 536547,
http://www.princeton.edu/fraassen/abstract/docs-publd/PSA04 Structure.pdf
(accessed 18 April 2015)
Discusses the relation between formal, mathematical models of reality and the
reality of which they are models, arguing that what I call Smiths gap presents
difficulties for scientific realism (which we discussed in Ch. 4).
10. On the nature of abstraction, see:
(a) Kramer, Jeff (2007), Is Abstraction the Key to Computing?, Communications of

the ACM 50(4) (April): 3642.
(b) Rapaport, William J. (1999), Implementation Is Semantic Interpretation, The
Monist 82 (1): 109130.
(c) Rapaport, William J. (2005), Implementation Is Semantic Interpretation: Further
Thoughts, Journal of Experimental and Theoretical Artificial Intelligence 17(4)
(December): 385417.
Chapter 18
Computer Ethics I:
Are There Decisions
Computers Should Never
Make?
The artificial pericardium will detect and treat a heart attack before any
symptoms appear. Its twenty years or more out there, [John] Rogers
said. But we can see a pathwayits not science fiction. Bit by bit,
our cells and tissues are becoming just another brand of hardware to be
upgraded and refined. I asked him whether eventually electronic parts
would let us live forever, and whether he thought this would come as a
relief, offer evidence that wed lost our souls and become robots, or both.
Thats a good thing to think about, and people should think about it, he
said But Im just an engineer, basically. [Tingley, 2013, 80]
629
630 CHAPTER 18. COMPUTER ETHICS I: DECISIONS

1. Required:
Moor, James H. (1979), Are There Decisions Computers Should Never Make?,
Nature and System 1: 217229,
http://www.researchgate.net/publication/242529825 Are there decisions computers should never make
(a) Friedman, Batya; & Kahn, Peter H., Jr. (1992), People Are Responsible,
Computers Are Not, excerpt from their Human Agency and Responsible
Computing: Implications for Computer System Design, Journal of Systems and
Software (1992): 714; excerpt reprinted in M. David Ermann, Mary B. Williams,
& Michele S. Shauf (eds.) (1997), Computers, Ethics, and Society, Second Edition
(New York: Oxford University Press): 303314.
Can be read as raising an objection to Moor 1979, esp. the section Delegating
Decision Making to Computational Systems (pp. 306307 in the Ermann et
al. 1997 reprint).
(b) Johnson, George (2002), To Err Is Human, New York Times (14 July),
http://www.nytimes.com/2002/07/14/weekinreview/deus-ex-machina-to-err-is-
human.html (accessed 14 November 2012).
An op-ed piece that provides an interesting, real-life, case study of Moors
problem and a possible counterexample to Friedman & Kahn 1992.
18.2 Introduction
Ethics is a branch of philosophy, so computer ethics should be a branch of
the philosophy of computer science. But computer ethics is a large and long-
standing discipline in its own right (for good introductions, see [Moor, 1985],
[Anderson and Anderson, 2006]). We will focus on two topics that I think are central
to the philosophy of computer science but, perhaps until recently, have not been the
focus of most discussions of computer ethics. This chapter will consider whether there
are decisions that computers should never make, and, after discussing the philosophy
of AI (whether we can build artificial intelligences, Ch. 19), Chapter 20 will consider
whether we should build them.
18.3 Do Computers Make Decisions?

Any time that a computer solves a problem, it is making a decision. And it certainly
seems that some computers can make rational decisions for us. Of course, it is not just
a physical computer that makes a decision; it is a computer program being executed
by a computer that makes the decision, but we will continue to speak as if it is the
computer that decides.
Presumably, if the decision is made by a computer that is following an algorithm
that does not involve any random or interactive procedure, then that decision is a purely
rational one. By rational, I dont necessarily mean that it is a purely logical decision.
It may, for instance, involve empirical data. It may even be illogical, but, as long as
there is an algorithm that can be studied to see how it works, or as long as the program
can explain how it came to its decision, I will consider it to be rational.
Now, if you believe in the possibility and value of program verification, it would
certainly be nice if such a decision-making program could be verified. But, given that
we may not be able to verify such programs, should we trust them?
And what is a decision? There are at least two senses of decision:
1. A decision could be made as the result of an arbitrary choice, such as flipping a

coin: heads, well go out to a movie; tails, well stay home and watch TV.
2. A decision could be made as the result of investigating the pros and cons of
various alternatives, evaluating these pros and cons, and then making a decision
based on this evaluation.
Computers can easily make the first kind of decision for us. Can they make the second
kind? The answer is, surely, yes: Computers can play games of strategy, such as
chess and checkers, and they can play them so well that they can beat the (human)
world champions. Such games involve choices among alternative moves that must be
evaluated, with, one hopes, the best (or least worst) choice being made. Computers can
do this.
Can computers be completely rational? As I suggested above, insofar as what
they do is algorithmic, their behavior is rational. But are there limits to a computers
rationality? Model-theoretic limits in the sense of [Smith, 1985] (as we discussed in
Ch. 17) affect us humans, too. So, if there are limits to our rationality, it shouldnt
surprise us that the same limits might apply to computers. But, at worst, this might
mean that they are no less rational than we are. Well return to this later in this section.
Another question about decisions is whether the process of decision making must
be conscious. Its one thing to go into an unconscious trance and emerge with a
decision. Its another to consciously think about an issue and then come up with a
decision that is clearly and rationally based on the thinking process. If consciousness is
a necessary condition of making a rational decision, and if computers are not conscious
(as many have argued), then they cannot make such rational decisions.
But there are many cases in which we humans make unconscious decisions (without
going into trances!). For instance, we make many decisions about what to eat or
wear without much conscious thought (of course, we also make many such decisions
consciously, with great thought). We also make decisions quickly when we are driving;
often, we have no time to stop and think, but must react (that is, make decisions)
instinctively.
Cognitive scientists tell us that there are two decision-making systems in our
brains: There is an unconscious, instinctive, intuitive, fast, tacit system (called
System 1), and there is a conscious, rational, slow, explicit system (called
System 2).1 The former helps us avoid immediate dangers; the latter helps us make
logical decisions. Unfortunately, sometimes they give us different answers to the same
question.
Arguably, a computer that can tell us how it made a decison, even if it is
unconscious, is better than one that cannot.
But can a computer really make a decision? I suggested above that it is really the
computer program being executed on a computer that is making the decision. And
programs are written by human programmers. So, perhaps it is only humans, using
computers, that make such decisions. Although that is surely sometimes the case,
there are cases in which humans do delegate decision-making powers to computers,
such as in traffic control. Typically, this happens in cases where large amounts of
data are involved in making the decision or in which decisions must be made quickly
and automatically. And, of course, given that one of the goals of computer science
is to determine what real-world tasks are computable, finding out which decisions are
computable is an aspect of that.
In any case, humans might so delegate such power. So, the ultimate question is:
What areas of our lives should be computer-controlled, and what areas should be left
to human control?
Are there decisions that non-human computers could not make as well as humans?
For instance, there might be situations in which there are sensory limitations that
prevent computer decisions from being fully rational. Or there might be situations
in which a computer could not be empathetic.
To answer this question, we need to distinguish between what is the case and what
could be the case. We could try to argue that there are some things that are in principle
1 For other adjectives characterizing these two systems, see
http://en.wikipedia.org/wiki/Dual process theory#Systems (accessed 20 April 2015).

For further discussion, see [Kahneman, 2011]; for further citations, see
http://www.cse.buffalo.edu/rapaport/575/rules-connections.html#uncs-cognition
18.3. DO COMPUTERS MAKE DECISIONS? 633
impossible for computers to do. Except for computationally impossible tasks (such as
deciding whether an arbitrary computer program halts), this might be hard to do. But
we should worry about the possible future now, so that we can be prepared for it if it
happens.
Whether there are, now, decisions that a computer could not make as well as a
human is an empirical question. It is capable of investigation, and, currently, the answer
is unknown. Many, if not most, of the objections to the limitations of computers are
best viewed as research problems: If someone says that computers cant do X, we
should try to make ones that do X.
Another version of our question is this: Are there decisions that non-human
computers could make better than humans? Well come back to this in the next section.
But this raises another question: How could computer decision-making competence be
judged? The best answer is: in the same way that human decision-making competence
is judged, namely, by means of its decision-making record and its justifications for its
decisions.
Why do justifications matter? After all, if a computer constantly bests humans at
some decision-making task, why does it matter how it does it? (Maybe we would be
better off not knowing! For a science-fiction treatment of this kind of issue, though not
in the context of computers, see Arthur C. Clarkes Childhoods End [Clarke, 1953].)
Presumably, however, decision-making computers do need to be accountable for their
decisions, and knowing their justifications helps this accounting. The justifications,
of course, need not be the same as human justifications. For one thing, human
justifications might be wrong or illogical.2
This suggests that the question Are there decisions that a computer should not
make? should really have nothing to do with computers! The question should really
be: Are there decisions that should not be made on a rational basis?
Again: Are there decisions that a computer should never make? The computer
scientist Joseph Weizenbaum has argued that, even if a computer could make decisions
as well as, or even better than, a human, they shouldnt, especially if their reasons differ
from ours [Weizenbaum, 1976].
And James Moor points out that, possibly, computers shouldnt have the power
to make (certain) decisions, even if they have the competence to do so (at least as
well as, if not better than, humans) [Moor, 1979]. But, if they have the competence,
why shouldnt they have the power? For instance, suppose a very superstitious group
of individuals makes poor medical decisions based entirely on their superstitions;
shouldnt a modern physicians outsider medicine take precedence? And does the
fact that computers are immune to human diseases mean that they lack the empathy to
recommend treatments to humans?
[Moor, 1979] suggests that, if computers can make certain decisions at least as
well as humans, then we should let them do so, and it would then be up to us humans
to accept or reject the computers decision. After all, when we ask for the advice of
an expert in medical or legal matters, we are free to accept or reject that advice. Why
2 See, especially, [Tversky and Kahneman, 1974]. Also see [Wainer, 2007] on humans difficulty in
reasoning about probability and statistics: Ignorance of how sample size affects statistical variation has
created havoc for nearly a millennium. For other citations on humans difficulty in reasoning about
probability and statistics, see http://www.cse.buffalo.edu/rapaport/575/reasoning.html
shouldnt the same be true for computer decision making?

This is crucial: Humans should be critical thinkers. There is a fallacy called the
Appeal to Authority: Just because an authority figure says that P is true, it does not
logically follow that P is true. Although logicians sometimes warn us about this fallacy,
it is acceptable to appeal to an authority (even a computer!) as long as the final decision
is yours. You canand mustdecide whether to believe the authority or to trust the
computer.
18.4 Emergency Decisions

But what about emergencies, or other situations in which there is no time for the human
who must act to include the computers recommendation in his or her deliberations?
On July 1, 2002, a Russian airliner crashed into a cargo jet over Germany, killing
all on board, mostly students [Johnson, 2002]. The Russian airliners flight recorder
had an automatic collision-avoidance system that instructed the pilot to go higher (to
fly over the cargo jet). The human air-traffic controller told the Russian pilot to go
lower (to fly under the cargo jet).
According to [Johnson, 2002, my emphasis], Pilots tend to listen to the air traffic
controller because they trust a human being and know that a person wants to keep them
safe. But the human air-traffic controller was tired and overworked. And the collision-
avoidance computer system didnt want anything; it simply made rational judgments.
The pilot followed the humans decision, not the computers, and a tragedy occurred.
There is an interesting contrasting case. In January 2009, after an accident
involving birds that got caught in its engines, a US Airways jet landed safely on
the Hudson River in New York City, saving all on board and making a hero out of
its pilot. Yet [Langewiesche, 2009] argues that the plane, with its computerized fly
by wire system, was the real hero. In other words, the pilots heroism was due to
his willingness to accept the computers decision. (For contrasting discussions of this
complicated case, see [Haberman, 2009], [Salter, 2010].)
A paragraph deeply embedded in [Kolata, 2004] suggests that people find it difficult
to accept the rational recommendations even of other people. The article reports on
evidence that a certain popular and common surgical procedure had just been shown
to be of no benefit: Dr. Hillis said he tried to explain the evidence to patients, to
little avail. You end up reaching a level of frustration, he said. I think they have
talked to someone along the line who convinced them that this procedure will save
their life. Perhaps the fundamental issue is not whether computers should make
rational decisions or recommendations, but whether or why humans should or dont
accept rational advice!
18.5 Moral Decisions

[Moor, 1979] suggests that, although a computer should make rational decisions for us,
a computer should not decide what our basic goals and values should be. Computers
18.5. MORAL DECISIONS 635
should help us reach those goals or satisfy those values, but they should not change
them.
But why not? Computers cant be legally or morally responsible for their decisions,
because theyre not persons. At least, not yet. But what if AI succeeds? Well return to
this in Chapters 19 and 20. Note, by the way, that for many legal purposes, non-human
corporations are considered to be persons.
[Friedman and Kahn, 1997] argue that humans arebut computers are not
capable of being moral agents and, therefore, computers should be designed so that:
1. humans are not in merely mechanical roles with a diminished sense of agency,
and
2. computers dont masquerade as agents with beliefs, desires, or intentions.
Lets consider point 1: They argue that computers should be designed so that
humans do realize that they (the humans) are moral agents. But what if the computer
has a better decision-making track record than humans? [Friedman and Kahn, 1997]
offer a case study of APACHE, a computer system that can make decisions about when
to withhold life support from a patient. It is acceptable if it is used as a tool to aid
human decision makers. But human users may experience a diminished sense of
moral agency when using it, presumably because a computer is involved.
But why? Suppose APACHE is replaced by a textbook on when to withhold life
support, or by a human expert. Would either of those diminish the human decision-
makers sense of moral agency? In fact, wouldnt human decision makers be remiss
if they failed to consult experts or the writings of experts? So wouldnt they also be
remiss if they failed to consult an expert computer?
Or would humans experience this diminished sense of moral agency for the
following reason? If APACHEs decisions exhibit good performance and are more
relied on, then humans may begin to yield to its decisions. But why would that be bad?
Turning to point 2, computers should be designed so that humans do realize that
computers are not moral agents. Does this mean that computers should be designed so
that humans cant take [Dennett, 1971]s intentional stance towards them?
But what if the computer did have beliefs, desires, and intentions? AI researchers
are actively designing computers that either really have them, or else that are best
understood as if they had them.3 Would they not then be moral agents? If not, why
not? According to [Dennett, 1971], some computers cant help masquerading as
belief-desire-intention agents, because thats the best way for us to understand them.
[Friedman and Kahn, 1997] argue that we should be careful about anthropomorphic
user-interfaces, because the appearance of beliefs, desires, and intentions does not
imply that they really have them. This is a classic theme, not only in the history of AI,4
3 See, e.g., [Kumar, 1994], [Kumar, 1996], https://www.google.com/search?q=belief+desire+intention,
and https://en.wikipedia.org/wiki/Belief%E2%80%93desire%E2%80%93intention software model

4 Here, I am thinking primarily of Joseph Weizenbaums Eliza program, which, in its
most famous version, allegedly simulated a Rogerian psychotherapist. See [Weizenbaum, 1966],
[Weizenbaum, 1967], [Weizenbaum, 1976], [Shapiro and Kwasny, 1975], [Winograd, 1983, Ch. 2], and
http://www.cse.buffalo.edu/rapaport/572/S02/proj1.html
but also in literature,5 and cinema.6 And this is at the heart of the Turing test in AI, to
which we now turn.
5 E.g., [Powers, 1995] (in which a cognitive-science grad student is assigned the task of programming
a computer to pass the PhD exam in English literature; he (the grad student) falls in love with her (the
computational cognitive agent)) and [Dick, 1968] (which was the basis of the film Blade Runner).
6 E.g., Steven Spielbergs A.I. Artificial Intelligence, Spike Jonzes Her, Alex Garlands Ex Machina, et al.

1. One thing that Moor et al. dont focus on is whether certain decisions might be
OK for computers to make, while others arent.
2. Whether computers ought to make decisions for us is equivalent to whether our
decisions ought to be made algorithmically. If there is an algorithm for making
a given decision, then why not? After all, that would be the rational thing to
do. One potential problem is if the algorithm requires exponential time or is
NP-complete, or even if it merely would take longer to come up with a decision
than the time needed for action. In that case, or if there is no such algorithm,
we would have to rely on a satisficing heuristic (in the sense of an algorithm
whose output is near enough to the correct solution). Perhaps thats where
the interesting ethical issues lie.
3. Related to both of the above two concerns:
One might argue that there is no such thing as computer ethics. All questions
about the morality of using computers to do something are really questions about
the morality of using algorithms. Surely, using algorithms is being rational, and
it seems implausible to argue that we shouldnt be rational.
Hence, perhaps the ethical issues really concern the nature of different kinds of
algorithms. If the algorithm is neat (in the sense of being based on formal
logic), then, if the algorithm is correct, surely there is no moral reason not to
use it (not to be correctly rational). If the algorithm is a heuristic (perhaps
as in the case of expert systems), then there is no more or less moral reason to
use it then there is to trust a human expert. If the algorithm is scruffy (not
logical, and not necessarily based on any formal theory), then one might have
moral qualms.
4. Here is one interesting aspect of computer decision making: It is impossible
(or at least less likely) to be swayed by such things as letter-writing campaigns,
parades, etc. Such tactics were used by GM in their campaign to persuade some
communities to open new plants for their Saturn cars. See, e.g., [Russo, 1986].
5. Of course, one crucial feature of whether decisions should be made by computers
(algorithms) is what kinds of decision-making criteria and data are encoded in
the algorithm, not to mention whether the algorithm is verifiable.
6. I may have already said this above somewhere (check to see!): whether there
are decisions computers should never make could mean: whether there are
decisions that should not be made algorithmically. Another (better?) question
is: whether there are decisions that should not be made by algorithms that are so
complex that we cannot understand them (but must merely have faith in them).
7. Even if we allow computers to make (certain) decisions for us, it is still important
for us to be able to understand those decisions and to be able follow through/act
on them. When my son was first learning how to drive, I did not want him to rely
on laser cruise control, because I wanted him to know how and when to slow
down or speed up on a superhighway. Once he knew how to do that, then he
could rely on the cars computer making those decisions for him, because, if the
computers decision-making skills went awry or were unavailable (for example,
the system wont work when the windshield wipers are on, or when the car
ahead of you is dirty!), he should know how to do those things himself. (See
Halpern 2015, Zremski 2009.)
8. This may have been mentioned earlier, but clearly this issue is highly relevant
today what with all the hype about self-driving cars. Some of the same issues
arise for self-flying planes (see Markoff 2015). Keep in mind that another vehicle
that many of us use all the time was once only human-operated but is now
completely automated (and I doubt that any of us would know what to do if
it failed): elevators.

Anderson, Michael; & Anderson, Susan Leigh (2007), Machine Ethics: Creating
an Ethical Intelligent Agent, AI Magazine 28(4) (Winter): 1526.
Anderson, Michael; & Anderson, Susan Leigh (2010), Robot Be Good, Scientific
American 303(4) (October): 7277.
Aref, Hassan (2004), Recipe for an Affordable Supercomputer: Take 1,100
Apples. . . , Chronicle of Higher Education (5 March): B14.
Suggests (but does not discuss) that supercomputers might make decisions that
we could not understand:
As we construct machines that rival the mental capability of humans,
will our analytical skills atrophy? Will we come to rely too much
on the ability to do brute-force simulations in a very short time,
rather than subject problems to careful analysis? Will we run to the
computer before thinking a problem through?. . . A major challenge
for the future of humanity is whether we can also learn to master
machines that outperform us mentally.
On the question will our analytical skills atrophy?, you might enjoy the
following science-fiction story about a human who rediscovers how to do
arithmetic after all arithemetical problems are handled by computers:
Asimov, Isaac (1957), The Feeling of Power, reprinted in Clifton
Fadiman (ed.), The Mathematical Magpie (New York: Simon and
Schuster, 1962): 314, accessed 14 November 2012 from
http://www.themathlab.com/writings/short%20stories/feeling.htm
Asimov, Isaac (1950), The Evitable Conflict, Astounding Science Fiction;
reprinted in Isaac Asimov, I, Robot (Garden City, NY: Doubleday), Ch. 9, pp. 195
218, accessed 14 November 2012 from:
http://preterhuman.net/texts/literature/books by author/A/Asimov,%20Isaac%20-
%20I,%20Robot.pdf or http://tinyurl.com/AsimovIRobot
A fictional approach to Moors question.
Brachman, Ronald J. (2002), Systems that Know What Theyre Doing, IEEE
Intelligent Systems (November/December): 6771.
A leading AI researcher suggests how and why decision-making computers
should be able to explain their decisions.
Greengard, Samuel (2009), Making Automation Work, Communications of the
ACM 52(12) (December): 1819.
Todays automated systems provide enormous safety and convenience.
However, when glitches, problems, or breakdowns occur, the results can be
catastrophic.
Hafner, Katie (2012), Could a Computer Outthink This Doctor?, New York Times
(4 December): D1, D6,
http://www.nytimes.com/2012/12/04/health/quest-to-eliminate-diagnostic-
lapses.html (accessed 10 October 2015).
For diagnosticians. . . , software offers a valuable backup. Whether it can ever
replace them is another question.
For a contrasting argument, see Mearian 2013.

Halpern, Sue (2015), How Robots & Algorithms Are Taking Over, New York
Review of Books 62(6) (2 April): 24, 26, 28,
http://www.nybooks.com/articles/archives/2015/apr/02/how-robots-algorithms-
are-taking-over/
Among other things, points out that an overreliance on automation, and on
a tendency to trust computer data even in the face of contradictory physical
evidence, can be dangerous, in part because the human in the decision-
making loop might not be paying attention: over half of all airplane accidents
were the result of the mental autopilot brought on by actual autopilot.
See also Zremski 2009.
Heingartner, Douglas (2006), Maybe We Should Leave That Up to the Computer,
New York Times (18 July),
http://www.nytimes.com/2006/07/18/technology/18model.html
On whether computers can make better decisions than humans.
. . . mathematical models generally make more accurate predictions han
humans do. . . . The main reason for computers edge is their consistency
or rather humans inconsistencyin applying their knowledge.
Markoff, John (2009), Ay Robot! Scientists Worry Machines May Outsmart Man,
New York Times (26 July): 1, 4; accessed 14 November 2012 from
http://www.nytimes.com/2009/07/26/science/26robot.html
Markoff, John (2015), Planes without Pilots, New York Times (7 April): D1, D4,
http://www.nytimes.com/2015/04/07/science/planes-without-pilots.html
(accessed 10 October).
In the wake of the commercial-airline pilot who who committed suicide,
thereby killing all aboard his plane, this article asks whether pilots are
necessary at all.
See also two follow-up letters to the editor, New York Times (14 April): D3,
http://www.nytimes.com/2015/04/14/science/letters-to-the-editor.html
(accessed 10 October 2015); they point out situations in which (a) computers
should take control over from humans or (b) humans might make better
because more creative or context-dependentdecisions than computers..
Mearian, Lucas (2013), AI Found Better than Doctors at Diagnosing, Treating
Patients, Computerworld (12 February),
http://www.computerworld.com/article/2494918/healthcare-it/ai-found-better-
than-doctors-at-diagnosingtreating-patients.html
(accessed 10 October 2015)
For a contrasting argument, see Hafner 2012.
Miller, Christopher A. (guest ed.) (2004), Human-Computer Etiquette: Managing
Expectations with Intentional Agents, Communications of the ACM 47(4) (April):
3061.
Friedman & Kahn (1992) argue that programmers should not design computer
systems so that users think that the systems are intelligent. The April 2004
issue of CACM has a whole section devoted to this.
Neumann, Peter G. (1993), Modeling and Simulation, Communications of the

ACM 36(4) (June): 124.
Some cautionary examples.
OHeigeartaigh, Sean (2013), Would You Hand Over a Moral Decision to a
Machine? Why Not? Moral Outsourcing and Artificial Intelligence, Practical
Ethics (9 August),
http://blog.practicalethics.ox.ac.uk/2013/08/would-you-hand-over-a-moral-
decision-to-a-machine-why-not-moral-outsourcing-and-artificial-intelligence/
A blog that considers many of the issues discussed in Moor 1979.
Sparrow, Robert (2007), Killer Robots, Journal of Applied Philosophy 24(1): 62
77.
Should artificial intelligences be allowed to kill? This paper considers the
ethics of the decision to send artificially intelligent robots into war. . . .
Wagner, Alan R.; & Arkin, Ronald C. (2011), Acting Deceptively: Providing
Robots with the Capacity for Deception, International Journal of Social Robotics
3(1): 526; accessed 14 November 2012 from
http://www.cc.gatech.edu/alanwags/pubs/Acting-Deceptively-Final.pdf
Wallach, Wendell; & Allen, Colin (2009), Moral Machines: Teaching Robots Right
from Wrong (New York: Oxford University Press).
Zremski, Jerry (2009), Perspectives Differ on Autopilot, Icing, Buffalo News
(18 February): A1A2,
http://www.buffalonews.com/Perspectives differ on autopilot icing Amid probe of crash no federal consensus for deadly problem.html
Explores the possibility that a decision-making computer might make things
more difficult for a human who is in the decision-making loop.
See also Halpern 2015.
2. Websites:
On computer ethics in general:
Entire courses and books have been devoted to computer ethics. For more
information, do a Google search on: computer ethics
AAAIs AI Topics website on Ethical & Social Issues: Implications of AI
for Society, is an excellent site, with many links:
http://aitopics.net/Ethics (accessed 14 November 2012).
The Research Center on Computing & Society
http://southernct.edu/organizations/rccs/ (accessed 14 November 2012).
Chapter 19
Philosophy of
Artificial Intelligence
Computers. . . are, after all, in the business of making mechanical what

smacks of vitalism to most scientists. [Collins and Quillian, 1972, 313]
The Turing Test(?): A problem is computable if a computer can convince

you it is. (anonymous undergraduate student in the authors course,
CSE 111, Great Ideas in Computer Science, 14 December 2000)1
Figure 19.1: 2005

c by Mediagraphics
1 http://www.cse.buffalo.edu/rapaport/111F04.html
643
644 CHAPTER 19. PHILOSOPHY OF ARTIFICIAL INTELLIGENCE

1. Required:
(a) Turing, Alan M. (1950), Computing Machinery and Intelligence,
Mind 59: 433460, accessed 11 October 2015 from
http://mind.oxfordjournals.org/content/LIX/236/433.full.pdf+html
(b) Searle, John R. (1980), Minds, Brains, and Programs, Behavioral and Brain
Sciences 3: 417457; online at:
http://www.class.uh.edu/philosophy/garson/MindsBrainsandPrograms.pdf
http://l3d.cs.colorado.edu/ ctg/classes/lib/cogsci/searle.pdf
http://cogprints.org/7150/1/10.1.1.83.5248.pdf
2. Recommended:
Rapaport, William J. (2000), How to Pass a Turing Test: Syntactic Semantics,
Natural-Language Understanding, and First-Person Cognition, Journal of Logic,
Language and Information 9(4): 467490; reprinted in James H. Moor (ed.),
The Turing Test: The Elusive Standard of Artificial Intelligence (Dordrecht, The
Netherlands: Kluwer Academic Publishers, 2003): 161184;
http://www.cse.buffalo.edu/rapaport/Papers/TURING.pdf
19.2 Introduction
As with computer ethics, the philosophy of artificial intelligence (AI) is another large
and long-standing discipline in its own right, and so we will only focus on two main
topics: What is AI? And is AI possible? For the second question, we will look at Alan
Turings classic 1950 paper on the Turing test of whether computers can think and at
John Searles 1980 Chinese Room Argument challenging that test.
19.3 What Is AI?

Many definitions of AI have been proposed (for a list of some, see the appendix to this
chapter). In this section, I want to focus on two nicely contrasting definitions. The first
is by Marvin Minsky, one of the pioneers of AI research; the second is by Margaret
Boden, one of the pioneers in cognitive science:
1. . . . artificial intelligence, the science of making machines do things that would

require intelligence if done by men.2 [Minsky, 1968, v]
2. By artificial intelligence I. . . mean the use of computer programs and

programming techniques to cast light on the principles of intelligence in general
and human thought in particular.3 [Boden, 1977, 5]
Minskys definition suggests that the methodology of AI is to study humans in order

to learn how to program computers. Bodens definition suggests that the methodology
goes in the opposite direction: to study computers in order to learn something about
humans.
Both views are consistent with Stuart C. Shapiros three views of AI
[Shapiro, 1992] (cf. [Rapaport, 1998], [Rapaport, 2000a], [Rapaport, 2003]):
AI as advanced computer science, or as engineering:
On this view, the goal of AI is to extend the frontiers of what we know

how to program, in order to reach the ultimate goal of self-programming,
natural-language-understanding computers, and to do this by whatever
means will do the job, not necessarily in a cognitive fashion.
The computer scientist John Case (personal communication) once
described AI understood in this way as being at the cutting edge of
computer science.4
AI as computational cognition: This view has two different versions:

2 I.e.,
by humans.
3 This is just one sentence from a lengthy discussion titled What Is Artificial Intelligence?,
[Boden, 1977, Ch. 1]
4 Another computer scientist, Anthony S. Ralston (personal communication) agreed with the topological
metaphor, except that instead of describing AI as being at the cutting edge, he said that it was at the
periphery of computer science!
1. AI as computational psychology: Here, the goal of AI is to write

programs as theories or models of human cognitive behavior. (Recall our
discussion of computer programs as theories in Ch. 15.)
2. AI as computational philosophy: Here, the goal of AI is to investigate

whether cognition is computable, that is, whether it is (expressible as)
one or more recursive functions.5 By cognition, I mean such mental
states and processes as belief, consciousness, emotion, language, learning,
memory, perception, planning, problem solving, reasoning, representation
(including categories, concepts, and mental imagery), sensation, thought,
etc. If these states and processes are computable, what algorithms are
necessary or sufficient for them? If not all of them are fully computable,
how much of cognition is computable?
Minskys view of AI as moving from humans to computers and Bodens view as

moving from computers to humans are both valid: AI is, in fact, a two-way street. This
is one reason why my preferred name for the field is computational cognition. Other
reasons have to do with the ambiguities of the word intelligence, which is often used
in the sense of IQ (AI is not necessarily concerned only with finding programs with
high IQ),6 and misleading implications of the word artificial (which can suggest that
artificial entities arent the real thing; synthetic might be a better adjective, because
an artificial diamond might not be a diamondit might be a cubic zirconiumwhereas
a synthetic diamond is a real diamond that just happened to be formed in a non-natural
way).
So, computational cognition is the branch of computer science (working with
other disciplines, such as cognitive anthropology, linguistics, cognitive neuroscience,
philosophy, and psychology, among others) that tries to answer the question: How
much of cognition is computable?
The working assumption of computational cognition is that all of cognition is
computable: The study [of AI] is to proceed on the basis of the conjecture that every
aspect of learning or any other feature of intelligence can in principle be so precisely
described that a machine can be made to simulate it [McCarthy et al., 1955].
And its main open research question is: Is what is not yet known to be computable
computable? If so, what are the implications of that fact? (Well investigate this when
we look at the Turing test in 19.4.) If not (which is the conclusion of the Chinese
Room Argument to be discussed in 19.5), why not, and what does that tell us about
cognition?
5 And, if cognition requires more than one recursive function, then, presumably, they will be interacting
functions, as discussed in Ch. 11.

6 But see my Artificial I.Q. Test at http://www.cse.buffalo.edu/rapaport/AIQ/aiq.html (accessed
11 October 2015), discussed in [Rapaport, 1986b], as well as [Ohlsson et al., 2015].

19.4. THE TURING TEST 647
19.4 The Turing Test

19.4.1 How Computers Can Think
We have seen that computational cognition (AI broadly construed, including what
is sometimes called computational cognitive science) holds that cognition is
computable. For our present purposes, it doesnt matter whether the computations are
of the classical, symbolic variety or the connectionist/artificial-neural-network variety.
Nor does it matter whether the neuron firings that produce cognition in the human brain
can be viewed as computations (for further discussion of this, see [Piccinini, 2005],
[Piccinini, 2007a], [Rapaport, 2012b]).
All that matters is this philosophical implication:
If (and to the extent that) cognitive states and processes can be expressed
as algorithms, then they can be implemented in non-human computers.
And this raises the following questions:
Are computers executing such cognitive algorithms merely simulating cognitive

states and processes?
Or are they actually exhibiting them?
In popular parlance, do such computers think?

In this section, we will look at one answer that has been given to this question, an
answer that arises from what is called the Turing test. In the next section, we will look
at an objection to it has been presented in the form of the Chinese Room Argument.
And after that, I will offer my own interpretation of the situation, introducing a theory
I call syntactic semantics.
19.4.2 The Imitation Game

Alan Turings second most important paper [Turing, 1950] never mentions a Turing
test (just as his most important paper [Turing, 1936] never mentions a Turing
machine). Instead, he introduces a parlor game (which you can actually playthis
is not a mere thought experiment) that he calls the Imitation Game.
The Imitation Game consists of three players: A man, a woman, and an interrogator
(who might be either a man or a woman; it might matter whether the interrogator is a
man rather than a woman, or the other way round, but well ignore this for now).7 The
three players are placed in separate rooms, so that they cannot see each other, and they
only communicate by means of something like a chat interface, so that they cannot
hear each other. The reason that they are not allowed to see or hear each other is that
the point of the game is for the interrogator to determine which room has the man,
and which room has the woman. To make things interesting, the woman is supposed
to tell the truth in order to convince the interrogator that she is the woman, but the
7 For that matter, the interrogator could also be a computer! But there are good reasons why this should
be ruled out: The point of the Turing test is for a human to judge whether something is intelligent (or as
intelligent as a human).
man is supposed to convince the interrogator that he (the man) is the woman, so he
will occasionally have to lie. The man wins if he convinces (fools) the interrogator
that he is the woman; the woman wins if she convinces the interrogator that she is the
woman. (Another way of thinking of this is that the interrogator wins if he correctly
figures out who is in which room.) If the man wins, then he is said to have passed
the test. Turing suggested that an average interrogator will not have more than 70
per cent. chance of making the right identification after five minutes of questioning
[Turing, 1950, 442]. But the actual amount of time may be irrelevant. One could
conduct a series of imitation games and calculate appropriate statistics on how likely
an interrogator is to make a correct determination after a given period of time.
What does this have to do with whether computers can think? What has come to be
known as the Turing test makes one small change in the Imitation Game:
We now ask the question, What will happen when a machine takes the
part of [the man] in this game? Will the interrogator decide wrongly
as often when the game is played like this as he [or she] does when the
game is played between a man and a woman? These questions replace our
original, Can machines think? [Turing, 1950, 434]
It turns out that there is some ambiguity in Turings question, and, in any case,
other modifications are possible. First, what is the machine (that is, the computer)
supposed to do? Is it supposed to convince the interrogator that it is the woman? (That
is, is it supposed to imitate a woman?) Or is it supposed to convince the interrogator
that it is a man who is trying to convince the interrogator that he is a woman? (That
is, is it supposed to imitate a man?) Usually, the Turing test is taken, more simply
and less ambiguously, to consist of a set up in which a computer, a human, and a
human interrogator are located in three different rooms, communicating over a chat
interface, and in which both the human and the computer are supposed to convince
the interrogator that each is a human. To the extent that the computer convinces the
interrogator and the human doesnt, under the same criteria for successful convincing
that obtains in the original imitation game, the computer is said to have passed the
Turing test.
And an even simpler version consists merely of two players: a human interrogator
and someone or something (a human or a computer) in two separate, chat-interfaced
rooms. If a computer convinces the interrogator that it is a human, then it passes the
Turing test. Which version of the test is better has been discussed in the literature,8 but
we will now turn to Turings view of the matter.
Here is his answer to the question that has now replaced Can machines think?:
I believe that at the end of the century [that is, by the year 2000] the use
of words and general educated opinion will have altered so much that
one will be able to speak of machines thinking without expecting to be
contradicted [Turing, 1950, 442, my boldface].
Before trying to understand exactly what this might mean, lets consider the Turing test
a bit further.
8 See, especially, [Piccinini, 2000], as well as [French, 2000], [Rapaport, 2006b].
19.4. THE TURING TEST 649
19.4.3 Thinking vs. Thinking

In 1993, The New Yorker magazine published a cartoon by Peter Steiner, showing a
dog sitting in front of a computer talking to another dog, the first one saying, On the
Internet, nobody knows youre a dog. This cartoons humor arises from the fact that
you do not know with whom you are communicating via computer! Yes, its unlikely
that theres a dog typing away at the other end of a chat session or an email, but could
it be a computer pretending to be a human? Or could it be a 30-year-old pedophile
pretending to be a 13-year-old classmate? Normally, we assume that we are talking
to people who really are whom they say they are. In particular, we assume that we
are talking to a human. But really all we know is that we are talking to an entity with
human cognitive capacities.
And that, I think, is precisely Turings point: An entity with human cognitive
capacities is all that we can ever be sure of, whether that entity is really a human or
merely a computer. This is a version of what philosophers have called the argument
from analogy for the existence of other minds. An argument from analogy is an
argument of the form:
Entity A is like (that is, is analogous to) entity B with respect to important
properties P1 , . . . , Pn .
A is (probably) like B with respect to some other property Q.
Another way to formulate it is this:
1. A is like B with respect to P1 , . . . , Pn .
2. In addition, B has property Q.
3. A also has (or probably has, or may have) property Q.
Such an argument is non-deductive: Its quite possible for the premise to be true but
for the conclusion to be false. But it has some inductive strength: The more alike two
objects are in many respects, the more likely it is that they will be alike in many other
respects (and maybe even all respects).
The problem of the existence of other minds is this: I know that I have a mind
(because I know what it means for me to think, to perceive, to solve problems, etc.).
How do I know whether you have a mind? Maybe you dont; maybe youre just some
kind of computer, or android,9 or philosophical zombie.10
The argument from analogy for the existence of other minds is therefore this
argument:
1. You are like me with respect to all physical and behavioral properties.
2. I have a mind. (Or: My behavioral properties can best be explained by the fact
that I have a mind.)
9 like Commander Data in Star Trek.
10 A philosophical zombie is not a horror-movie zombie. Rather, it is an entity who is exactly like us in all
respects but who lacks a mind or consciousness. See [Kirk, 1974], [Chalmers, 1996a], and other references
at http://www.cse.buffalo.edu/rapaport/719/csnessrdgs.html#zombies
3. (probably) you have a mind. (Or: Your behavioral properties can best be
explained if it is assumed that you also have a mind.)
Of course, this argument is deductively invalid. I could be wrong about whether you
are biologically human. In that case, the best explanation of your behavior might be, not
that you have a mind, but that you are a computer who has been cleverly and suitably
programmed. Now, there are two ways to understand this: One way to understand it
is to say that you dont have a mind; youre just a cleverly programmed robot. But
another way to understand it is to say that being cleverly programmed in that way is
exactly what it means to have a mind; perhaps we are both cleverly programmed in that
way, or perhaps (a) you are programmed in that way, whereas (b) I have a brain that
behaves in that way, but (c) these are simply two different implementations of having
a mind.
In either case, am I wrong about your being able to think? That is, am I wrong about
your (human) cognitive abilities? Turings answer is: No! More cautiously, perhaps,
his answer is that whether Im wrong depends on the definition of (human) cognitive
abilities (or thinking):
If human-like cognition requires a (human) brain, and you lack one, then,
technically speaking, you dont have human-like cognition (even if you pass the Turing
test). On this view, I really do think, but you can only think.11
But, if human-like cognition can be implemented in different waysthat is, if it
does not require a (human) brainthen we both have human-like cognition (and thats
why you pass the test). On this view, we both can think.
Heres an analogy: Everyone can agree that birds fly.12 Do people fly? Well, we
certainly speak as if they do; we say things like, I flew from Buffalo to JFK last week.
But we also know that I dont literally mean that I flapped my arms when flying from
Buffalo to JFK; rather, I flew in an airplane: It wasnt I who was flying; it was the
airplane. OK; do planes fly? Well, they dont flap their wings, either. In what sense are
they flying?
There are two ways to understand what it means to say that planes fly: One
way is by what I will call metaphorical extension [Lakoff and Johnson, 1980a],
[Lakoff and Johnson, 1980b]. The reason we say that planes fly is that what they are
doing is very much like what birds do when they fly (they move through the air, even
if their methods of doing so are different). So, instead of using a simile, saying that
planes move through the air like birds fly, we use a metaphor, saying directly that
planes fly. And then that metaphor becomes frozen; it becomes a legitimate part of
our language, so much so that we no longer realize that it is metaphorical. This is just
like what happens when we say that time is money: We say things like, Youre wasting
time, This will save you time, How did you spend your vacation?, and so on. But
were usually not aware that we are speaking metaphorically (until someone points it
11 That is, you are not really thinking, but doing something else that I will call thinking only in a
metaphorical sense.
12 This is true in general, even if most birds actually dont fly: Not only do penguins, ostriches, etc.,
not fly, but baby birds, birds with injured wings, dead birds, etc., also dont fly. Handling the logic of
statements like this is a branch of AI called non-monotonic reasoning; see 2.6.1.4 and [Ginsberg, 1987],
[Strasser and Antonelli, 2015].
19.5. THE CHINESE ROOM ARGUMENT 651
out), and theres often no other (convenient) way to express the same ideas. As Turing
said, our use of words has changed.
The other way to understand what it means to say that planes fly is that we have
realized that flapping wings is not essential to flying. There are deeper similarities
between what birds and planes do when they move through the air that have nothing
to do with wing-flapping but that have everything to do with the shape of wings and,
more importantly, with the physics of flight. We have developed a more abstract, and
therefore more general, theory of flight, one that applies to both birds and planes. And
so we can promote the word fly from its use for birds to a more general use that also
applies to planes. Flying can be implemented by both biological and non-biological
entities. As Turing said, general educated opinion has changed.
In fact, both the use of words and general educated opinion has changed. Perhaps
the change in one facilitated the change in the other; perhaps the abstract, general
theory can account for the metaphorical extension.
What does this have to do with the philosophy of AI? Well, what is a computer?
As we saw in 6.2, originally, a computer was a human who computed. That was
the case till about the 1950s, but, a half-century later, a computer is a machine. (To
emphasize this, before around 1950, what we now call computers were called digital
or electronic computers to distinguish them from the human kind. But, now, it is very
confusing to read pre-1950 papers without thinking of the word computer as meaning,
by default, a non-human machine.) Now, in the beginning of the 21st century, general
educated opinion holds that computers are viewed, not as implementing devices, but
in functional, input-output terms. The same, as we will see, is slowly becoming true
for the word think. But some philosophers argue that what computers do is not really
thinking. To understand why, lets think about one of those philosophers.
19.5 The Chinese Room Argument

Thirty years after Turings publication of the Turing test, John Searle published
a thought experiment called the Chinese Room Argument [Searle, 1980] (cf.
[Searle, 1982]). In this experiment, a human (John Searle himself, as it happens)
who knows no Chinese is placed in a room (the Chinese room) along with paper,
pencils, and a book containing an English-language algorithm for manipulating certain
squiggles (symbols that are meaningless to Searle-in-the-room13 ). Outside the room
is a native speaker of Chinese. There is something like a mail slot in one wall of the
Chinese room. Through that slot, the native speaker passes pieces of paper that contain
a text written in Chinese and reading-comprehension questions about that text. When
Searle-in-the-room gets these pieces of paper, which, from his point of view, contain
nothing but meaningless squiggles, he consults his book and follows its instructions.
Those instructions tell him to manipulate the symbols in certain ways, to write certain
symbols down on a clean piece of paper, and to slip those responses through the
mail slot. The native speaker who reads them sees that whoever (or whatever) is in the
room has answered all the questions correctly, demonstrating a fluent understanding of
13 I will distinguish between the real John Searle, philosopher and author of [Searle, 1980], and the
Chineseless John Searle who occupies the Chinese room by referring to the latter as Searle-in-the-room.
Chinese. This is because the instruction book is a complete, Chinese natural-language-

understanding algorithm. But, by hypothesis, Searle-in-the-room does not understand
Chinese. We seem to have a contradiction.
The Chinese Room Argument (CRA) is offered as a counterexample to the Turing
test, concluding from this thought experiment that it is possible to pass a Turing test,
yet not really think. The setup of the CRA is identical to the simplified version of the
Turing test: The interrogator is the native Chinese speaker, who has to decide whether
the entity in the room understands Chinese. Because the interrogator determines that
the entity in the room does understand Chinese (which is analogous to deciding that the
entity in the simplified Turing test is a human, rather than a computer), but the entity
in fact does not understand Chinese (which is analogous to the entity in the simplified
Turing test being a computer), the test fails.
Or does it?
19.5.1 Two Chinese Room Arguments

Searle actually bases two arguments on the Chinese Room thought experiment:
The Argument from Biology:
B1 Computer programs are non-biological.
B2 Cognition is biological.
B3 No (non-biological) computer program can exhibit (biological) cognition.
The Argument from Semantics:
S1 Computer programs are purely syntactic.
S2 Cognition is semantic.
S3 Syntax alone is not sufficient for semantics.
S4 No (purely syntactic) computer program can exhibit (semantic) cognition.
Objections have been raised against both arguments. We will look at a few of them.
The principal objection to the Argument from Biology is that premise B2 is at
least misleading and probably false: Cognition can be characterized abstractly, and
implemented in different media.
The principal objection to the Argument from Semantics is that premise S3 is false:
Syntaxthat is, symbol manipulationdoes suffice for semantics.
After investigating these objections (and others), we will consider whether there
are other approaches that can be taken to circumvent the CRA. One of them is to try to
build a real analogue of a Chinese room; to do that, we will need to answer the question
of what is needed for natural-language understanding.
19.5.2 The Argument from Biology

19.5.2.1 Causal Powers
Lets begin by considering some of the things that Searle says about the CRA:
I still dont understand a word of Chinese and neither does any other digital
computer because all the computer has is what I have: a formal program
that attaches no meaning, interpretation, or content to any of the symbols.
What this simple argument shows is that no formal program by itself is
sufficient for understanding. . . . [Searle, 1982, 5]
Note that a program that did attach meaning, etc., might understand. But Searle
denies that, too:
I see no reason in principle why we couldnt give a machine the capacity

to understand English or Chinese, since in an important sense our bodies
with our brains are precisely such machines. But. . . we could not give such
a thing to a machine where the operation of the machine is defined solely
in terms of computational processes over formally defined elements. . . .
[Searle, 1980, 422]
Why not? Because only something that has the same causal powers as brains
can have intentionality [Searle, 1980, 423]. By intentionality here, Searle means
cognition more generally. So he is saying that, if something exhibits cognition, then
it must have the same causal powers as brains.
All right; what are these causal powers? After all, if they turn out to be something
that can be computationally implemented, then computers can have them (which Searle
thinks they cannot). So, what does he say they are? He says that these causal powers
are due to the fact that I am a certain sort of organism with a certain biological
(i.e. chemical and physical) structure [Searle, 1980, 422]. Weve narrowed down the
nature of these causal powers a little bit. If we could figure out what this biological
structure is, and if we could figure out how to implement that structure computationally,
then we should be able to get computers to understand. Admittedly, those are big ifs,
but they are worth trying to satisfy.
So, what is this biological structure? Before we see what Searle says about it,
lets think for a moment about what a structure is. What is the structure of the
brain? One plausible answer is that the brain is a network of neurons, and the way
those neurons are organized is its structure. Presumably, if you made a model of the
brain using (lets say) string to model the neurons, then, if the strings were arranged
in the same way that the neurons were, we could say that the model had the same
structure as the brain. Of course, string is static (it doesnt do anything), and neurons
are dynamic, so structure alone wont suffice, but its a start.
But Searle doesnt think so. He says that a simulated human brain made
entirely of. . . millions (or billions) of old beer cans that are rigged up to levers and
powered by windmills would not really exhibit cognition even though it appeared
to [Searle, 1982].14 In other words, structure plus the ability to do something is not
enough: Cognition must (also) be biological, according to Searle; it must be made of
the right stuff.
14 [Weizenbaum, 1976, Ch. 5] considers computers made of bailing wire, chewing gum, and adhesive
tape. [Weizenbaum, 1976, Ch. 2] considers a Turing machine made of toilet paper and pebbles. And recall
our discussion in 8.8 of Hilberts tables, chairs, and beer mugs.
But now consider what Searle is saying: Only biological systems have the requisite
causal properties to produce cognition. So were back at our first question: What
are those causal properties? According to Searle, they are the ones that are causally
capable of producing perception, action, understanding, learning, and other intentional
[i.e., cognitive] phenomena [Searle, 1980, 422]. Again: What are the causal properties
that produce cognition? They are the ones that produce cognition! Thats not a very
helpful answer.
Elsewhere, Searle does say some things that give a possible clue as to what the
causal powers are: mental states are both caused by the operations of the brain and
realized in the structure of the brain [Searle, 1983, 265]. In other words, they are
implemented in the brain. And this suggests a way to avoid Searles argument from
biology.
19.5.2.2 The Implementation Counterargument

Remember, Searle says:
[M]ental states are as real as any other biological phenomena, as real
as lactation, photosynthesis, mitosis, or digestion. Like these other
phenomena, mental states are caused by biological phenomena and in turn
cause other biological phenomena. [Searle, 1983, 264, my emphasis]
Searles mental states are biological implementations. But, if they are
implementations, then they must be implementations of something else, in this case, of
something more abstract: abstract mental states. For every implementation I, there is
an Abstraction A such that I implements A (call this implementation thesis 1).
Lets review what Searle says a bit more closely:
1. . . . intentional states. . . are both caused by and realized in the structure of the
brain. [Searle, 1980, 451, my emphasis]
2. But brains and beer-cans/levers/windmills can share structure. This is a simple
fact about the nature of structure.
3. point 1 must be false: It cant be a single thingintentional (that is, mental)
statesthat is both caused by and realized in the brain. Instead, the brain causes
(a) implemented mental states but (b) realizes abstract mental states, and those
are two distinct things.
Recall our discussion of implementation from Ch. 14. There, we saw that
Abstractions (a generalization of the computer science notion of abstract data type)
can be implemented in more than one waythey can be multiply realized. We
saw that stacks can be implemented as arrays or as lists, that any sequence of items
that satisfy Peanos axioms is an implementation of the natural numbers, that any two
performances of the same play or music are different implementations of the script or
score, and so on. For every Abstraction A, there are more than one implementations
I, I 0 , etc. (call this implementation thesis 2).
So, Searle says that the human brain can understand Chinese because understanding
is biological, whereas a computer executing a Chinese natural-language-understanding
program cannot understand Chinese, because it is not biological. But the

implementation counterargument says that, on an abstract, functional, computational
notion of understanding as an Abstraction, understanding can be implemented in both
human brains and in computers, and, therefore, both can understand.
More generally, if we put implementation theses 1 and 2 together, we can see that if
we begin with an implementation (say, real, biological mental states), we can develop
an abstract theory about them (this is what cognitive science, including computational
cognitive science, tries to do). But once we have an abstract theory, we can re-
implement it in a different medium. If our abstract theory is computable, then we can
re-implement it in a computer. When this happens, our use of words changes, because
general educated opinion changes, as Turing predicted.
19.5.3 The Argument from Semantics

19.5.3.1 Which Premise Is at Fault?
Recall the Argument from Semantics:
S1 Computer programs are purely syntactic.
S2 Cognition is semantic.
S3 Syntax alone is not sufficient for semantics.
S4 No purely syntactic computer program can exhibit semantic cognition.
In this section, we will look at reasons for thinking that (S3) is false, that syntax does
suffice for semantics.
First, lets consider this argument in a little bit more detail. Premise (S1) says
that computer programs merely tell a computer to (or describe how a computer can)
manipulate symbols on the basis of their properties as marks and their relations
among themselves, completely independently of their semantic relations, that is, of the
relations that the symbols have to other items, items that are external to the computer
and that are the meanings or interpretations of the symbols, the things in the real world
that the symbols represent.
Premise (S2) says that cognition is centrally concerned with such external
relations. Cognition, roughly speaking, is whatever the brain does with the sensory
inputs from the external world. To fully understand cognition, according to this
premise, it is necessary to understand the internal workings of the brain, the external
world, and the relations between them. That is a semantic enterprise.
It seems clear that the study of relations among the symbols alone could not
possibly suffice to account for the relations between those symbols and anything else.
Hence premise (S3): Syntax and semantics are two different, though related, subjects.
Conclusion (S4) seems to follow validly. So, any questions about the goodness of
the argument must concern its soundness: Are the premises true? Doubts have been
raised about each of them.
Lets look at (S1) first: Although it is not a computer program, the World Wide Web
is generally considered to be a syntactic object: a collection of nodes (for example,
websites) related by links, that is, a mathematical graph. Some researchers have felt
that there are limitations to this syntactic web and have proposed the Semantic Web.
By attaching meanings to websites (as Searle might say), they hope to make the Web
more. . . well. . . meaningful, more useful. In fact, however, the way they do this is by
adding more syntax!15 So, for now, well accept premise (S1).
Next, lets look at (S2): At least one major philosopher, Jerry Fodor, has argued
that the study of cognition need not involve the study of the external world that is being
cognized [Fodor, 1980]. For one thing, we would need to solve all of the problems of
physics, cosmology, chemistry, biology, etc.in other words, all of the problems of
understanding the external worldbefore we could fully understand cognition. That
by itself is not conclusive; after all, it has been argued that any single branch of AI
is AI complete, for instance, that the problem of natural-language understanding
cannot be fully solved before fully solving all of the other problems in in AI. More
importantly, Fodor argues that cognition is what takes place internally to the brain.
Whether the brain correctly or incorrectly interprets its input from the external world,
its the interpretation that matters, not the actual state of the external world. This view
is called methodological solipsism. Solipsism is, roughly, the view that I am the only
thing that exists; you are all figments of my imagination. Note that you cannot make
the same claim, because, after all, if solipsism is true, then you dont exist!16 (Theres
a story that, at a lecture that Bertrand Russell once gave on solipsism, someone in the
audience commented that it was such a good theory, why didnt more people believe it?)
Methodological solipsism is the view that, as a methodology for studying cognition, we
can pretend that the external world doesnt exist; we only have to investigate what the
brain does with the inputs that it gets, not where those inputs come from or what they
are really like.
Of course, if understanding cognition only depends on the workings of the brain
and not on its relations with the external world, then the study of cognition might be
purely syntactic. And so were ready to consider premise (S3). Can we somehow get
semantics from syntax? There are three, interrelated reasons for thinking that we can.
First, we can try to show that semantics, which is the study of relations between
symbols and meanings, can be turned into a syntactic study, a study of relations among
symbols and symbolized meanings. Second, it can be argued that semantics is
recursive in the sense that we understand a syntactic domain in terms of an antecedently
understood semantic domain, but that there must be a base case, and that this base case
is a case of syntactic understanding. Third, we can take the methodologically solipsistic
approach and argue that an internal, narrow, first-person point of view is (all that is)
needed for understanding or modeling cognition.
Before looking at each of these, remember that Searle claims that syntax cannot
suffice for semantics because the former is missing the links to the external world.
This kind of claim relies on two assumptions, both of which are faulty. First, Searle is
assuming that computers have no links to the external world, that they are really (and
not just methodologically) solipsistic. But this is obviously not true, and is certainly
15 See [Rapaport, 2012b, 3.2 and note 25] for details.
16 Actually, solipsism is not really the claim that only I exist, but that I live in a world of my own completely
cut off from the external world, and so do you. This is reminiscent of the philosopher Gottfried Leibnizs
monads. But thats beyond our present scope.
inconsistent with the arguments about the relation of computing to the world that we
looked at in our discussion of [Smith, 1985] in Chapter 17. Second, Searle assumes
that external links are really needed in order to attach meanings to symbols. But, if so,
then why couldnt computers have them just as well as humans do? Both humans and
computers exist and act in the world. If we humans have the appropriate links, what
reason is there to think that computers could not? (Well, there is the argument from
biology, but this is still an issue that needs to be addressed.)
19.5.3.2 Semiotics
The first reason for thinking that syntax might suffice for semantics comes from
semiotics, the study of signs and symmbols. According to one major semiotician,
Charles Morris [Morris, 1938], semiotics has three branches: syntax, semantics, and
pragmatics.
Given a formal system of marks (symbols without meanings)sometimes called
a (formal) symbol systemsyntax is the study of relations among the marks: how
to recognize, identify, and construct them (in other words, what they look like, for
instance, their grammar), and how to manipulate them (for instance, their proof theory).
Importantly, syntax does not study any relations between marks and non-marks.
Semantics is the study of relations between the marks and their meanings.
Meanings are part of a different domain of semantic interpretations (an ontology).
Therefore, syntax cannot and does not suffice for semantics! (Or so it would seem.)
Pragmatics has been variously characterized as the study of relations among marks,
meanings, and the cognitive agents that interpret them; or as the study of relations
among marks, meanings, interpreters, and contexts. Some philosophers have suggested
that pragmatics is the study of everything that is interesting about symbols systems that
isnt covered under syntax or semantics!
For our purposes, we only need to consider syntax and semantics. Again, lets be
clear. Syntax studies the properties of, and relations among, the elements of a single
set of objects (which we are calling marks); for convenience, call this set SYN.
Semantics studies the relations between the members of two sets, SYN and a set that
well call SEM.
Now, take the set-theoretical union of these two setsthe set of marks and the set
of meanings: M = SYN SEM. Consider M as a new set of marks. In other words, we
have now internalized the previously external meanings into a new symbol system.
And the study of the properties of, and the relations among, the members of M is Ms
syntax!
In other words, what was formerly semantics (that is, relations between the marks
in SYN and their meanings in SEM) is now syntax (that is, relations among the new
marks in M.) This is how syntax can suffice for semantics.17
This can be made clear with the diagram in Figure 19.2.
Consider the top picture: a set, which I will call a SYNtactic DOMain, showing
two of its members and a relation between them. Imagine that there are many members,
17 In fact, any study of semantics in linguisticsin the sense of a study of the meanings of linguistic
expressionsthat focuses only on relations among the expressions and not on their relations to the world is
a syntactic enterprise. This is the nature of, for instance, cognitive semantics.
Figure 19.2:
each with several properties, and many with relations between them. The study of this
set, its members, their properties, and their relations is the syntax of SYN-DOM.
Now consider the middle picture: two sets, SYN-DOM and another, which I will
call a SEMantic DOMain. SYN-DOM, of course, has its syntax. But so does SEM-
DOM. (Often, in AI, the syntax of a semantic domain is called its ontology.) But now
there are additional relations between (some or all of) the members of SYN-DOM and
(some or all of) the members of SEM-DOM. Note that these relations are external to
both domains: You really cant describe these relations using only the language used to
describe SYN-DOM or the language used to describe SEM-DOM; instead, you need a
language that can talk about both domains, and that language cannot be internal to
either domain. The study of these relations is what is called semantics. The usual
idea is that the members of SEM-DOM are the meanings of the members of SYN-
DOM, espcially if SYN-DOM is the language used to describe SEM-DOM. So, for
instance, you might think of SEM-DOM as the actual world and SYN-DOM as either
a language like English that talks about the actual world or perhaps a scientific theory
about the actual world, perhaps expressed in some mathematical (or computational!)
language. Another way to think about this is that SEM-DOM gives us the facilities
needed to understand SYN-DOM: We understand SYN-DOM in terms of SEM-DOM.
In the bottom picture, we have taken the union of these two domains. Now, the
formerly external semantic relations have become internal relations of the new,
unioned domain. But, as such, they are now no different in principle from the
previous internal, syntactic relations of SYN-DOM or the previous internal, syntactic
or ontological relations of SEM-DOM. Thus, these previous semantic relations have
also become syntactic ones. I call this syntactic semantics.18
This semiotic technique for turning semantics into syntax raises a number of
questions: Can the semantic domain be internalized? Yes, under the conditions
obtaining for human language understanding: How do we learn the meaning of a word?
How, for instance, do I learn that the word tree means tree? The common view is
that this relation is learned by associating trees with the word tree. But really what
happens is that my internal representation of a tree is associated with my internal
representation of the word tree. Those internal representations could be activated
neurons. In whatever way that neurons are bound together when, for instance, we
perceive a pink cube (perhaps with shape neurons firing simultaneously with, and
thereby binding with, color neurons that are firing), the neurons that fire when we
see a tree might bind with the neurons that fire when we are thinking of, or hearing, or
reading the word tree.
And the same thing can happen in a computational cognitive agent. Suppose we
have such an agent (a robot, perhaps; call her Cassie) whose computational mind
is implemented as a semantic network whose nodes represent concepts and whose arcs
represent structural relations between concepts, as in figure 19.3: There is a real tree
external to Cassies mind. Light reflecting off the tree enter Cassies eyes; this is the
causal link between the tree and Cassies brain. The end result of the visual process is
an internal representation of the tree in Cassies brain. But she also has an internal
18 For more details on syntactic semantics, see, e.g., [Rapaport, 1988], [Rapaport, 2012b],
[Rapaport, 2015].
Figure 19.3:
representation of the word tree, and those two representations can be associated.
What Cassie now has is an enlarged set of marks, including a mark for a word and
a mark for the words meaning. But they are both marks in her mind.19
This is akin the Robot Reply to the CRA [Searle, 1980, 420], in which sensors and
effectors are added to the Chinese Room so that Searle-in-the-room can both perceive
the external world as well as act in it. Searles response to the Robot Reply is to say
that it is just more symbols. The reply to Searle is to say that that is exactly how human
cognition works! In our brains, all cognition is the result of neuron firings, and the
study of that single set of neuron firings is a syntactic study (because it is the study
of the properties of, and relations among, a single set of marksin this case, the
marks are neuron firings). The same is true for computers: If I say something to
Cassie in English, she builds internal nodes that represent my utterance in her semantic
network. If I show pictures to her, or if she sees something, she builds other internal
nodes representing what she sees. This set of nodes forms a single computational
knowledge base, whose study is syntactic in nature (because it is the study of the
properties of, and relations among, a single set of marksin this case, the marks
are nodes in a semantic network). In the same way, both truth tables and the kind
of formal semantics that logicians study are syntactic ways of doing semantics: The
method of truth tables syntactically manipulates symbols that represent semantic truth
values. And formal semantics syntactically manipulates symbols that represent the
objects in the domain of semantic interpretation.
19.5.3.3 Points of View
The second prong of our reply to the Argument from Semantics concerns the differing
points of view of the native, Chinese-speaking interrogator and Searle-in-the-room.
To understand how a cognitive agent understands, and to construct a computational
cognitive agent, we must take the first-person point of view. We must construct
a cognitive agent (a robot, if you will) from the agents point of view, from the
perspective of whats going on inside the agents head. In other words, we must
be methodologically solipsistic and develop or implement a narrow or internal
model of cognition. Such a model is called narrow, as opposed to wide, because
it ignores the wider outside world and focuses only on the narrow inner world of the
agents point of view. We dont need to understand the causal or historical orgins of
the agents internal symbols; we only need to understand the symbols.
But there are two different points of view: There is Searle-in-the-rooms point of
view and there is the interrogators point of view. In Searles CRA, Searle-in-the-
rooms point of view trumps the interrogators; in the Turing test (and in the kind
of syntactic semantics that I am advocating), the interrogators trumps Searle-in-the-
rooms. How should we resolve this?
Here is an analogy that I think helps clarify the situation. Consider the following
passage from The Wizard of Oz (the novel, not the movie):
19 For more about Cassie, see A Conversation with Cassie at
http://www.cse.buffalo.edu/rapaport/675w/cassie.conversation.new.html (accessed 12 October 2015) and

[Rapaport, 1988], [Shapiro, 1989], [Shapiro and Rapaport, 1995], [Shapiro, 1998], [Rapaport, 2006a].
When Boq [a Munchkin] saw her silver shoes, he said,

You must be a great sorceress.
Why? asked [Dorothy].
Because you wear silver shoes and have killed the wicked witch. Besides,
you have white in your frock, and only witches and sorceresses wear
white.
My dress is blue and white checked, said Dorothy. . . .
It is kind of you to wear that, said Boq. Blue is the color of the
Munchkins, and white is the witch color; so we know you are a friendly
witch.
Dorothy did not know what to say to this, for all the people seemed to
think her a witch, and she knew very well she was only an ordinary little
girl who had come by the chance of a cyclone into a strange land.
[Baum, 1900, 3435]
Is Dorothy a witch? From her point of view, the answer is no; from Boqs point
of view, the answer is yes. Whose point of view should trump the others? Dorothy
certainly believes that shes not a witch, at least as she understands the word witch
(you knowblack hat, broomstick, Halloween, and all that). Now, it is certainly
possible that Dorothy is (such a) witch while believing (mistakenly in that case) that she
is not (such a) witch. So, what counts as being a witch (in these circumstances)? Note
that the dispute between Dorothy and Boq is not about whether Dorothy is really a
witch in some context-independent sense. The dispute is about whether Dorothy is a
witch in Boqs sense, from Boqs point of view. And, because Dorothy is in Oz, Boqs
point of view trumps hers!
Now compare this to the Chinese room situation: Here, instead of asking whether
Dorothy is a witch, we ask: Does Searle-in-the-room understand Chinese? From his
point of view, the answer is no; from the native Chinese speakers point of view,
the answer is yes. Whose point of view should trump the others? Searle-in-the-
room certainly believes that he does not understand Chinese, at least as he understands
understanding Chinese (you knowthe way you understand your native language as
opposed to the way you understand the foreign language that you may have (poorly)
learned in high school or college). Now, it is certainly possible that Searle-in-the-room
does understand Chinese while believing (mistakenly, in that case) that he does not
understand it. So, what counts as understanding Chinese (in these circumstances)? For
the same reason as in the witch case, it must be the native Chinese speakers point of
view that trumps Searle-in-the-rooms!
Of course, it would be perfectly reasonable for Searle-in-the-room to insist that he
doesnt understand Chinese. Compare Searle-in-the-rooms situation to mine: I studied
French in high school; spent a summer living with a French family in Vichy, France;
spent a summer studying French (but mostly speaking English!) at the University of
Aix-en-Provence; and have visited French friends in France many times. I believe
that I understand about 80% of the French that I hear in a one-on-one conversation
(considerably less if Im hearing it on TV or radio) and can express myself the way
that I want about 75% of the time (I have, however, been known to give directions to
Parisian taxi drivers), but I always feel that Im missing something. Should I believe
my native French-speaking friends when they tell me that I am fluent in French? Searle
would say no.
But Searle-in-the-room isnt me. Searle-in-the-room cant insist that he alone
doesnt understand Chinese and that, therefore, his point of view should trump the
native, Chinese-speaking interrogators. And this is because Searle-in-the-room isnt
alone: Searle-in-the-room has the Chinese natural-language-processing instruction
book (even if he doesnt know that thats what it is). This is the core of what is known
as the Systems Reply [Searle, 1980, 419420]that it is the system consisting of
Searle-in-the-room together with the rule book that understands Chinese. After all, it
is not a computers CPU that would understand Chinese (or do arithmetic, or do word-
processing), but it is the system, or combination, consisting of the CPU executing a
computer program that would understand Chinese (or do arithmetic, or process words).
And Searle-in-the-room together with the rulebook, stranded on a desert island, could
communicate (fluently) with a native, Chinese-speaking Friday.20
Does it make sense for a system like this to exhibit cognition? Doesnt cognition
have to be something exhibited by a single entity, like a person, an animal, or a robot?
The cognitive anthropologist Edwin Hutchins [Hutchins, 1995a], [Hutchins, 1995b],
[Hollan et al., 2000] has argued that it is an extended or distributed cognitive
system (such as a ships crew together with their navigation instruments) that navigates
a ship [Hutchins, 1995a], [Hutchins, 1995b], [Hollan et al., 2000]. This is a real-
life counterpart of Searle-in-the-room together with his rulebook: Cognitive science
normally takes the individual agent as its unit of analysis.. . . [But] systems that are
larger than an individual may have cognitive properties in their own right that cannot
be reduced to the cognitive properties of individual persons [Hutchins, 1995b, 265,
266]. So, Searle-in-the-room plus his external rulebook can have the cognitive property
of understanding Chinese, even though Searle-in-the-room all by himself lacks that
property.
In fact, if the property of understanding Chinese (that is, the knowledge of Chinese)
has to be located in some smaller unit than the entire system, it would probably have
to be in the rulebook, not Searle-in-the-room! In an episode of the 1950s TV comedy
series I Love Lucy, Lucy tries to convince her Cuban in-laws that she speaks fluent
Spanish, even though she doesnt. To accomplish this, she hires a native Spanish
speaker to hide in her kitchen and to communicate with her via a hidden, two-way radio,
while she is in the living room conversing with her in-law interrogators. Here, it is
quite clear that the knowledge of Spanish resides in the man in the kitchen. Similarly,
the knowledge of Chinese resides in the rulebook. It is the ability to execute or process
that knowledge that resides in Searle-in-the-room. Together, the system understands
Chinese.
It can be argued that cognitive agents have no direct access to external entities.
When I point to a tree, what I am aware of is, not my actual hand pointing to the
actual tree, but an internal visual image of: my hand pointing to a tree. This is
consistent with the 18th-century philosopher Immanuel Kants theory of phenomena
and noumena. We are not directly aware of (for Kant, we have no knowledge
of) the real world as it is in itself; he called this the world of noumena (singular:
20 Friday was the name of the native resident of the island t at Robinson Crusoe was stranded on.
noumenon). All that we are aware of is the world filtered through our senses and our
mental concepts; he called this the world of phenomena (singular: phenomenon). My
access to the external world of noumena is mediated by internal representatives. There
are several reasons for thinking that this is really the case (no matter how Matrix-like
it may sound!): There is an argument from illusion that says that, because we see
different things with each eye, what we see is, not whats out there, but the outputs of
what our eyes have conveyed to our brains and that our brains have processed.21 There
is an argument from time delay: Because it takes time (no matter how short) for light
reflected off an object to reach our eyes, we see events after they happen; so, what we
are seeing is in our heads, not out there.22
Now, someone who takes a third-person point of view would say that you can have
access to the external world. For instance, as a computer scientist programming a
robot, it seems that I can have access to the world external to the robot as well as to
the robots internal mind (and I can compare the two, to determine if the robot has any
misperceptions). If the robot (or you) and I are both looking at a tree, we see the same
tree, dont we? From the first-person point of view, the answer is no: As the robots
programmer, I have access only to my internal representation of the external world
and to my internal representation of the robots internal world. And the same goes for
you with respect to me, and for me with respect to you. If you and I are looking at a
tree, we are each aware only of our two, separate internal representatives of that tree:
one in your mind, one in mine; one produced by your neuron firings, one produced by
mine. We cannot get outside of our heads to see whats really going on: Kant was
rightly impressed by the thought that if we ask whether we have a correct conception
of the world, we cannot step entirely outside our actual conceptions and theories so as
to compare them with a world that is not conceptualized at all, a bare whatever there
is. [Williams, 1998, 40]
So, by merging internalized semantic marks with internal syntactic marks, the
semantic project of mapping meanings to symbols can by handled by syntax, that is,
by symbol manipulation. That is why syntax suffices for the first-person, semantic
enterprise, and why Searles Argument from Semantics is unsound.
But there is another reason, too.
19.5.3.4 A Recursive Theory of Understanding

Semantics, as we have seen, requires there to be two domains and one binary relation:
There is a syntactic domain of marks (which we have called SYN), characterized by
syntactic formation and inference rules. There is a semantic domain of meanings or
interpretation, (which we have called SEM), also characterized by syntactic formation
and inference rules (called its ontology by AI researchers). And there is a binary,
semantic interpretation function, I : SY N SEM, that assigns meanings from SEM
to marks in SYN.
We use SEM to understand SYN. Therefore, we must antecedently understand
SEM; otherwise, we would be understanding one thing in terms of something else
that we do not understand, and that should hardly count as understanding.
21 See [Ayer, 1956, Ch. 3, (ii), pp. 8795].
22 See [Russell, 1912, Ch. 3, p. 33], [Changizi et al., 2008].
So, how do we understand SEM? In the same way that we understand SYN: by
treating SEM as a new syntactic domain, and then finding a new semantic domain,
SEM0 , in terms of which to understand it. [Smith, 1987] called this a correspondence
continuum, because it can be continued indefinitely, understanding the SEM0 in terms
of yet another SEM00 , and so on. To stop an infinite regress, there must be a base
case, a last semantic domain that we understand directly, in terms of itself rather
than in terms of something else. But to understand a domain in terms of itself is to
understand its members in terms of their properties and relations to each other. And
that is syntax. It is a kind of understanding that I call syntactic understanding. We
understand a domain syntactically by being conversant with manipulating its marks (or
by knowing which wffs are theorems). On this view, the meaning of a mark is its
location in a network of other marks, with the connections between the marks being
their properties and relations to the other marks in the network. (This is a kind of
structural understanding; cf. [Saussure, 1959].)
Here is another way to think about it: When I understand what you say, I do this
by interpreting what you say, that is, by mapping what you say into my concepts.
Similarly, I (semantically) understand a purely syntactic formal system by interpreting
it, that is, by providing a (model-theoretic) semantics for it. Now, lets turn the tables:
What would it be for a formal system to understand me? (Does that even make sense?
Sure: Robots that could understand natural language, or even simple commands, are
merely programsformal systemsbeing executed.) The answer is this: A formal
system could understand me in the same way that I could understand itby treating
what I say as a formal system and interpreting it. Note that links to the external world
are irrelevant; the semantic interpretation of a formal system is a purely syntactic
enterprise.23
19.5.4 A Better Way

So, the really interesting question is: Whats in the rulebook? What is needed for
(computational) natural-language understanding? To understand language, a cognitive
agent must (at least):
take discourse as input; it does not suffice for it to be able to understand isolated
sentences
understand ungrammatical input; we do this all the time, often without realizing
it, and, even when we realize it, we have to be able to recover from any
misinterpretations
make inferences and revise our beliefs; after all, what you say will often cause
me to think about other things (a kind of inferencing) or to change my mind
about things (belief revision)
make plans: We make plans for speech acts (how should I ask you to pass the
salt? Should I demand Gimme the salt!, or should I politely ask May I please
have the salt??, or should I merely make the observation Gee; this food needs
23 For more details, see [Rapaport, 1988], [Rapaport, 2012b].
some salt.?), we make plans to ask and to answer questions, and we make plans
about how to initiate or end conversations.
understand plans, especially the speech-act plans of our interlocutors (when you
said, Its chilly in here, did you really mean that you wanted me to close the
window?)
construct a user model, that is, a model of our interlocutors beliefs
learn about the world and about language
have background knowledge (sometimes also called world knowledge or
commonsense knowledge)
remember what it heard, what it learned, what it inferred, and what beliefs it has
revised.
When you consider all of these things, you realize that, to understand natural language,
you need to have a mind! And this mind can be constructed as a syntactic system.
In other words, the rulebook must be a computer program for complete AI. Natural-
language understanding is an AI-complete problem [Shapiro, 1992].
A robot with such a syntactic (or computational) mind would be like Searle-in-the-
room, manipulating symbols that are highly interconnected and that include internal
representatives of external objects. It would be causally linked to the external world
(for this is where it gets its input), which provides grounding and a kind of external,
third-person, semantic understanding. Such a robot could (or, more optimistically,
will be able to) pass a Turing test and escape from the Chinese room.
19.6 Appendix: Some Definitions of AI

1. We propose that a 2 month, 10 man study of artificial intelligence be carried out
during the summer of 1956 at Dartmouth College in Hanover, New Hampshire.
The study is to proceed on the basis of the conjecture that every aspect of
learning or any other feature of intelligence can in principle be so precisely
described that a machine can be made to simulate it. An attempt will be made
to find how to make machines use language, form abstractions and concepts,
solve kinds of problems now reserved for humans, and improve themselves. We
think that a significant advance can be made in one or more of these problems
if a carefully selected group of scientists work on it together for a summer.
[McCarthy et al., 1955, my emphasis]
2. The goal of work in artificial intelligence is to build machines that perform
tasks normally requiring human intelligence. [Nilsson, 1971, vii] (See also
[Nilsson, 1983].)
3. Research scientists in Artificial Intelligence try to get machines to exhibit
behavior that we call intelligent behavior when we observe it in human beings.
[Slagle, 1971, 1]
19.6. APPENDIX: SOME DEFINITIONS OF AI 667
4. B[ertram] Raphael. . . has suggested that AI is a collective name for problems

which we do not yet know how to solve properly by computer.24 [Michie, 1971,
101]
5. What is or should be [AI researchers] main scientific activitystudying
the structure of information and the structure of problem solving processes
independently of applications and independently of its realization in animals or
humans. [McCarthy, 1974, 317]
6. By artificial intelligence I therefore mean the use of computer programs and
programming techniques to cast light on the principles of intelligence in general
and human thought in particular. [Boden, 1977, 5]
7. Artificial Intelligence is the branch of Computer Science that attempts to solve
problems for which there is no known efficient solution, but which we know
are efficiently solvable, (typically) because some intelligence can solve the
problem (often in real time). A side benefit of AI is that it helps us learn
how intelligences solve these problems, and thus how natural intelligence works.
[Leler, 1985]
8. Artificial intelligence is concerned with the attempt to develop complex
computer programs that will be capable of performing difficult cognitive tasks.
[Eysenck, 1990, 22]
9. AI is making computers act like those in movies. [Brown, 1992]
10. We define AI as the study of agents that receive percepts from the environment
and perform actions [Russell and Norvig, 2003, vii]. . . . the study of rational-
agent design [Russell and Norvig, 2003, 5].
Some discussions of the nature of AI cannot be summarized neatly in a one-

sentence definition:
A valuable discussion of the nature of AI may be found in a debate between
two AI researchers, Roger C. Schank and Alan Bundy; see [Schank, 1983],
[Bundy, 1983].
A discussion by a software engineer comparing two different definitions, in the
context of the US governments Strategic Defense Initiative, is in [Parnas, 1985].
John McCarthy has written several essays on the nature of AI; see
[McCarthy, 1988], [McCarthy, 2007].
A philosopher-turned-AI researcher, Aaron Sloman, discussed the nature of AI
in a newsgroup, [Sloman, 1989].
A discussion among three computer scientists on what AI is
[Shapiro et al., 1992]
24 Note that it follows that, once we do know how to solve them, they are no longer AI!

1. [Quillian, 1994, 440442] has some interesting arguments based on the history
of the physics of flight to the effect that studying what might be called artificial
flight was crucial. [S]tudying the animals that fly, no matter in how great detail
and for how many years, would not have yielded any useful information on how
humans might be able to fly. Rather, attempt[ing] to construct devices that
fly. . . attempts to build flying machines resulted in our entire understanding of
flight today. Even if ones aim is to understand how birds or insects fly, one will
look to aeronautics for the key principles. . . . This supports my claim (in this
book? Or merely in some of my syntactic-semantics papers?) that the study of
computational theories of cognition helps us generalize the more parochial study
of human (or even, more generally, animal) cognition. (Quilliana pioneer in
AI researchuses this argument to support his explanation of why the natural
sciences are more effective than the social sciences.)
2. Is the full power of a TM needed for AI? [Sloman, 2002, 3.3] says no. This
seems correct; after all, even NLP might not need the full power of a TM (a
context-free grammar might suffice). On the other hand, cant a human do
anything that a TM can do? TMs, after all, are computational models of human
computing ability.
3. The analytical engine has no pretentions whatever to originate

anything. It can do whatever we know how to order it to perform.
(Ada Lovelace, cited in [Menabrea, 1843, 722].
That it has no pretentions means that it wasnt designed that way; nevertheless,
it might still be able to originate things. Also, if we can find out how to order
it to perform cognitive activities, then it can do them!
Finding out how requires us to be conscious of something that we ordinarily do
unconsciously:
Regardless of what one calls the work of a digital computer

[specifically, regardless of whether one says that it can think], the
unfortunate fact remains that more rather than less human thinking
is required to solve a problem using a present day machine since
every possible contingency which might arise during the course of the
computation must be thought through in advance. The jocular advice
recently published to the effect, Dont Think! Let UNIVAC do it for
you, cannot be taken seriously. Perhaps, if IBMs familiar motto
[namely, Think!] needs amending, it should be Think: Think
harder when you use the ULTIMAC. [Samuel, 1953, 1225].
Surely this is related to the theme of [Dennett, 2009].
4. There is a nice discussion of marks and mark manipulation systems in

[Kearns, 1997, 2, pp. 273274]
5. Recently, Michael Scherer, a Time magazine bureau chief, received

a phone call from a young lady, Samantha West, asking him if he
wanted a deal on health insurance. After she responded to a number
of his queries in what sounded like prerecorded fashion, he asked her
point-blank whether she was a robot, to which he got the reply I am
human. When he repeated the question, the connection was cut off.
Samantha West turned out to be a system of recorded messages that
were part of a computer program created by the brokers for health
insurance. [Skidelsky, 2014, 36]
6. On Turings strange inversion:
[T]he thought experiment [most recently described by Ned Block,

in which a computer programmed with a large table-lookup passes
the Turing test] is meant to show that seemingly intelligent behavior
can be produced by unintelligent means, whereas real intelligence
cannot. . . . [McDermott, 2014, 144]
This is also the underlying motivation of Turings strange inversion. Actually,

the latter is an even stronger version than the table-lookup version, because the
latter allows for the underlying computer program to be more than merely table-
lookup. McDermotts point is that even table-lookup is more than mere.
7. On the CRA:
Rescorlas (and Clelands) emphasis on the interpretation of the symbols used by
a computer is relevant to the CRA. Here is a nice description of computation that
matches the CRA:
Consider again the scenario described by Turing: an idealized
human computor manipulates symbols inscribed on paper. The
computor manipulates these symbols because he [sic] wants to
calculate the value some number-theoretic function assumes on
some input. The computor starts with a symbolic representation for
the input, performs a series of syntactic operations, and arrives at a
symbolic representation for the output. This procedure succeeds only
when the computor can understand the symbolic representations he
manipulates. The computor need not know in advance which number
a given symbol represents, but he must be capable, in principle, of
determining which number the symbol represents. Only then does his
syntactic activity constitute a computation of the relevant number-
theoretic function. If the computor lacks any potential understanding
of the relevant syntactic items, then his activity counts as mere
manipulation of syntax, rather than calculation of one number from
another. [Rescorla, 2007, 261262, my boldface, Rescorlas italics]
Without the boldfaced sentence, this is a nice description of the CR. The
difference is that, in the CR, Searle-in-the-room does not want to communicate
in Chinese; he doesnt know what hes doing, in that sense of the phrase. Still,
hes doing it, according to the intepretation of the native speaker outside the
room.
8. The basic strategy of AI has always been to seek out progressively

more complex human tasks and show how computers can do them,
in humanoid ways or by brute force. With a half-century of steady
progress, we have assembled a solid body of tested theory on the
processes of human thinking and the ways to simulate and supplement
them. (Herbert Simon, quoted in [Hearst and Hirsh, 2000, 8].)
The phrase human tasks nicely avoids any issues involved with the notion of
intelligence.
9. On the idea that neurons have a syntax (19.5.3.2): [Rescorla, 2012a, 19]
says, neural properties are not multiply realizable. So neural properties are
not syntactic properties. But he gives no argument for why they are not
multiply realizable, nor is it clear why he thinks that they are not syntactic as a
consequence of their non-realizability. In fact, while I would agree that neurons
are not multiply realizable, I fail to see why their properties are not. Properties,
after all, are universals, and, if anything is multiply realizable, surely universals
are.
10. On strong-vs.-weak AI:

Here is Simons version of the distinction:
Those of us who regard computer programs as theories rather

than metaphors are probably still in the minority. Weak AI, as
the metaphoric view is sometimes called, still probably has more
advocates than strong AI. [Simon, 1996a, 161]
So, strong AI is the view that a computational theory of cognition expressed as

a computer program is a legitimate theory, whereas weak AI is the view that
such a program is merely a metaphor. Presumably, if a computational theory
of cognition is correct, then, when it is being executed, it is really exhibiting
cognitive behavior (and not merely simulating it).
11. The subject of computation is essentially that of artificial intelligence

since the development of computation is in the direction of making
machines carry out ever more complex and sophisticated processes,
i.e. to behave as intelligently as possible. [McCarthy, 1963, 4.2,
p. 38].
This follows from his earlier definition (see Ch. 3) of computation as the science
of how to get machines to carry out intellectual processes. The latter is surely
equivalent to AI.
12. On the idea of abstracting from a single example and re-implementing

computationally:
Quite typically, an abstract structure underlies some human cognitive

activity that is not at all apparent in superficial phenomenology
or practice. Often, that structure is related in interesting ways to
the structures we would invent if we constructed an ideal machine
to perform that cognitive activity. (We might think of artificial
intelligence as a normative enterprise). But that structure is rarely
identical to the ideal machines structure. [Gopnik, 1996, 489]
The underying abstract structure could be computational in nature. Hence, it

could be (re-)implemented in an ideal machine. The abstract computational
theory might be thought of as having the form: such-and-such a human cognitive
activity can/ought to be performed in this computational way even if the way that
humans in fact do it is not identical to that ideal structure.
Gopnik goes on to say:
This process may seem like analogy or metaphor, but it involves more
serious conceptual changes. It is not simply that the new idea is the
old idea applied to a new domain, but that the earlier idea is itself
modified to fit its role in the new theory. [Gopnik, 1996, 498]
I have said elsewhere that the process is metaphorical. I dont think that that is
inconsistent with it also involving more serious conceptual changes: We come
to see the old idea in a new way. This is all relevant to my interpretation of what
Turing meant by the Turing test.
13. Comments on [Turing, 1951]:
(a) In this essay, Turing seems to come out a bit more strongly than he does in
[Turing, 1950] about the possibility of computers thinking:
You cannot make a machine to think for you. This is. . . usually
accepted without question. It will be the purpose of this paper to
question it. (p. 256)
Although it is possible to read that last sentence neutrally, to my ears it
sounds like a challenge strongly suggesting that Turing thinks that you
can make a machine think. Indeed, later he says that his contention is
that machines can be constructed which will simulate the behaviour of
the human mind very closely (p. 257). This is cautiously wordedis
simulation of thinking (that is, the behavior of the human mind) the same
as real thinking?but his ultimate claim here is that it will come so close
to human thinking as to make no difference: on the whole the output of
them [that is, of such thinking machines] will be worth attention to the
same sort of extent as the output of a human mind (p. 257, my italics).
And how would this be proved? By the Turing test: It would be the actual
reaction of the machine to circumstances that would prove my contention,
if indeed it can be proved at all (p. 257)
(b) He also suggests that the algorithm for such a machine must be based on
what is now called machine learning: If the machine were able in some
way to learn by experience it would be much more impressive (p. 257).
Moreover, he also suggests that the machine should be an oracle machine!
(Recall our discussion of these in Ch. 11.)
There is. . . one feature that I would like to suggest should be
incorporated in the machines, and that is a random element.
Each machine should be supplied with a tape bearing a random
series of figures, e.g., 0 and 1 in equal quantities, and this series
of figures should be used in the choices made by the machine.
(p. 259)
Note, however, that Turing seems to consider them to be a (small) extension
of Turing machines.
(c) Also interesting is his anticipation of what is now called The Singularity
(see Ch. 11), and the question that we will return to in Ch. 20 about whether
we should build artificial intelligences:
Let us now assume, for the sake of argument, that these
machines ae a genuine possibility, and look at the consequences
of constructing them. To do so would of course meet with
great opposition, unless we have advanced greatly in religious
toleration from the days of Galileo. There would be great
opposition from the intellectuals who were afraid of being put
out of a job. . . . it seems probably that once the machine thinking
method had started, it would not take long to outstrip our feeble
powers. There would be no question of the machines dying,
and they would be able to convers with each other to sharpen
their wits. At some stage therefore we should have to expect the
machines to take control. . . . (pp. 259260)
14. On Turings end of the century prediction, also cite Simon & Newells
ten year prediction about chess. Modify the following passage from my
[Rapaport, 2000b]:
I believe that a suitably programmed computer could pass a Turing

test. I dont think that this has happened yet, examples such as Eliza,
Parry, or the Loebner competitions notwithstanding.25 Nor do I think
that it is going to happen in the near future. As I write, 2001 is
close upon us, but HAL is not (cf. Stork 1997), and I wont venture
to make any more precise predictions: Both Turing (who, in 1950,
predicted 2000) and Herbert A. Simon and Allen Newell (who, in
1957, predicted 1967 for the chess version of a Turing test, missing
25 On Eliza, see Weizenbaum 1966, 1967, 1976. On Parry, see Colby, Weber, & Hilf 1971; Colby et al.
1972; Colby 1981. And on the Loebner competitions, see Loebner 1994, 1998; Rees et al. 1994; Shieber
1994ab.
by 30 years; see Simon & Newell 1958) were way off, and I couldnt
hope to compete with the likes of them.26
26 Although Simon says that it had nothing to do with the Turing Test and that (a) I regard the predictions
as a highly successful exercise in futurology, and (b) placed in the equivalent position today, I would make
them again, and for the same reasons. (Some people never seem to learn.) (personal communication, 24
September 1998). At the end of the next millennium, no doubt, historians looking back will find the 40-year
distance between the time of Newell and Simons prediction and the time of Kasparovs defeat to have been
insignificant.

1. ADD MATERIAL FROM MY WEBSITES AT
http://www.cse.buffalo.edu/rapaport/584/philai.html
2. Bernstein, Jeremy (1976), When the Computer Procreates, New York Times Magazine
(15 February),
http://vasulka.org/archive/Writings/WhenComputerProcreate.pdf
Its most profound impact arises not from its usefulness as calculating machine, but
from its potential for self-replicationeven mutation.
3. Cole, David (2014), Alan Turing & the Chinese Room Argument,
http://www.thecritique.com/articles/alan-turing-the-chinese-room-argument/
Part of a series of online articles providing background for the movie The Imitation
Game. This is a critique of the Chinese Room Argument by a philosopher who has
written widely on the topic.
4. Horst, Steven (2014), The Computational Theory of Mind: Alan Turing & the Cartesian
Challenge,
http://www.thecritique.com/articles/the-computational-theory-of-mind-alan-turing-the-
cartesian-challenge/
Game. This is a critique of Turing 1950 and the Chinese Room Argument,
suggesting that Searles two main points. . . are (1) that computation alone is not
sufficient for understanding and (2) that passing the Turing test would not be proof
that understanding (hence thinking) is involved.
5. Husbands, Phil; & Holland, Owen (2012), Warren McCulloch and the British
Cyberneticians, Interdisciplinary Science Reviews 37(3) (September): 237253,
http://www.mathcomp.leeds.ac.uk/turing2012/Images/mcculloch.pdf
6. Kahn, Amina (2014), Rover Disguised as Penguin Chick Does Research better than
Scientists, Los Angeles Times Science Now (3 November), accessed 10 November 2014
from
http://www.latimes.com/science/sciencenow/la-sci-sn-penguin-robot-chick-study-
20141103-story.html
Did this penguin robot pass a kind of Turing test?
7. McDermott, Drew (2014), What Was Alan Turings Imitation Game?,

http://www.thecritique.com/articles/what-was-alan-turings-imitation-game/
Game. This critique of Turing 1950 was written by a well-known AI researcher.
8. McCarthy, John; & Hayes, Patrick J. (1969), Some Philosophical Problems from
the Standpoint of Artificial Intelligence, in B. Meltzer & D. Michie (eds.), Machine
Intelligence 4 (Edinburgh: Edinburgh University Press),
http://www-formal.stanford.edu/jmc/mcchay69.html
For a commentary, see:

Sloman, Aaron (1971), Interactions between Philosophy and Artificial
Intelligence: the Role of Intuition and Non-Logical Reasoning in Intelligence,
Second International Joint Conference on Artificial Intelligence, Imperial College,
London (London: British Computer Society): 270278,
http://ijcai.org/Past%20Proceedings/IJCAI-1971/PDF/022.pdf
(accessed 23 November 2015); updated verion in Artificial Intelligence 2(1971):
209225,
http://www.cs.bham.ac.uk/research/projects/cogaff/sloman-analogical-1971/
9. McCarthy, J.; Minsky, M.L.; Rocheser, N.; & Shannone, C.E. (1955), A Proposal for the
Dartmouth Summer Research Project on Artificial Intelligence,
http://www-formal.stanford.edu/jmc/history/dartmouth.pdf
The founding document of AI. The study is to proceed on the basis of the
conjecture that every aspect of learning or any other feature of intelligence can
in principle be so precisely described that a machine can be made to simulate
it. An attempt will be made to find how to make machines use language, form
abstractions and concepts, solve kinds of problems now reserved for humans, and
improve themselves. We think that a significant advance can be made in one
or more of these problems if a carefully selected group of scientists work on it
together for a summer. An overly optimistic goal, but Newell, Simon, & Shaw
presented a working AI program, the Logic Theorist (see [Newell et al., 1958] and
https://en.wikipedia.org/wiki/Logic Theorist).
10. Pylyshyn, Zenon W. (1992), Computers and the Symbolization of Knowledge, in

Ralph Morelli, W. Miller Brown, Dina Anselmi, Karl Haberlandt, & Dan Lloyd (eds.),
Minds, Brains & Computers: Perspectives in Cognitive Science and Artificial Intelligence
(Norwood, NJ: Ablex): 8294.
Preprint (1987) accessed 9 April 2014 from:

(see esp. p. 7)
11. Rey, Georges (2012), The Turing Thesis vs. the Turing Test, The Philosophers
12. Shannon, Claude E. (1953), Computers and Automata, Proceedings of the Institute of
Radio Engineers 41(10) (October): 12341241.
A review of game-playing and learning machines by one of the pioneers of

information theory.
13. Simon, Herbert A. (1966), Thinking by Computers, in Robert G. Colodny (ed.), Mind
and Cosmos: Essays in Contemporary Science and Philosophy, University of Pittsburgh
Series in the Philosophy of Science, Vol. 3 (Pittsburgh: University of Pittsburgh Press),
321, accessed 30 September 2015 from:
http://digitalcollections.library.cmu.edu/awweb/awarchive?type=file&item=39897
Among other things, contains a counterargument to the claim that computers cannot
think because they do only what they are programmed to do (p. 18).
A sequel to this essay is Simons Scientific Discovery and the Psychology of
Problem Solving, in the same volume, pp. 2240, in which he argues that
scientific discovery is a form of problem solving, and hence computational in
nature.
14. Wilkes, M.V. (1953), Can Machines Think?, Discovery 14 (May): 151; reprinted in
Proceedings of the Institute of Radio Engineers 41(10) (October): 12301234.
Speculations by one of the pioneers of computers. Discusses the Turing test,
learning machines, and the role of external input.
Chapter 20
Computer Ethics II:

Should We Build
Artificial Intelligences?
Douglas Engelbart, who more than anyone else invented the modern user
interface, modern networking and modern information management. . .
met Marvin Minskyone of the founders of the field of AIand Minsky
told him how the AI lab would create intelligent machines. Engelbart
replied, Youre going to do all that for the machines? What are you going
to do for the people?
[Lanier, 2005, 365]
677
678 CHAPTER 20. COMPUTER ETHICS II: AI

1. Required:
LaChat, Michael R. (1986), Artificial Intelligence and Ethics: An Exercise in the
Moral Imagination, AI Magazine 7(2): 7079.
Lem, Stanisaw (1971), Non Serviam, in S. Lem, A Perfect Vacuum, trans. by
Michael Kandel (New York: Harcourt Brace Jovanovich, 1979).
Reprinted as:
The Experiment (A Review of Non Serviam, by James Dobb), The New
Yorker (24 July 1978): 26ff.
Reprinted in:
Hofstadter, Douglas R.; & Dennett, Daniel C. (eds.) (1981), The Minds I:
Fantasies and Reflections on Self and Soul (New York: Basic Books): 296
317.
Moor, James H. (1985), What Is Computer Ethics?, Metaphilosophy 16(4)
(October): 266275.
4. Highly Recommended:
Dietrich, Eric (2001), Homo sapiens 2.0: Why We Should Build the Better Robots
of our Nature, Journal of Experimental and Theoretical Artificial Intelligence
13(4) (October): 323328.
Turkle, Sherry (2004), How Computers Change the Way We Think, Chronicle of
Higher Education (January 30): B26B28,
http://web.mit.edu/sturkle/www/pdfsforstwebpage/Turkle how computers change way we think.pdf
Petersen, Stephen (2007), The Ethics of Robot Servitude, Journal of
Experimental and Theoretical Artificial Intelligence 19(1) (March): 4354; online
at:
http://stevepetersen.net/professional/petersen-robot-servitude.pdf
5. Recommended:
Powers, Richard (1995), Galatea 2.2 (New York: Farrar, Straus, Giroux).
20.2 Introduction
Stanisaw Lems short story, Non Serviam [Lem, 1971], concerns what is now
called artificial life (or A-life). In the story, an A-life researcher constructs
a computational world of intelligent entities, and follows their evolution and
development of language and philosophy. These artificial intelligences discuss the
existence of God in much the same way that human philosophers have. The difference
(if it is a difference) is that both the researcher and reader realize that he, the researcher,
is their God, and, although he created them, he is neither omniscient nor omnipotent,
and, worse, when his funding runs out, he will have to literally pull the plug on the
computer and thereby destroy them.
This raises the question of whether such an experiment should even be begun,
which is the second ethical issue and the final philosophical topic that we will look
at.
20.3 Artificial Persons

One of the earliest philosophical investigations of this issue is an essay by Michael R.
LaChat that appeared in AI Magazine, [LaChat, 1986]. LaChat argued that it is
worthwhile to consider the moral implications of creating an artificial intelligence, an
artificial person. One reason is that it might happen, so we should be prepared for
it. Another reason is that, even if it turns out to be improbable, such a discussion
illuminates what it means to be a person, which is an important goal in any case. Lets
consider each of these.
20.4 Is AI Possible in Principle?

An artificial intelligence might be created if and only if the field of artificial intelligence
is capable of succeeding in its goal of developing a computational theory of cognition.
Most objections to AI stem from what is known as mind-brain dualism. This is,
roughly, the view that the mind and the brain are two, distinct kinds of entities that
somehow interact. On this view, the mind is not physical, but the brain is. And nothing
physicalneither the brain nor a computercan give rise to something non-physical;
hence, AI cannot create a mind.
But, LaChat notes, there is no evidence of conscious thought without a physical
brain. So, having a brain seems to be a necessary condition of having a mind, even if
we dont believe that thoughts are identical tothat is, are the very same things as
brain states or processes. Hence, it might be sufficient to simulate a brain (perhaps
computationally) in order to create a mind.
On LaChats view, AI is possible in principle if it is possible that there exists a
functional isomorphism between (1) the neural network that constitutes our brain
(that is, brain states and processes) and (2) any other physical implementation of
the functional (that is, psychological) behavior that that neural network implements
[LaChat, 1986, 72]. In other words, psychology is an Abstraction that can be

implemented in either brains or other physical media.
Functionalism is roughly the view that the mindbetter: cognitionis the
function of the brain, that is, the mind is what the brain does. Or, as the philosopher
Hilary Putnam suggested [Putnam, 1960], a Turing-machine program stands in the
same relation to computer states and processes as mental states and processes stand
to brain states and processes (sometimes summarized as the mind is to the brain as
software is to hardware). (For a good overview of functionalism, see [Fodor, 1981].)
Functionalism, as a way of resolving the mind-brain problem, has the advantage of
allowing all mental states and processes to be implemented in some physical states and
processes; this is the principle of multiple realization. And it has another advantage:
It allows the mind to be an emergent property of a certain level of organization
[LaChat, 1986, 72].
There are, of course, problems, both for functionalism in particular and
for AI in general. One is the problem of personality: Would [a] personal
intelligence. . . have personality? [LaChat, 1986, 73] thinks this is almost
impossible, but there is considerable computational work on emotions, so I
would not rule this out of hand. (See Reid Simmonss Social Robots Project,
[http://www.conscious.cmu.edu/afs/conscious/project/robocomp/social/www/]; the
Cog project at
MIT, [http://www.ai.mit.edu/projects/humanoid-robotics-group/cognitive/]; and work
by Rosalind Picard [Picard, 1997], Aaron Sloman [Sloman and Croucher, 1981], and
Paul Thagard [Thagard, 2006].) Another is the problem of pain and other qualia,
that is, qualitative feelings and experiences such as colors and sounds. (Do red fire
engines look the same to you and to me? Or do fire engines for you seem to have the
color that grass has for me? Why does the sound of a bell give rise to the experience
that it does rather than the experience that the smell of garlic has?) The problem here
is whether computers could experience qualia and, even if they could, how we would
know that. This is a vast topic well beyond our present scope, but for a brief argument
that qualia are not out of the question for artificial intelligences, see Shapiros example
of a computer that can feel pain, discussed in [Rapaport, 2005b, 2.3.1].
20.5 What Is the Measure of Success?

How would we know if we have achieved a personal artificial intelligence? One way,
of course, would be by having it pass a Turing test. LaChat offers a different criterion:
Joseph Fletchers analysis of personhood [Fletcher, 1972].
The question of what kinds of entities count as persons is not limited to AI.
The issue arises most prominently in the abortion debate: If fetuses are persons, and
if killing persons is immoral, then abortion is immoral (to oversimplify matters). It
also arises in animal ethics: Are dolphins intelligent enough to be considered persons?
(And how about extraterrestrials? Or corporations?) The point is that there is a
distinction between the biological category of being human and an ethical or legal
category of being a person. The question is: How can personhood be characterized in
an implementation-independent way?
20.5. WHAT IS THE MEASURE OF SUCCESS? 681
On Fletchers analysis, x is a person1 if and only if x has (1) the following positive
characteristics:
1. minimal intelligence
This might mean, for example, having an IQ greater than about 30 or 40
(if you believe that IQ measures intelligence). That is, to be minimally
intelligent is not to be mere biological life (presumably, a bacterium
would not be minimally intelligent; according to Douglas Hofstadter, what
Fletcher is calling minimal intelligence would only apply to lifeforms
evolutionarily higher than a mosquito [Hofstadter, 2007]). For example
minimal intelligence might include some level of rationality, or perhaps
even language use.
2. a sense of self
That is, persons must be self-aware and exhibit self-control.
3. a sense of time
Persons must have a sense of the past, hence some kind of culture; a sense
of the future (so that they have the ability to make plans); and a sense of
the passage of time.
4. a social role
Persons must have an ability to relate to others, to have concern for others,
and to communicate with others (hence the need for language as part of
minimal rationality).
5. curiosity
That is, persons must not be indifferent.
6. changeability
Persons must be creative and be able to change their minds.
7. idiosyncrasy, or uniqueness
8. neo-cortical function
The cerebral cortex is where all the cognitive action occurs in the
brain, so, for Fletcher, a person must have (something whose function
is equivalent to) a cortex. (For more on neo-cortical function, see
[Cardoso, 1997].)
and (2) the following negative characteristics:
1. neither essentially non-artificial nor essentially anti-artificial
1 Fletcher actually uses the term human, not person, but I dont think that this is terminologically
important.
This clause allows for multiple realization and does not restrict personhood
to biological entities. It also allows for test-tube babies to count as
persons.
2. not essentially sexual
So, a cloned entity could be a person.
3. not essentially a bundle of rights
Fletcher argues that there are no essential rights; hence, the notion of
rights cannot be used to characterize persons.
4. not essentially a worshipper
You dont have to be religious to be a person.
20.6 Rights
Does a personal AI have rights? That is, does an artificial intelligence that either
passes a Turing test or that satisfies (something like) Fletchers 12 characteristics have
rights? For instance, would it have the right not to be a slave?
At first glance, you might think so. But isnt that what most robots are intended to
be? So, if they are persons of either the Turing or the Fletcher variety, do they have
the right not to do what we created them to do? At least one philosopher has suggested
that they do not, that robot servitude is permissible [Petersen, 2007].
By robot servitude, Petersen does not mean voluntary assistance, where you do
something or help someone because you want to, not because you are being paid
to. Nor does he mean slavery, forced work that is contrary to your will. By robot
servitude, he is thinking of robots who are initially programmed to want to serve us,
in particular, to want to do tasks that humans find either unpleasant or inconvenient.
For example, think of a robot programmed to love to do laundry. (On the other hand,
compare this to a dog genetically programmed to want to fetch. This is reminiscent
of the epsilon caste in Aldous Huxleys Brave New World [Huxley, 1932, Ch. 5, 1],
who are genetically programmed to have limited desiresthose destined to be elevator
operators desire nothing more than to operate elevators.)
So, how would robots programmed to want to do unpleasant or humanly
inconvenient tasks different from genetically engineering humans from scratch to
desire to do such tasks? One assumption is that doing this to humans is morally wrong;
hence, doing it to robots would also be morally wrong. Another assumption underlying
this argument is that the two cases are alike.
But on most standard theories of ethics, either this is not wrong or it is not like
the robot case. For instance, Aristotle might argue that engineering humans is wrong
because humans have a particular function or purpose, and it would be wrong to
engineer them away from it. In this case, there is no parallel with robots. In fact, a
robots function might be to do the laundry!
20.7. IMMORAL EXPERIMENTS 683
On a Kantian theory, such an engineered human would either be potentially

autonomous (following its own moral rules that are universally generalizable) or would
not be potentially autonomous. In the first case, it would be wrong to prevent such a
robot from doing laundry, and it would not be harmful to let it do what it autonomously
wants to do. In the second case, we cant do wrong to the robot by having it do our
laundry any more than we can do wrong to a washing machine.
Alternatively, it could be argued that more autonomy is always better. But an
engineered human would only have limited autonomy and therefore could have had
a better life had it not been engineered. And similarly for a robot. (Does this mean that
a Turing machine has less autonomy than a universal Turing machine?)
John Stuart Mill might argue that such an engineered human life would be
unfulfilling, because it substitutes a lower pleasure for a higher one. But whats
wrong with that? Were not taking a potential philosopher and turning her into a
laundry-bot. Rather, were introducing a laundry-bot into the world. Surely, this is
no worse than when a dolphin is born.
What if such a robot could reason its way into renouncing laundry? That would be
OK; otherwise, it would be robot slavery.
20.7 Immoral Experiments

Is the construction of a personal AI (that is, an artificial intelligence that (who?) satisfies
the criteria for being considered a person) an immoral experiment?
Clearly, some experiments are immoral. The existence of institutional review
boards at universities is testament to this. There are many experiments that have
been performed that are now considered to have been immoral (such as WHOSE
experiments in which subjects were told to give what they thought were deadly electric
shocks to people whom they thought were other subjects). And there are many that
are clearly so (such as raising newborns on desert islands to see what kind of language
they might naturally develop; such as experiment was proposed by WHOM?).
The most famousand most relevantliterary example of such an experiment is
the construction of Frankensteins monster. In Mary Shelleys novel, the monster
(who is the most sympathetic character) laments as follows:
Like Adam, I was apparently united by no link to any other being in
existence, but his state was far different from mine in every other respect.
He had come forth from the hands of God a perfect creature, happy and
prosperous, guarded by the especial care of his creator, he was allowed to
converse with, and acquire knowledge from, beings of a superior nature,
but I was wretched, helpless, and alone. Many times I considered Satan
was the fitter emblem of my condition. For often, like him, when I
saw the bliss of my protectors, the bitter gall of envy rose up within
me. . . . Hateful day when I received life! . . . Accursed creator! Why did
you form a monster so hideous that even you turned from me in disgust?
[Shelley, 1818, Ch. 15]
Frankenstein tries to justify his experiment in terms of how it advanced knowlege,
but he realizes that the advancement of knowledge must be balanced against other
considerations:
When younger,. . . I believed myself destined for some great enterprise.

. . . I possessed a coolness of judgment that fitted me for illustrious
achievements. This sentiment of the worth of my nature supported me
when others would have been oppressed; for I deemed it criminal to throw
away in useless grief those talents that might be useful to my fellow-
creatures. When I reflected on the work I had completed, no less a one
than the creation of a sensitive and rational animal, I could not rank myself
with the herd of common projectors. But this thought, which supported
me in the commencement of my career, now serves only to plunge me
lower in the dust. All my speculations and hopes are as nothing; and, like
the archangel who aspired to omnipotence, I am chained in an eternal hell.
My imagination was vivid, yet my powers of analysis and application were
intense; by the union of these qualities I conceived the idea and executed
the creation of a man. Even now I cannot recollect without passion my
reveries while the work was incomplete. I trod heaven in my thoughts,
now exulting in my powers, now burning with the idea of their effects.
From my infancy I was imbued with high hopes and a lofty ambition; but
how am I sunk! Oh! my friend, if you had known me as I once was you
would not recognise me in this state of degradation. Despondency rarely
visited my heart; a high destiny seemed to bear me on until I fell, never,
never again to rise.
Is there ever a situation in which the negative side-effects of an otherwise

praiseworthy goal do not override that goal? (Compare this question with whether
there are ever just wars.) The early cybernetics researcher Norbert Wiener struggled
with this issue:
If we adhere to all these taboos, we may acquire a great reputation as

conservative and sound thinkers, but we shall contribute very little to
the further advance of knowledge. It is the part of the scientistof the
intelligent man of letters and of the honest clergyman as wellto entertain
heretical and forbidden opinions experimentally, even if he is finally to
reject them. [Wiener, 1964, 5].
The basic ethical principle here seems to be what LaChat calls non-maleficence,
or Do No Harm. This is more stringent than beneficence, or Do Good, because
benevolence (doing good) might allow or require doing harm to a few for the benefit
of the many (at least, according to utilitarianism), whereas non-malevolence would
restrict doing good in order to avoid doing harm.
Is creating an artificial intelligence beneficial or not to the AI that is created? Does
creating it do harm to that which is created? One way to think about this is to ask
whether conscious life is better than no life at all. If it isnt, then creating an artificial
life is not a therapeutic experiment, hence not allowable by human-subjects review
boards. Why? Because the subject of the experimentthe artificial person that the
20.8. AIS AND MORALITY 685
experiment will create if it is successful (or, perhaps more so, if it is only partially
successful)does not exist before the experiment is begun, and so the experimenter
is not making it better. Here, we approach the philosophy of existentialism, one of
whose tenets is that existence precedes essence. This slogan means that who you
are, what kind of person you are, is something that is only determinable after you are
born (after you come to exist)and it is not immutable, because, by your actions, you
can change whom you are. The opposite sloganessence precedes existenceis
Aristotelian in flavor: You are a certain kind of person, and cannot change this fact.
On the existentialist view, you exist first, and then you determine what you will be.
Frankenstein did an existential experiment, creating an AI without an essence, and both
Frankenstein and his monster were surprised with the results. On the Aristotelian
view, an essence (something like an Abstraction, as discussed in Ch. 14) is implemented
(or realized). In AI, we canindeed, mustplan out the essence of an entity before
bringing it into existence (before implementing it). In either case, we cant guarantee
that it would come out OK. Hence, creating an AI is probably immoral! So, LaChat
sides with Frankensteins monster, not Frankenstein (or Wiener).
20.8 AIs and Morality

You might think that all of this is a bit silly or, at least, premature. But it is always
better to be prepared: It is better to think about the consequences of our actions while
we have the time and leisure to do so, so that, if those consequences come to be, then
we wont be taken by surprise.
We have looked at whether it is moral to create an AI. Suppose we succeed in doing
so. Could the AI that we create itself be moral?
If AIs are programmed, then one might say that they are not free, hence they are
amoral. This is different from being immoral! Being amoral merely means that
morality is irrelevant to whom or what you are. To oversimplify a bit, good people are
moral, bad people are immoral, a pencil is amoral. The current question is whether AIs
are amoral or not.
And so we have bumped up against one of the Big Questions of philosophy: Is
there such a thing as free will? Do humans have it? Might robots have it? We will not
attempt to investigate this issue here, but merely note that at least one AI researcher,
Drew McDermott, has argued that free will may be a necessary illusion arising from
our being self-aware [McDermott, 2001].
A different perspective has been taken by Eric Dietrich in a pair of papers
[Dietrich, 2001], [Dietrich, 2007]. He argues that robots could be programmed to
be better than humans (presumably because their essence precedes their existence).
Hence, we could decrease the amount of evil in the world by building moral robots and
letting them inherit the Earth.

1. Discuss Bostrums computer simulation argument; see Further Sources of
Inofrmation for Ch. 15.
2. Add to list of movies, etc., re robots:

Mendelsohn, Daniel (2015), The Robots Are Winning!, New York Review
of Books 62(10) (4 June):
5154, http://www.nybooks.com/articles/archives/2015/jun/04/robots-are-
winning/ (accessed 14 June 2015).
A review of the films Her and Ex Machina, but also a history of robots
in literature from The Iliad on.

Bostrom, Nick (2003), Are You Living in a Computer Simulation?, Philosophical
Quarterly 53(211): 243255.
An interesting philosophical analysis not unrelated to Lems story.
Available online, along with many related items, at:
http://www.simulation-argument.com/
Choi, Charles Q. (2008), Not Tonight, Dear, I Have to Reboot, Scientific
American (March): 9497.
Is love and marriage with robots an institute you can disparage? Computing
pioneer David Levy doesnt think sohe expects people to wed droids by
midcentury. Is that a good thing?
Clarke, Roger (1993/1994), Asimovs Laws of Robotics: Implications for
Information Technology:
Part 1, IEEE Computer 26(12) (December 1993): 5361.
Part 2, IEEE Computer 27(1) (January 1994): 5766.
Hofstadter, Douglas R.; & Dennett, Daniel C. (1981), Reflections [on Lems Non
Serviam], in Douglas R. Hofstadter & Daniel C. Dennett (eds.) (1981), The
Minds I: Fantasies and Reflections on Self and Soul (New York: Basic Books):
317320.
La Chat, Michael Ray (2003), Moral Stages in the Evolution of the Artificial
Superego: A Cost-Benefits Trajectory, in Iva Smit, Wendell Wallach, &
Goerge E. Lasker (eds.), Cognitive, Emotive and Ethical Aspects of Decision
Making in Humans and in Artificial Intelligence, Vol. II (Windsor, ON, CANADA:
International Institute for Advanced Studies in Systems Research and Cybernetics):
1824.
A follow-up essay by LaChat, in which he argues that a moral robot will
have to possess sentient properties, chiefly pain perception and emotion, in
order to develop an empathetic superego which human persons would find
necessary and permissible in a morally autonomous AI.
Sparrow, Robert (2007), Killer Robots, Journal of Applied Philosophy 24(1): 62
77.
Should artificial intelligences be allowed to kill? This paper considers the
ethics of the decision to send artificially intelligent robots into war. . . .
Tanaka, Fumihide; Cicourel, Aaron; & Movellan, Javier R. (2007), Socialization
between Toddlers and Robots at an Early Childhood Education Center,
Proceedings of the National Academy of Sciences 104(46) (13 November): 17954
17958.
2. Websites:
(a) On Karel Capek and R.U.R:
i. Karel Capek
http://www.catch22.com/SF/ARB/SFC/Capek,Karel.php3
ii. Josef and Karel Capek

http://www.imagi-nation.com/moonstruck/clsc23.html
iii. Jerz, Dennis G. (2002), R.U.R. (Rossums Universal Robots)
http://jerz.setonhill.edu/resources/RUR/index.html
Discussion of various versions of the play, with numerous links.
Online versions of the play:
A. Capek, Karel (1920), R.U.R.: Rossoms Universal Robots, trans. by
Paul Selver
http://www.mindfully.org/Reform/RUR-Capek-1920.htm
B. Capek, Karel, R.U.R. (Rossums Universal Robots, trans. by David
Wyllie
http://ebooks.adelaide.edu.au/c/capek/karel/rur/
(b) On AI research into personality and emotion:
Reid Simmonss Social Robots Project at Carnegie-Mellon University
http://www.cs.cmu.edu/afs/cs/project/robocomp/social/www/
MITs COG project on humanoid robotics
http://www.ai.mit.edu/projects/humanoid-robotics-group/cog/
Bibliography on the cognitive science of emotion
http://www.cse.buffalo.edu/rapaport/575/emotion.html
Chapter 21
Summary
Version of 3 June 2014
Philosophical reflection. . . is not static, and fixed, but ongoing and

dynamic. The conflict of opinions not only isnt something to worry about,
in fact, it is precisely how things ought to be. . . .
For. . . only after youve considered all sides will you be in a
meaningful position to choose onewhen that time comes to decide.
. . . the philosopher within me cannot make that decision for you. His
job, he reminds me, is merely to rouse the philosopher within you and to
get you thinkingnot to tell you what to think.
Thats your philosophers job.
689
690 CHAPTER 21. SUMMARY

1. Required:
Smith, Brian Cantwell (2002), The Foundations of Computing, in Matthias
Scheutz (ed.), Computationalism: New Directions (Cambridge, MA: MIT Press):
2358.
Earlier version online at:
http://www.ageofsignificance.org/people/bcsmith/print/smith-foundtns.pdf
DeJohn, Jerry; & Dietrich, Eric (2003), Editorial: Subvert the Dominant
Paradigm! A Review of Computationalism: New Directions, edited by Matthias
Scheutz, Journal of Experimental & Theoretical Artificial Intelligence 15(4)
(October-December): 375382, esp. 3.1, pp. 378379.
Scheutz, Matthias (2002), Computation, Philosophical Issues about, Encyclope-
dia of Cognitive Science (London: Macmillan): 604610; online at:
http://staffweb.worc.ac.uk/DrC/Courses%202006-7/Comp%204070/Reading%20Materials/scheutz%20Comp%2
21.2. WHAT IS PHILOSOPHY? 691
Let us take stock of where we have been. MAKE SURE THAT EVERYTHING
HERE IS ACTUALLY DISCUSSED IN THE BOOK!
21.2 What Is Philosophy?

We began with a brief history of western philosophy in terms of its relevance to
computer science. I offered a definition of philosophy as the personal search for truth,
in any field, by rational means. We reviewed some of the principles of rationality, in
particular, of argument analysis, looking at how an argument consists of premises and a
conclusion, how the conclusion might follow validly from the premises if the argument
is truth preserving, and how the argument might be sound if the premises are true. And
we briefly looked at the main branches of philosophy, asserting the existence, for any
field X of a philosophy of X. So, the philosophy of computer science would be the
study of the main goals and fundamental assumptions of computer science.
21.3 What Is Computer Science?

We then turned to an ontological question: What is computer science? We looked
at both political and philosophical motivations for asking the question, and then we
examined various answers: [Newell et al., 1967] said that computer science is the
science of computers and surrounding phenomena such as algorithms. [Knuth, 1974b]
said that computer science is the study of algorithms and surrounding phenomena
such as the computers that they run on. One way of adjudication between these two
apparently opposed viewpoints is to take them as being extensionally equivalent
computer science is the study of algorithms and the computers that execute thembut
intensionally distinct because each focuses on a different portion of computer sciences
single subject matter.
A slightly different view is taken by [Newell and Simon, 1976], who said that
computer science is the artificial science (as opposed to the natural science) of the
phenomena surrounding computers, where both kinds of science are empirical studies,
with the latter studying phenomena that occur in nature and the former studying human-
made artifacts. This can be contrasted with [Shapiro, 2001], who argues that computer
science is a natural science, but not of computers; rather, it is the natural science of
procedures.
Others, such as [Hartmanis and Lin, 1992], add to the subject matter the notion
of information: Computer science is the study of how to represent and process
information and of the machines and systems that do this. Still others, such as
[Brooks, 1996], say that it is not a science at all, but a branch of engineering. And
[Denning et al., 1989], [Denning and Freeman, 2009] say that it is a new kind of
science (neither physical, biological, or social) of natural and artificial information
processes.
21.4 Is Computer Science a Kind of Science or of

Engineering?
So the question of what computer science is breaks down into two smaller questions:
What does it study (as summarized in the previous section)? And how does it study it?
That is, is it a science? Is it a branch of engineering? Or is it neitheris it some other,
or new, kind of study? To answer this, we considered what science and engineering
were.
21.4.1 What Is Science?

We looked at two dichotomies in trying to determine what science is: Is its goal to
describe, or to explain? And are its descriptions or explanations intended to be about
reality (realism) or merely useful summaries that enable us to make predictions
(instrumentalism)?
And what is the scientific method? Is it the experimental method advocated by
Bacon in the 1600s and described by Kemenys loop? That is:
while there is a new fact to observe, do:

begin
observe it;
induce a general hypothesis (to explain or describe it);
deduce future observations;
verify your predictions
end.
Or is it what Popper suggested in the 1950s: Science makes conjectures and

refutations. A statement is scientific if and only if it is falsifiable.
Or is it what Kuhn suggested in the 1960s: Science proceeds by paradigm
revolutions alternating with periods of normal science?
And what about subjects like mathematics, which seems to be scientific, yet is
non-empirical?
21.4.2 What Is Engineering?

Perhaps we need to distinguish between pure and applied sciences; perhaps computer
science is an applied science, or a branch of engineering. We looked at [Davis, 1998]s
history of engineering, which suggests that engineering is defined by its curriculum,
which teaches how to apply science for the use and convenience of people and to
improve the means of production. By contrast, [Petroski, 2003] suggests that the
fundamental activity of engineering is design. But [Loui, 1987] suggests that computer
science is a new kind of engineering that studies the theory, design, analysis, and
implementation of information-processing algorithms.
21.5. WHAT DOES COMPUTER SCIENCE STUDY? 693
21.5 What Does Computer Science Study?

So, whether it is a science, a branch of engineering, or something else, what does it
study? Does it study two things: computing and computers? Or are these merely two
aspects of a single, underlying subject matter? To answer this, we asked four questions:
21.5.1 What Is a Computer? Historical Answer

We saw that there were two intertwined branches of the history of computers, each
of which had a slightly different goal: The goal of one branch was to build a
computing machine. This is the history in which prominent roles were played by
Pascal, Leibniz, Babbage, Aiken, Atansoff and Berry, Turing, Eckert and Mauchly,
and von Neumann, among others. The goal of the other branch was to provide a
foundation for mathematics. This is the history in which prominent roles were played
by Leibniz (again), Boole, Frege, Russell, Hilbet, Turing (again), Church, and Godel,
among others.
21.5.2 What Is an Algorithm? Mathematical Answer

We began this investigation by asking what computation is. A function f is computable
means by definition that there is an algorithm that computes f , that is, that there is an
algorithm A such that, for all input i, A(i) = f (i), and such that A specfies how f s input
and output are related.
So, what is an algorithm? Roughly, an algorithm for a problem P is, by definition,
a finite procedure (that is, a finite set of explicit instructions) for solving P that is
(i) unambiguous for the computer or human who will execute itthat is, all steps of the
procedure must be clear and well-defined for the executorand (ii) it must eventually
halt, outputting a correct solution to P. Of course, this is only a rough definition:
Spelling out exactly what clear and well-defined means was an accomplishment of
the highest order.
We looked at the most successful solution to this problem of what an algorith is,
namely, the Turing machine, as well as the history of the term computable. We
discussed Turings thesis that a function is computable if and only if it is computable by
a Turing machine, as well as Churchs thesis that a function is computable if and only
if it is lambda-definable. That both of these notions turn out to be logically equivalent
to each other as well as to other analyses, such as that of recursive functions, allows us
to talk of the Church-Turing Computability Thesis.
21.5.3 What Is a Computer? Philosophical Answer

Armed with the history of computers and the mathematics of computation, we returned
to the question of what a computer is. We began our philosophical investigation with
[Searle, 1990]s claim that everything is (interpretable as) a digital computer. And we
looked at some alternatives: that a computer is magic paper that can take as input
patterns that describe changes to themselves and to other patterns, and that causes the
described changes to occur [Hayes, 1997], that a computer is a device that changes
values of variables [Thomason, 2003] (see 21.7.2, (C4), below), and that the universe
is a computer [Lloyd and Ng, 2004], [Wolfram, 2002].
21.5.4 What Is an Algorithm? Philosophical Answer

And we then returned to the question of what an algorithm is, now from a philosophical
point of view. This divided into three questions:
21.5.4.1 What Is a Procedure?
We saw that [Shapiro, 2001] argued for the more general notion of a procedure as
being the subject matter of computer science. So we looked at what some philosophers
have said about procedures.
[Cleland, 1993] (and in subsequent papers) argued that mundane procedures
(such as causal processes, including recipes) are effective procedures that are not
computable by Turing machines, because their effectiveness depends on conditions
in the external world. And [Preston, 2006] pointed out important differences between
improvisational recipes (and music) and precise algorithms, suggesting that recipes are
more like specifications of programs than they are like computer programs.
21.5.4.2 What Is Hypercomputation?
Next, we looked at the idea that there might be functions that can be computed in
some more general sense than by Turing machines, what is called hypercomputation.
We looked at Turings oracle machines, Boolos and Jeffreys infinitely accelerating,
Zeus machines, Wegners interaction machines and Putnams and Golds trial and
error machines (which are TMs that can change their mind: it is the last answer that
matters, not the first one), which [Kugel, 2002] argued are necessary in order for AI to
succeed.
21.5.4.3 What Is a Computer Program?
This led us to the third part of our investigation of algorithms: their implementation in
computer programs. Here we went deep into the depths of the philosophy of computer
science, looking at implementation, software, legal protections for software, and the
relation of programs to scientific theories and to the real world.
21.5.4.3.1 What Is Implementation? [Chalmers, 2011] (and elsewhere) argued

against [Searle, 1990] that implementation is an isomorphism and that a computer
is an implementation of a computation or of a universal Turing machine. And
[Rapaport, 1999] argued that implementation is the semantic interpretation of an
Abstraction in some medium. To understand this, we looked at the syntax and
semantics of formal systems.
21.5. WHAT DOES COMPUTER SCIENCE STUDY? 695
21.5.4.3.2 What Is Software? Paralleling the computer-computation distinction is

the hardware-software distinction. According to [Moor, 1978], one can understand
computers as physical objects or on a symbolic level (and we compared these two levels
of understanding to [Dennett, 1971]s three levels: the physical, design, and intentional
stances). A computer program, for Moor, is a set of instructions that a computer can
execute. But the notion of software is relative to both a computer and a person: X is
software for computer C and person P if and only if X is a computer program for C that
is changeable by P. Hardware is similarly relative: X is hardware for C and P if and
only if X is (physically) part of C, and X is not software for C and P.
In contrast, [Suber, 1988] argued that software is simply syntactic form, and
[Colburn, 1999] argued that it is a concrete abstraction that has an abstract medium
of description (a text in a formal language) and a concrete medium of execution
(circuits and semiconductors).
We also looked at how the software-hardware distinction compared to Spinozas
dual-aspect theory concerning the mind-brain distinction, and whether any of the
interpretations of Spinozas theory might apply to the computer case.
21.5.4.3.3 Can (Should) Software Be Patented, or Copyrighted? In order to try

to understand the software-hardware relationship, we looked at the issue of whether
software could, or should, be patented or else copyrighted. After all, if a program is a
piece of text, then copyright is appropriate. But if a program is a piece of hardware,
then patent is appropriate. But programs seem to be both, yet nothing can (legally) be
both patented and copyrighted.
To resolve this paradox, [Newell, 1986] suggested that we need to devise
good models (ontologies) of algorithms and other computational entities. And
[Koepsell, 2000] suggested that we need to revise the models of legal protection.
21.5.4.3.4 Are Programs Scientific Theories? As part of our endeavor to

understand what a program is, we considered the claim made by several philosophers
and computer scientists that some programs are scientific theories.
Programs are (a language for expressing) theories, which can then be their own
models. We looked at the differences and relations between theories and models,
simulations and the real thing, and simulation vs. emulation. We also briefly looked
at philosophical theories of scientific explanation and of scientific models.
21.5.4.3.5 What Is the Relation of Programs to the World? A final question

related to what computer programs are is how they relate to the real world. Here,
we looked at [Smith, 1985], who argued that there is a gap between the world and our
models of it, and that computers rely on models of the world, but must act in the real
world.
One related issue here concerns [Cleland, 1993]s views about the Church-Turing
Computability Thesis: Could the computability of a problem depend in part on the real
world, and not exclusively on the program for solving it?
Another related issue concerns whether Smiths observations pertain only to
computers or also to us humans. After all, we must act in the real world, but perhaps
our actions are also based on mental models of the world that are separate from the real
world.
21.6 Computer Ethics

We looked at two topics in computer ethics.
21.6.1 Are There Decisions Computers Should Never Make?

Given the limitations on the verifiability of program correctness, it becomes important
to ask this question, first asked in [Moor, 1979].
His answer has two parts: First, no, there are no decisions that computers should
never make as long as their track record is better than that of humans. After all, the
question seems to be logically equivalent to this one: Are there decisions that should
not be made rationally? Presumably, computer programs that make decisions make
them on the basis of rational evaluation of the facts. And surely we always want to
make our decisions rationally.
Second, it is up to us to accept or reject the decisions made by a computer. This is
the case no more and no less than it is for decisions made (for us) by others. Thus, the
decision made by a computer should be evaluated in the same way as advice offered by
an expert or in a textbook.
On the other hand, [Friedman and Kahn, 1997] argue that, yes, there are decisions
that computers should never make, on the grounds that only humans are capable of
being moral agents. Whether that is really the case is part of the second issue in
computer ethics (see 21.6.2). But to err is human, as shown in the case of the airline
crash caused by following a humans decision instead of a computers. This must be
contrasted with cases in which tragedy occurred by blindly following a computers
decision, as in the radiation scenario.
21.6.2 Should We Build an Artificial Intelligence?

The second issue in computer ethics that we looked at was, given the fact that it is
plausible that we could build an AI, it becomes important to ask whether we should
build one.
[Lem, 1971] pointed out that, if we do succeed in building AIs, we may someday
have to pull the plug on them. And [LaChat, 1986] suggested that maybe we shouldnt
even begin. But [LaChat, 1986] also argued that considering the possibility of building
one and considering the moral consequences of doing so enables us to deal with
important issues such as: What is a person? Would an AI with personhood have (moral)
rights? (See 21.6.1, above.) And could the AI itself be a moral agent?
21.6.3 Philosophy of AI
Our final topic concerned the philosophy of artificial intelligence, in which we looked
at the Turing test and the Chinese Room Argument.
21.7. WHAT IS THE PHILOSOPHY OF COMPUTER SCIENCE? 697
[Turing, 1950]s article that introduced what is now called the Turing test suggested
that a computer will be said to be able to think if we cannot distinguish its cognitive
behavior from that of a human.
[Searle, 1980]s rebuttal argued that a computer could pass a Turing test without
really being able to think.
We then looked at how multiple implementation and how syntax can suffice
for semantic interpretation of the kind needed for computational cognition might
overthrow Searles position.
21.7 What Is the Philosophy of Computer Science?

Lets now return to our original question and examine it in the light of an essay directed
to that issue by Brian Cantwell Smith, [Smith, 2002].1
21.7.1 The Major Question

The major question in the philosophy of computer science, according to Smith, is:
What is computation? And the answer is that it is a lot more than just the theory of
Turing machines.
Smith says that computing is:
empirical: Computation in the wild includes every aspect of the nature and use of
computing and computers, presumably including things like iPads and Facebook.
conceptual: Computing and computers are intimately related to issues in semantics

such as interpretation and representation. This is probably where the theory of
Turing machines fits in.
cognitive: The theory of computation must provide a foundastion for a theory of mind
(that is, for what I have called computational cognitive science). This is probably
where the Turing test and the Chinese Room Argument fit in.
21.7.2 9 Theses about Computation

Smith offers 9 theses about computation, each of which is worthy of a book-length
discussion. Here, I will present them, not in the order in which Smith does, but in what
seems to me to be a more logical order (but I retain Smiths numbering scheme).
(C2) A theory of computation needs a full theory of semantics and intentionality.

By intentionality, I think Smith means what the 19th-century cognitive scientist
Franz Brentano meant, namely, directedness to an object, that is, being about
something.2
1 Fora critique of [Smith, 2002], see [DeJohn and Dietrich, 2003], esp. 3.1, pp. 378379.
2 On intentionality, see [Jacob, 2010]. On Brentano, see [Huemer, 2010]. On directedness and
aboutness, see [Rapaport, 2012a].
(C1) However, none of the following provide such a theory:
1. effective computability: the study of what can, cant, or is impractical to

be done by an abstract machine
2. execution of algorithms, or rule followingincluding recipes
3. calculation of functions: the behavior of taking input and producing
output
4. digital state machines: that is, automata with finite sets of states
5. information processing: that is, storing, manipulating, and displaying
information.3
6. physical symbol systems: that is, the claim that a computers interaction
with symbols depends on its physical implementation
7. formal symbol manipulation by a machine without regard to meaning:
(C3) Formal symbol manipulation separates the syntactic and
semantic domains.
But real-world computational processes are participatory
That is, the syntax and semantics of real-world computational
processes intersect in two ways:
(a) As argued in [Smith, 1985], computers get their input from, and
must act in (that is, produce output to), the real world.
(b) Computers are part of the real world:
i. Some computer-internal symbols refer to other computer-
internal symbols; others refer to external objects.
ii. Some computer-external symbols refer to computer-internal
ones; others refer to other external ones.
(C6) Computation is not formal in any of the following senses of formal:
anti-semantic: the sense in which syntax is independent of
semantics
syntactic: for example, proof theory only relies on pattern
matching and symbol manipulation.
determinate or well-defined (that is, not vague): for example,
digital or discrete
mathematical: I think Smith has in mind here the kind of formality
that Bertrand Russell was discussing in this remark:
Pure mathematics consists entirely of such assertions as
that, if such and such a proposition is true of anything, then
such and such another proposition is true of that thing. It is
essential not to discuss whether the first proposition is really
true, and not to mention what the anything is, of which it is
supposed to be true. . . . If our hypothesis is about anything,
3 On various meanings of information, see [Piccinini and Scarantino, 2011] and
http://en.wikipedia.org/wiki/Information
21.7. WHAT IS THE PHILOSOPHY OF COMPUTER SCIENCE? 699
and not about some one or more particular things, then our
deductions constitute mathematics. Thus mathematics may
be defined as the subject in which we never know what
we are talking about, nor whether what we are saying is
true. [Russell, 1917, Ch. 4, p. 75; my boldface]4
analytic method: This might be related to [Simon, 1962]s theory
of how to analyze complexity hierarchically, that is, recursively.
(C4) Turing machines dont use marks to represent numbers; they use numbers to
represent marks!
Recall that syntax concerns mark-manipulation, as we saw in CITE. (This
is the kind of mathematics that Russell was discussing in the above
quote.) The real world of mathematics concerns number-manipulation.
And semantics is the study of the relation between them.
But, according to Smith,5 In science, we use numbers to represent stuff;
computation only studies marks.
A corollary is this: The theory of computation is a theory [that is, a language]

of how. . . patches of the world in one physical configuration [can] change
into another physical configuration [Smith, 2002, 42]. (Compare this to
Thomasons definition of a computer, 21.5.3, above.) That is:
(C5) Computability theory is a mathematical theory of the flow of causality.
(C7) A theory of computation needs a full theory of ontology. (For a beginning, see
[Smith, 1996]. For critical reviews, see [Koepsell, 1998], [Loui, 1998].)
(C8) Computers are not natural kinds. (But recall that [Shapiro, 2001] argued
that procedures are natural kinds.)
There can be no theory of computation.
(C9) Therefore, the existence of computation is extremely important, because any
theory of it will be a theory of intentional artifacts, hence a theory of everything!
And that seems to be a good note on which to end.
4 The passage humorously concludes: People who have been puzzled by the beginnings of mathematics
will, I hope, find comfort in this definition, and will probably agree that it is accurate.
5 In a lecture given at UB in 20xx??.
21.8 NOTE FOR NEXT DRAFT

1. See discussion of real computers, which are interaction machines, with
theoretical computers, in 20, above.
Chapter 22
Syntax and Semantics
Version of 19 March 2014

USE http://www.cse.buffalo.edu/rapaport/563S05/synsem.html,
BUT ALSO USE Chs. 2 and 3 from
http://www.cse.buffalo.edu/rapaport/Papers/book.pdf
and their published version:
Rapaport, William J. (1995), Understanding Understanding: Syntactic

Semantics and Computational Cognition, in James E. Tomberlin (ed.), AI,
Connectionism, and Philosophical Psychology, Philosophical Perspectives
Vol. 9 (Atascadero, CA: Ridgeview): 49-88; reprinted in Toribio, Josefa, &
Clark, Andy (eds.) (1998), Language and Meaning in Cognitive Science:
Cognitive Issues and Semantic Theory, Artificial Intelligence and Cognitive
Science, Vol. 4: Conceptual Issues (New York: Garland): 7388.
And Ch. 3 of Understanding Understanding,

http://www.cse.buffalo.edu/rapaport/Papers/book.pdf
And also use [Suber, 1997a].
701
702 CHAPTER 22. SYNTAX AND SEMANTICS
This Is Not the End
Lots of things never end. Space. Time. Numbers. The questions little kids ask.1
And philosophy.
You try to convince somebody of somethingeven yourselfby offering reasons
to believe the thing. But then your belief is only as valid as your reasons are, so you
offer reasons to accept your reasons. But then those reasons need further reasons and
youre off. As a result it often seems that there arent any answers to philosophical
questions: there are just more arguments, more objections, more replies. And so it may
easily seem that its not worth even getting started. Why bother? Youll never finish.
You may as well try to count all the numbers.
But there is another way of thinking about it.
I went snorkeling for the first time a few years ago. It was an amazing experience.
There was a whole world under that water to which Id been oblivious my entire life.
This world was populated with countless amazing creatures with all sorts of complex
relationships to each other in that tangled ecosystemic way. Indeed every single thing
was connected to every other thing: this one is food for that one, which excretes
chemicals used by another one, which excretes waste products used by others, and
so on. Stunning, fascinating, and absolutely, deeply, beautiful. It had been there all
along, jsut waiting for me to dive in.
If you were now to tell me that that ocean goes on forever, filled with ever more
amazing creatures in more amazing relationshipsI wouldnt say, Well then why
bother entering? Rather, Id say, Where can a guy get a wetsuit around here?
But that is philosophy. Its filled with countless amazing ideas, concepts, beings,
which exist in all sorts of complex logical relationships with each other. And unlike
the actual ocean this one is infinitely deep: Wherever you enter you can keep going,
and going, and going. What you should be thinking, then, is not: Why enter? It is,
rather, this: thank youvery much.
But of course, that world just is this world, the world that youre in. This great
ocean you may be looking for, youre alrady in it. You just have to start thinking about
it. The very first drop in that bucket is a splash into the infinite.
1 The cells on a Turing-machine tape. Infinite loops.items added by WJR.
703
704 CHAPTER 22. SYNTAX AND SEMANTICS
This is the beginning.

Appendix A
Argument Analysis and

Evaluation
Version of 7 July 2014
[Sidney Harris cartoon showing boy sitting in front of a computer,

thinking:]
Liebniz, Boole and Godel worked with logic.

I work with logic.
I am Leibniz, Boole and Godel.
American Scientist 73 (January-February): 19.
705
706 APPENDIX A. ARGUMENT ANALYSIS AND EVALUATION
A.1 Recommended Readings:

Required:
Chudnoff, Elijah (2007), A Guide to Philosophical Writing, (Cambridge, MA:
Harvard Center for Expository Writing).
http://www.fas.harvard.edu/phildept/files/GuidetoPhilosophicalWriting.pdf
A.2. INTRODUCTION 707
A.2 Introduction
In 2.3, I said that the methodology of philosophy involved rational means for
seeking truth, and in 2.6, we looked at different kinds of rational methods. In this
appendix, well look more closely at one of those methods: argument analysis and
evaluation.
Unless you are gullible, willing to believe everything you read or anything that an
authority figure tells you, you should want to know why you should believe something
that you read or something that you are told. If someone tells you that some proposition
C is true because some other propositions R1 and R2 are true, you might then consider,
first, whether those reasons (R1 and R2 ) really do support the conclusion C and, second,
whether you believe the reasons.
So, lets consider how you might go about doing this.
A.3 A Question-Answer Game

Consider two players, Q and A, in a question-answer game:
Q asks a question (see step A.3, below).
A gives an alleged answer (see step A.3, below).
Q must verify As alleged answer (see steps A.3A.3, below).
Step 1 Q asks whether C is true.

Step 2 A answers: C, because R1 and R2 .
That is, A gives an argument for conclusion C with reasons (also
called premises) R1 and R2 .
Note, by the way, that this use of the word argument has nothing
directly to do with the kind of fighting argument that you might have
with your roommate; rather, its more like the legal arguments that
lawyers present to a jury.
Also, for the sake of simplicity, Im assuming that A gives only two
reasons for believing C; in a real case, there might be only one reason
(for example: Fred is a computer scientist; therefore; someone is a
computer scientist), or there might be more than two reasons (for
examples, see any of the arguments for analysis and evaluation in
Appendix B).
Step 3 Q must analyze or verify As arguments. Q can do this by asking three
questions:
Do I believe (that is, do I agree with) R1 ?
Do I believe (that is, do I agree with) R2 ?
Does C follow validly from R1 and R2 ?
There are a few comments to make about these steps. First, steps A.3
and A.3 are recursive: For each of them, Q could ask A (or
someone else!) whether Ri is true, and A (or the other person) could
give an argument for conclusion Ri with new premises R3 and R4 .
Clearly, this process could continue (this is what toddlers do when
they continually ask their parents Why?!). It is an interesting
philosophical question, but fortunately beyond our present scope, to
consider where, if at all, this process might stop.
Second, to ask whether C follows validly from the premises is
to assume that As argument is a deductive one. For the sake of
simplicity, all (or at least most) of the arguments in Appendix B
are deductive, but, in real life, most arguments are not completely
deductive, or not even deductive at all. So, more generally, in
Step A.3, Q should ask whether C follows rationally from the
premises: If it does not follow deductively, does it follow inductively?
Abductively? And so on.
Third, unlike the first steps of considering the truth value of the
premises, this step of determining whether the relation between the
premises of an argument and its conclusion is a rational one is not
similarly recursive, on pain of infinite regress. The classical source
of this observation is due to Lewis Carroll (of Alice in Wonderland
fame).1 Carroll was a logician by profession, and wrote a classic
philosophy essay involving Achilles and the Tortoise [Carroll, 1895].
Finally, it should be pointed out that the order of doing these steps
is irrelevant. Q could first analyze the validity (or rationality) of the
argument and then analyze the truth value of (that is, decide whether
to agree with) the premises, rather than the other way round.
Step 4 Having analyzed As argument, Q now has to evaluate it, by reasoning in one
of the following ways;
If I agree with R1 ,
and if I agree with R2 ,
and if C follows validly (or rationally) from R1 and R2 ,
then I logically must agree with C.
But, bf if I really dont agree with C,
then I must reconsider my agreement either:
with R1 ,
or with R2 ,
or with the logic of the inference from R1 &R2 to C.
If I agree with R1 ,
and if I agree with R2 ,
but the argument is invalid, is there a missing premisean extra reason (see
below)that would validate the argument and that I would agree with?
1 More properly known as Alices Adventures in Wonderland and Through the Looking Glass.
A.4. MISSING PREMISES 709
[(i)]
If so, then I can accept C,
else I should not yet reject C,
but I do need a new argument for (that is, a new set of reasons for
believing) C.
If I disagree with R1 or with R2 (or both),
bf then this argument is not a reason for me to believe C;
so, I need a new argument for C.
There is one other option for Q in this case: Q might want to
go back and reconsider the premises. Maybe Q was too hasty in
rejecting them.
What if Q cannot find a good argument for believing C? Then it might
be time to consider whether C is false. In that case, Q needs to find an
argument for Cs negation: Not-C (sometime symbolized C).
This process of argument analysis and evaluation is summarized in the flowchart in
Figure A.1.
A.4 Missing Premises

One of the trickiest parts of argument analysis can be identifying missing premises.
Often, this is tricky because the missing premise seems so obvious that youre not
even aware that its missing. But, equally often, its the missing premise that can make
or break an argument.
Heres an example from the Textual Entailment Challenge, a competition for
researchers interested in knowledge representation and in information extraction.2 In a
typical challenge, a system is given one or two premises and a conclusion (to use our
terminology) and asked to determine whether the conclusion follows from the premise.
And follows is taken fairly liberally to include all kinds of non-deductive inference.
Here is an example:
Premise 1 (P): Bountiful arrived after wars end, sailing into San Francisco Bay
21 August 1945.
Premise 2: Bountiful was then assigned as hospital ship at Yokosuka, Japan, departing
San Francisco 1 November 1945.
Conclusion (C): Bountiful reached San Francisco in August 1945.
The idea is that the two premises might be sentences from a news article, and the
conclusion is something that a typical reader of the article might be expected to
understand from reading it.
I hope you can agree that this conclusion does, indeed, follow from these premises.
In fact, it follows from Premise 1 alone. Premise 2 is a distractor, in this case.
But what rule of inference allows us to infer C from P?
2 For more information on extual entailment in general, and the Challenge in particular, link to:
http://pascallin.ecs.soton.ac.uk/Challenges/RTE3/ (accessed accessed 7 January 2013).

P talks of arrival and sailing into, but C talks only of reaching
P talks of San Franciso Bay, but C talks only of San Francisco
There are no logical rules that connect these concepts.

Most people, I suspect, would think that no such rules would be needed; after all,
isnt it obvious that, if you arrive somewhere, then you have reached it? And isnt it
obvious that San Francisco Bay must be in San Francisco?
Well, maybe. But, whereas people might know these things, computers wont,
unless we tell them. In other words, computers need some lexical knowledge and some
simple geographical knowledge. (If you dont like the word knowledge here, you can
substitute information.)
So, we need to supply some extra premises that link P with C more closely. These
are the missing premises. And the argument from P to C is called an enthymeme
(because the missing premises are in (Greek en-) the arguers mind (Greek
thymos)).
So, we might elaborate the argument as follows (there are other ways to do it; this
is one that comes to my mind):
P: Bountiful arrived after wars end, sailing into San Francisco Bay 21 August 1945.
Pa : If something sails into a place, then it arrives at that place.
C1 : Bountiful arrived at San Francisco Bay 21 August 1945.
In this first step, Ive added a missing premise, Pa , and derived an intermediate
conclusion C1 . Hopefully, you agree that C1 follows validly (or at least logically in
some way, that is, rationally) from P and Pa .
We have no way of knowing whether P is true, and must, for the sake of the
argument, simply assume that it is true. (Well, we could look it up, I suppose; but
were not asked if the argument is sound (see below), only if it is valid: Does C
follow from P?)
Pa , on the other hand, doesnt have to be accepted at all; after all, we are imposing
it on the (unknown) author of the argument. So, we had better impose something that
is likely to be true. Pa is offered as part of the meaning of sail into. I wont defend
its truth any further here, but if you think that its not true, then you should either reject
the argument or else find a better missing premise.
We might have chosen a slightly different missing premise:
Pb : If something arrives in a place named X Bay, then it arrives at a place named X.
C2 : Bountiful arrived at San Francisco 21 August 1945.
C2 will follow from C1 and Pb , but is Pb true? Can you think of any bays named X
Bay that are not located in a place named X? If you can, then we cant use Pb . Lets
assume the worst: Then well need something like:
Pb .1: If something arrives in San Francisco Bay, then it arrives at San Francisco.
A.4. MISSING PREMISES 711
C2 will follow from C1 and Pb .1, and we can easily check the likely truth of Pb .1 by
looking at a map.
So far, so good. Weve now got Bountiful arriving at San Francisco on 21 August
1945. But what we need is Bountiful reaching San Francisco in August 1945. So
lets add:
Pc : If something arrives somewhere, then it reaches that place.
Again, this is proposed as an explication of part of the meaning of arrive, and, in

particular, of that part of its meaning that connects it to C.
From Pc and C2 , we can infer:
C3 : Bountiful reached San Francisco 21 August 1945.
Are we done? Does C3 = C? Look at them:
C: Bountiful reached San Francisco in August 1945.
Think like a computer! What do you need to know in order to know whether C3 = C?
You need to know whether the final missing premise, Pd , is true:
Pd : If something occurs (on) DATE MONTH YEAR, then it occurs in MONTH

YEAR.
And thats true by virtue of the way (some) people talk. From Pd and C3 , we can infer
C.
So, the simple argument that we started with, ignoring its irrelevant premise,
becomes this rather more elaborate one:
P: Bountiful arrived after wars end, sailing into San Francisco Bay 21 August 1945.
Pa : If something sails into a place, then it arrives at that place.
C1 : Bountiful arrived at San Francisco Bay 21 August 1945.
Pb : If something arrives in a place named X Bay, then it arrives at a place named X.

(or Pb .1: If something arrives in San Francisco Bay, then it arrives at San
Francisco.)
C2 : Bountiful arrived at San Francisco 21 August 1945.
Pc : If something arrives somewhere, then it reaches that place.
Pd : If something occurs (on) DATE MONTH YEAR, then it occurs in MONTH

YEAR.
C: Bountiful reached San Francisco in August 1945.

A.5 When Is an Argument a Good Argument?

As we have seen, Q needs to do two things in analyzing and evaluating an argument:
1. deciding whether the premises are true (that is, deciding whether to agree with,
or believe, the premises), and
2. deciding whether the inference (that is, the reasoning) from the premises to the
conclusion is a good one.
That is, there are two separate conditions for the goodness of an argument:
1. factual goodness
2. logical goodness
Factual goodnesstruthis beyond the scope of logic, although it is definitely not

beyond the scope of deciding whether to accept the conclusion of an argument. As we
saw in 2.4, there are several ways of defining truth and of determining whether a
proposition is true. Two of the most obvious (though not the simplest to apply!) are
(1) constructing a (good!) argument for a proposition whose truth value is in question
and (2) making an empirical investigation to determine its truth-value (for instance,
performing some scientific experiments or doing some kind of scholarly research).
Logical goodness (for deductive arguments) is called validity. I will define this
in a moment. But, for now, note that these two conditions must both obtain for an
argument to be really good: A really good argument is said to be soundan
argument is sound if and only if it is both valid and factually good, that is, if and
only if it is both valid and all of its premises really are true.
Just to drive this point home: If the premises of an argument are all true (or if you
believe all of them), that by itself does not make the argument sound (really good).
For one thing, your belief in the truth of the premises might be mistaken. But, more
importantly, the argument might not be valid.
And, if an argument is valid, that by itself does not make the argument sound
(really good). All of its premises need to be true, also; that is, it needs to be factually
good.
So, what does it mean for a (deductive) argument to be valid? An argument is valid
=d e f it is truth-preserving. That is, an argument is valid if and only if, whenever all
of its premises are true, then its conclusion must also be true. Heres another way to
put it: An argument is valid if and only if it is impossible for all of its premises to be
simultaneously true, yet its conclusion is false. Note that none of these formulations
say that the premises are true. They only say that, if the premises were to be true,
then the conclusion would preserve that truthit would inherit that truth from the
premisesand so it would also (have to) be true.
Thats fine as far as it goes, but it really isnt very helpful in deciding whether an
argument really is valid. How can you tell if an argument is truth-preserving? There is
a simple, recursive definition, but, to state it, well need to be a bit more precise in how
we define an argument.
A.5. WHEN IS AN ARGUMENT A GOOD ARGUMENT? 713
Definition 1: An argument from premises P1 , . . . , Pn to conclusion C isdef a

sequence of propositions hP1 , . . . , Pn ,Ci, where C is alleged to follow logically
from the Pi .
Definition 2: An argument is valid (or: is truth-preserving) if and only if it is an

argument in which each proposition Pi or conclusion C either:
is a tautology
or is a premise
or follows validly from previous propositions in the sequence by one or
more rules of inference.
This needs some commenting! First, what is a tautology? A tautology is

a proposition that must always be true. How can that be? Most tautologies are
(uninformative) logical truths, such as Either P or not-P, or If P, then P. (Note
that, if P is true (or, if you believe P), then Either P or not-P has to be true (or, you
are logically obligated to also believe Either P or not-P), and, if P is false, then
not-P is true, and so Either P or not-P still has to be true (or, you are logically
obligated to also believe Either P or not-P); so, in either case, the disjunction has
to be true (or, you are logically obligated to believe it. Similar considerations hold
for If P, then P.) Sometimes, statements of mathematics are also considered to
be tautologies (whether they are informative or not is an interesting philosophical
puzzle; see [Wittgenstein, 1921]).
Second, a premise is one of the initial reasons given for C, or one of the missing
premises added later. Premises, of course, need not be true, but, when evaluating an
argument for validity, we must assume that they are true for the sake of the argument.
Third, clause A.5 might look circular, but it isnt; rather, its recursive. In fact, this
entire definition is recursive. The base cases of the recursion are the first two clauses:
Tautologies must be true, and premises are assumed to be true. The recursive case
consists of rules of inference, which are sub-proofs that are clearly valid (truth-
preserving) when analyzed by means of truth tables.
So, what are these primitive valid arguments known as rules of inference? The
most famous is called Modus Ponens:
From P
and If P, then C,
you may validly infer C.
Why may you validly infer C? Because the truth table for If P, then C says that
that conditional proposition is false in only one circumstance: when its antecedent
(P) is true and its consequent (C) is false; in all other circumstances, the conditional
proposition is true. So, if the antecedent of a conditional is true, and the conditional
itself is true, then its consequent must also be true. Modus Ponens preserves truth.
Another important rule of inference is called Universal Elimination (or Universal
Instantiation):
From For all x, F(x),

you may infer F(a), for any individual a in the domain of discourse (that
is, in the set of things that you are talking about).
A truth-table analysis wont help here, because this is a rule of inference from first-
order predicate logic, not from propositional logic, and the formal definition of truth
for first-order predicate logic is beyond our scope, but it should be pretty obvious that,
if it is true that everything in the domain of discourse has some property F, then it must
also be true that any particular thing in the domain (say, a) has that property. (For more
rules of inference and for the formal definition of truth in first-order predicate logic,
see any good introductory logic text.)
NEED A SECTION WITH EXAMPLES OF
VALID/SOUND/INVALID/UNSOUND ARGS.
A.6 Summary
So, to analyze an argument, you must identify its premises and conclusion, and supply
any missing premises to help make it valid. To evaluate the argument, you should then
determine whether it is valid (that is, truth preserving), and then decide whether you
agree with its premises.
If you agree with the premises of a valid argument, then you are logically obligated
to believe its conclusion. If you dont believe its conclusion, even after your analysis
and evaluation, then you need to revisit both your evaluation of its validity (maybe
you erred in determining its validity) as well as your agreement with its premises: If
you really disagree with the conclusion of a valid argument, then you must (logically)
disagree with at least one of its premises.
You should try your hand at analyzing and evaluating the arguments in Appendix B!
A.7 NOTES FOR NEXT DRAFT

1. MISCELLANEOUS THOUGHTS ABOUT PROOFS
(a) There are lots of ways to have invalid arguments!

(b) More importantly, it is possible to have an invalid argument whose
conclusion is true. Heres an example:
All birds fly. (true)

Tweety the canary flies. (true)
------------------------
Therefore, Tweety is a bird. (true)
This is invalid, despite the fact that both premises and the conclusion are
true: It is invalid, because the argument form can have true premises and a
A.7. NOTES FOR NEXT DRAFT 715
false conclusion:
x[P(x) Q(x)
R(a) Q(a)
P(a)
Heres a counterexample, that is, an argument with this form that has true
premises but a false conclusion:
All birds fly.

Bob the bat flies.
------------------
Therefore, Bob is a bird.
Just having a true conclusion doesnt make an argument valid. And such an
argument doesnt prove its conclusion (even though the conclusion is true).
(c) Can any proposition (or its negation) be proved?
That is, given a proposition P, we know that either P is true or else P is
false (that is, that P is true). So, whichever one is true should be provable.
Is it?
No!
First, there are propositions whose truth value we dont know yet. For
example, once upon a time, no one knew if Fermats Last Theorem could
be proved. Thats the proposition that the equation
xn + yn = zn
has no solutions for integers x, y, z and integer n > 2. But it has now been
proved.3
For another example, no one knows (yet) if Goldbachs Conjecture is true.
Goldbachs Conjecture says that all positive even integers are the sum of 2
primes; for example, 28 = 5 + 23. Expressed in first-order logic, this would
be:
x[Integer(x) (x > 0) Even(x) yz[Prime(y) Prime(z) (x = y+z)]]
For yet another example, no one knows (yet) if the Twin Prime Conjecture
is true. The Twin Prime Conjecture says that there are an infinite number
of twin primes, that is, primes m, n such that n = m + 2; for example,
2 and 3, 3 and 5, 5 and 7, 9 and 11, 11 and 13, etc. Expressed in first-order
logic, this would be:
mn[Prime(m) Prime(n) (n = m+2) xy[Prime(x) Prime(y) (y = x+2) (y > m) (x > n)]]
Secondand much more astounding than our mere inability so far to prove
or disprove any of these conjecturesthere are propositions whose truth
3 See http://en.wikipedia.org/wiki/Fermats Last Theorem
value is known to be true, but which we can prove that we cannot prove!
This is the essence of Godels Incompleteness Theorem. Stated informally,
it asks us to consider this proposition, which is a slight variation on the Liar
Paradox (that is the proposition This proposition is false: If its false, then
its true; if its true then its false):
(G) This proposition is true but unprovable.
We can assume that G is either true or false. Is it false? If it is false, then
it is provable. But any provable proposition has to be true (because proofs
are truth-preserving). So it isnt false. Therefore, it must be true. But if its
true, then its unprovable. End of story; no paradox!
That is, there are true propositions (moreover, Godel showed that they are
propositions that are true in the mathematical system consisting of first-
order predicate logic plus Peanos axioms; that is, they are true propositions
of arithmetic!) that cannot be proved. (For more information on Godels
proof, see [Nagel et al., 2001], [Hofstadter, 1979], and [Franzen, 2005].)
2. Application of Predicate Logic to AI
Heres an application of predicate logic to artificial intelligence (AI).
In the late 1950s, one of the founders of AI, John McCarthy, proposed a computer
program to be called The Advice Taker, as part of a project that he called
programs with common sense. (McCarthy is famous for at least the following
things: He came up with the name artificial intelligence, he invented the
programming language Lisp, and he helped develop time sharing. For more
information on him, see:
http://en.wikipedia.org/wiki/John McCarthy (computer scientist)
and
http://aitopics.org/search/site/John%20McCarthy)
The idea behind The Advice Taker was that problems to be solved would
be expressed in a predicate-logic language (only a little bit more expressive
than first-order logic), a set of premises or assumptions describing required
background information would be given, and then the problem would be solved
by logically deducing it from the assumptions.
He gave an example: getting from his desk at home to the airport. It begins with
premises like
at(I,desk)
meaning I am at my desk, and rules like
xyz[at(x, y) at(y, z) at(x, z)],
which expresses the transitivity of the at predicate, and slightly more

complicated rules (which go slightly beyond the expressive power of FOPL)
like:
xyz[walkable(x) at(y, x) at(z, x) at(I, y) can(I, go(y, z, walking))];

A.7. NOTES FOR NEXT DRAFT 717
that is, if x is walkable, and if y and z are at x, and if I am at y, then I can go from
y to z by walking.
The proposition to be proved from these (plus lots of others) is:
want(at(I,airport))
To see it all worked out, take a look at [McCarthy, 1959].

Figure A.1: This flowchart uses Pi instead of Ri , as the text does. Also, the symbol
exists should be read: Does there exist.
Appendix B
Position-Paper Assignments
719
720 APPENDIX B. POSITION-PAPER ASSIGNMENTS
B.1 Recommended Readings:

Required:
Chudnoff, Elijah (2007), A Guide to Philosophical Writing, (Cambridge, MA:
Harvard Center for Expository Writing).
http://www.fas.harvard.edu/phildept/files/GuidetoPhilosophicalWriting.pdf
B.2. INTRODUCTION 721
B.2 Introduction
One of the best ways to learn how to do philosophy and, perhaps more importantly, to
find out what your beliefs are about important issues (as well as what your reasons for
your beliefs are!) is to write about them and then to discuss what youve written with
others who have also thought and written about about the issuesyour peers.
So, the writing assignments take the form of position papers, in which you will
be:
presented with a logical argument about a topic in the philosophy of computer
science,
asked to analyze and evaluate the argument,
given an opportunity to clarify and defend your analysis and evaluation,
and simultaneously be asked to help your peers clarify and defend their analyses
and evaluations of the same argument in an exercise called peer editing.
This should help you to clarify your own position on the topic.
The arguments that are presented in the following sections are those that I have
used when I have taught this course. Instructors are invited to modify these or to create
their own arguments. The assignments can be scheduled at the convenience of the
instructor, which is why I have collected them in this appendix rather than placing
them throughout the text. In general, however, I suggest putting each assignment
approximately one week after the relevant topic has been covered in class. This way,
the students will have the benefit of having thought about the readings before forming
their own opinions. An alternative, of course, is to schedule them one week before
the topic is begun to be covered in class, so that the students will be foreced to think
about the issues before reading what experts have had to say. Another option is to
do both: Assign the first draft before the topic is begun, then have the students do the
required readings and participate in class discussions of the topic, then follow this with
an optional revision or second draft of the position paper and the peer-editing session,
with a third draft (or second, if the optional, post-class-discussion draft is omitted) to
be turned in for instructor evaluation.
Peer-editing sessions should take up a full class period. The general method is to
divide the class into small groups of three or four students (two students will work
if the number of students in the classor latecomers!demands it, but is not ideal;
five-student groups are too large to enable everyone to participate, especially if time is
limited). For each student in a group:
1. Have the group read the students position paper.
2. Have the group challenge the students position, ask for clarification, and make
recommendations for improving the students paper.
If there are s students in a group and the peer-editing session lasts for m minutes,
then the group should spend no more than m/s minutes on each students paper. The
instructor should visit each group at least once to ensure that all is going well, to answer
questions, and to suggest questions if the group seems to be having trouble. If a group
ends early and there is a lot of time left in the session, ask each student in the group
to join another group (even if only to listen in to that groups ongoing discussion, but
if that other group also ended early, then the newcomer should peer-edit one of their
papers). Specific information for peer-editing sessions is given with each assignment.
After peer-editing, students should revise their position papers in the light of the
editing suggestions and hand in all drafts to the instructor. I usually give the students
one week for this revision.
B.3 Position Paper #1: What Is Computer Science?

B.3.1 Assignment
B.3.1.1 Introduction
The purpose of this position paper is to give you an opportunity to clarify your beliefs
about what computer science is, so that, as we continue to discuss the topic in class,
and as you continue to read about it, youll know where you standwhat your beliefs
are.
Later, when your beliefs have been contaminated by further readings and by our
discussions, you may wish to revise your beliefs. But you cant revise a belief that you
dont have (you can only acquire new beliefs). So, here I am forcing you to discover,
clarify, and defend the beliefs that you now have, by turning them into words and
putting them on paper.
B.3.1.2 The Argument

Imagine that you are the newly-appointed Dean of the School of Science at the
University of X. In an attempt to build up the rival School of Engineering, the newly-
appointed Dean of Engineering has proposed to the Provost (the boss of both deans)
that the Department of Computer Science be movedlock, stock, and computer, so to
speak1 to Engineering, on the following grounds:
1. Science is the systematic observation, description, experimental investigation,

and theoretical explanation of natural phenomena.
2. Computer science is the study of computers and related phenomena.
3. Therefore, computer science is not a science.
(The Dean of Engineering has not yet argued that computer science is an
engineering discipline; that may come later.)
How do you respond to the Dean of Engineerings argument?
You may agree with it, or not (but there are several ways that might happen; see
below).
1 http://www.worldwidewords.org/qa/qa-loc1.htm
B.3. POSITION PAPER #1: WHAT IS COMPUTER SCIENCE? 723
You should ignore political considerations: You may suppose that the move from
Science to Engineering involves no loss or gain of money, prestige, or anything else,
and it is to be done, if at all, only on strictly intellectual grounds.
The Provost is eagerly awaiting your reply, and will abide by your decision. . . if ,
that is, you give a well-argued defense of your position.
To formulate and defend your position, you should:
a) Say whether you agree that conclusion 3 logically follows from premises 1 and 2,
(whether or not you agree with 1 and 2)
and why you think it follows or doesnt.
(I agree that conclusion 3 follows from premises 1 and 2 because. . .
OR I do not agree that conclusion 3 follows from premises 1 and 2
because. . . )
If you think that conclusion 3 doesnt follow, is there some (interesting, non-
trivial) missing premise (that is, a missing link between the premises and
conclusion) that would make it follow?2
b) Say whether you agree with premise 1, and why you do or dont.
(I agree with premise 1 because . . . OR I disagree with premise 1

because . . . )
c) Say whether you agree with premise 2, and why you do or dont.
(I agree with premise 2 because . . . OR I disagree with premise 2

because . . . )
d) If you thought that there were missing premises that validated the argument, say
whether you agree with them, and why you do or dont.
e) If you think that the argument is logically invalid, you might still agree or disagree
with conclusion 3 independently of the reasons given for it by premises 1 and 2 (and
any missing premises).
If so, state whether you agree with 3, and why.
Its also possible that you might neither agree nor disagree with 3; alternatively,
you might both agree and disagree with it.
For example, you might believe that computer science is both a science and an
engineering discipline (or, alternatively, that it is neither).
If so, then please give your reasons for this.
2 For a discussion of missing premises, see
http://www.cse.buffalo.edu/rapaport/584/S10/EMAIL/20100118-MissingPremises.txt
Argument Analysis: Finding Missing Premises. INCLUDE THIS IN THE BOOK SOMEWHERE
And, if you are unsure about any of your answers, try to be very precise about why you
are unsure and what further information would help you decide.
Other responses:
You might not agree with any of these ways to respond. However, I believe that any
other response can, perhaps with a bit of force, be seen to fall under one of the above
responses. But if you really feel that your position is not exactly characterized by any
of the above responses, then please say:
what your position is,
why you believe it,
and why you think it is not one of the above.
For general assistance on analyzing arguments, see

http://www.cse.buffalo.edu/rapaport/584/S10/arganal.html
A Heuristic for Argument Analysis, and the links at
http://www.cse.buffalo.edu/rapaport/584/arganalhome.html
Argument Analysis ADD THESE TO THE TEXT
Ground Rules:
1. Your answer should honestly reflect your beliefs (not what you think the fictional
Provost or Dean of Engineering wants to hear!).
2. If you resort to a dictionary, textbook, article, website, etc., be sure to say which
one. Give as much detailed information as you can that would assist someone
else to locate the item by themselves. (See the
http://www.cse.buffalo.edu/rapaport/howtowrite.html#citations
How to Handle Citations section of my website on How to Write, for the
proper way to do this.)
3. Your position paper should be approximately 1 typed page and double-spaced

(that is, about 250 words) (not including any bibliographic citations).
To help keep your paper short, you do not need any fancy introductory
paragraph; you can assume that your reader is a fellow student in this course
who has just done the same assignment.
If you write:
1 paragraph analyzing validity,
1 paragraph each analyzing the premises,
and 1 paragraph analyzing the conclusion,
you will have (more than) enough material.
4. Please bring 5 copies to lecture on the due date.
5. At the top of the page, please put all and only the following information:
a) the title Position Paper #1

b) your name
c) the course you are enrolled in
d) the due date.
(The space taken up by this will not count against your total pages.)
6. For general assistance with writing (including my preferred method of paper
preparation and format, as well as advice on grammar), see my website
http://www.cse.buffalo.edu/rapaport/howtowrite.html
DUE AT THE BEGINNING OF LECTURE, 1 WEEK FROM TODAY
B.3.2 Suggestions and Guidelines for Peer-Group Editing

1. When you get into your small groups:
introduce yourselves quickly,

share copies of your papers with each other,
and write each others names on your paper (so that we have a record of
who peer-reviewed whom)
2. Choose one paper to discuss first.

Suggestion: Go in alphabetical order by family name.
3. The other people in the group might find it useful to imagine themselves as
members of a committee set up by the Provost to make a recommendation. Their
purpose is to try to help the author clarify his or her beliefs and arguments, so
that they will be able to make a recommendation to the Provost on purely logical
grounds (again: ignore politics!).
4. Start by asking the author to state (or read) his or her beliefs about whether
computer science is a science, giving his or her reasons for those beliefs.
5. Be sure that the author has discussed:
a) the validity of the argument

b) the truth value of (or their (dis)agreement with) premise 1
c) the truth value of (or their (dis)agreement with) premise 2
d) the truth value of (or their (dis)agreement with) any missing premises.
e) the truth value of (or their (dis)agreement with) the conclusion
And for each of the above, their reasons
6. Any time you have a question, ask it. Here are some suggestions:
Why did you say rather than ?

What did you mean when you said ?

Can you give me an example of ?
Can you give me more details about ?
Do you think that is always true?
Why? (This is always a good question to ask.)
How?
7. The author should not get defensive. The committee members are friendly.
Critical, but friendly.
8. Keep a written record of the questions and replies. This will be useful to the
author, for revision.
9. After spending about 10 minutes3 on the first paper, move on to the next, going
back to step 2 above, changing roles. Spend no more than 15 minutes4 per paper
(because youve only got about 45 minutes5 at most). Perhaps one member of
the group can be a timekeeper.
10. At home, over the next week, please revise your paper to take into consideration
the comments made by your fellow students (that is, your peers):
Perhaps defend your claims better, or clarify statements that were misunderstood,
etc. For help, see your instructor.
At the top of the first page of your revision, please put all and only the
following information:
a) the title Position Paper #1: 2nd Draft
b) your name
c) the course you are enrolled in
d) the due date
Please staple copies of your first draft, (with peer-editing comments, if any)
to your second draft.
Your second draft should be substantially different from your first draft!
12 PAGE (250500 WORD) REVISION, 1 COPY, TYPED, DUE ONE

WEEK FROM TODAY. NO LATE PAPERS WILL BE ACCEPTED!
3 Actually,some number n of minutes close to 10 but less than or equal to m/s, where m is the total
number of minutes in the class, and s is the total number of students in the group.
4 Actually, n + 5.
5 Actually, m.
B.3.3 Grading
B.3.3.1 Philosophy of Grading
To make grading these essays easier on the instructor and easy for students (in case the
instructor decides to have students grade each others essays), I recommend using the
triage method of grading. On this method, each item to be graded is given:
full credit (for example, 3 points) if it is clearly done in a completely acceptable

manner (even if it might not be entirely correct)
partial credit (for example, 2 points) if it is done, but is not clearly worth either
full credit or minimal credit
minimal credit (for example, 1 point) if it is done, but is clearly not done in an
acceptable manner.
no credit (that is, 0 points) if it is omitted.
The advantage to this method of grading is that the grader only has to decide if a
response is worth full credit (that is, shows clear understanding of what is expected) or
minimal credit (that is, shows clear mis- or lack of understanding of what is expected).
Any response that is not clearly one or the other is given partial credit. And failure to
respond, or omission of some requirement, is given no credit. This helps make grading
slightly more objective (and certainly easier for novice graders). Details and the theory
behind the method are given in [Rapaport, 2011a] and online at:
http://www.cse.buffalo.edu/rapaport/howigrade.html
B.3.4 Grading Rubric

1. Premise 1: Did you state clearly whether you agreed or disagreed with it?
(Doesnt matter whether you agreed or didnt agree, only with whether you said
so.)
3 pts = clearly stated whether you agreed or not

2 pts = not clearly stated but implied
1 pts = stated, but incorrect terminology
0 pts = did not clearly state whether you agreed
2. Did you give your reasons for your (dis)agreement?
3 = reasons given, clearly stated, & pertinent

2 = partial credit: I couldnt decide between 1 & 3
1 = reasons given, but not clearly stated or not pertinent
0 = no reasons
3. Premise 2: Did you state clearly whether you agreed or disagreed with it?
(Doesnt matter whether you agreed or didnt agree, only with whether you said
so.)
0, 1, 2, or 3, as for Premise 1
0,1,2, or 3, as for Premise 1
5. Valid?
no answer: 0
yes; XOR no, but no MP: 1
no, with wrong MP

or yes, with right MP: 2
no, with right MP: 3

(MP [Missing Premise] = computers are not natural phenomena)
6. Evaluation of MP (agree? why?):
no evaluation 0
dis/agree, but no reason 1
dis/agree, w/ unclear reason 2
dis/agree, w/ clear/pertinent reason 3
7. Conclusion: Did you state clearly whether you agreed or disagreed with it?
3 = clearly stated whether you agreed
2 = not clearly stated, but implied
0 = did not clearly state whether you agreed
0,1,2, or 3 points, as for Premise 1
9. Citation style:
used sources w/o citing 3
used sources w/ incomp & incorrect citation 2
used sources w/ incomp XOR incorrect citation 1
not applicable 0
10. Attached draft 1 & list of peer editors to demonstrate that draft 2 6=
draft 1? 0
Didnt 1
The total is 24 points, which, following my grading theory, maps into letter grades
as follows:
A 2324
A 22
B+ 21
B 1920
B 18
C+ 17
C 1416
C 1113
D+ 910
D 58
F 04
On my grading scheme,
A means understood the material for all practical purposes,
(here, thats 24 pts = 8 parts 3 pts full credit)
B has no direct interpretation, but comes about when averaging A and C grades
C means average,
(here, thats max 16 pts = 8 2 pts partial credit)
D means did not understand the material, (here, thats max 8 pts = 8 1 pt
minimum credit)
F usually means did not do the work (that is, 0 pts), but can also come about when
omitting some parts and doing D work on others.
B.4 Position Paper #2: What Is Computation?

B.4.1 Assignment
For this position paper, please evaluate the following argument:
1. [Knuth, 1973, 46] characterizes the informal, intuitive notion of algorithm as

follows:
a) Finiteness. An algorithm must always terminate after a finite number of

steps. . .
b) Definiteness. Each step of an algorithm must be precisely defined; the
actions to be carried out must be rigorously and unambiguously specified
for each case. . .
c) Input. An algorithm has zero or more inputs. . .
d) Output. An algorithm has one6 or more outputs. . .
6 My former student Albert Goldfain made the following observation on Knuths characterization of an
algorithm:
Knuths characterization of an algorithm includes:
1d Output. An algorithm has one or more outputs. . . (my emphasis)
Contrast this with (1c) zero or more inputs (my emphasis). This is not a typo, and most
likely not a mental slip by someone as precise as Knuth. There are several, modern, high-level
languages that will allow you to return nothing from a function. For example, consider the C
function:
int store-and-return(int x)
int i;
i=x;
return;
In some sense, there are no outputs from this function, yet it is still somehow an algorithm.
(In the Pascal programming language, there was a distinction between procedures and functions;
some other languages call procedures subroutines. But the point is that these are not functions
in the mathematical sense of ordered pairs.) In Lisp, which may have been more dominant when
Knuth wrote, every function had to return at least one value.
We can, however, re-interpret the output of an algorithm so that it is not always the return-
value of a function. So, in store-and-return, perhaps it is the memory location of the local
variable i that gets the output of the algorithm. In a program like Cs Hello World,
main()
printf("Hello World
n");
return;
perhaps it is the screen that you are printing to that gets the output of the algorithm.
Notice also that if you think of the store-and-return function as a black box (that is, a
device whose internal workings are hidden from you but whose external behavior is observable),
there would be no way to tell that it was doing anything. Behaviorally, you would hand an input
into the black box and observe nothing else! It may be that Knuth wants to exclude behaviorally
unobservable algorithms.
B.4. POSITION PAPER #2: WHAT IS COMPUTATION? 731
e) Effectiveness. [A]ll of the operations to be performed in the algorithm must

be sufficiently basic that they can in principle be done exactly and in a finite
length of time by a [hu]man using pencil and paper. . .
Note: We can also say that A is an algorithm for computing a function f means
that A is an algorithm as characterized above and that, for any input i, As output
for i = f s output for i; that is, for any i, A(i) = f (i).
2. Computer programming languages (like Java, Lisp, Fortran, etc.) are formal
languages for expressing (or implementing) algorithms.
3. Every computer programming language is equivalent in expressibility to a

Turing-machine programming language.
a) That is, every program in any programming language can be translated into
the language for programming Turing machines, and vice versa.
b) That is, any function that is computable by any programming language is
computable by a Turing machine, and vice versa.
4. Some real computer programs violate parts of definition 1:
a) For example, airline-reservation systems, ATMs, operating systems, etc.,

never terminate.
b) For example, heuristic AI programs dont always compute exactly the
function that they were written for, but only come very close. BE SURE
TO PUT IN A SECTION ON HEURISTIC PROGRAMS, AND ALSO SITE
KAHNEMANS BOOK
c) For example, the effectiveness of mundane or everyday procedures (like
recipes) depends on the environment in which they are executed.
For example, using a different brand of some ingredient can ruin

a recipe, or one chefs pinch of salt might be anothers 1/8th
teaspoon. And can you really execute a recipe using pencil and
paper?
d) For example, algorithms can apparently be written that can perform an infinite
computation in a finite amount of time (by continually accelerating).
For example, we can sum the terms of an infinite sequence in a finite

amount of time if we take 21n second to add the nth term.
And so on.
Bottom line: the one or more from premise 1d should not trip you up from analyzing the rest
of the argument.
5. Therefore, these programs (that is, the real programs referred to in premise 4)
do not implement Turing machines (contrary to premise 3).
6. Therefore, they (that is, the real programs referred to in premise 4) are not
computable. (But how can a real computer program not be computable?!)
To evaluate this argument, you must state whether the argument is valid, and you
must state whether and why you agree or disagree with each premise and conclusion.
If it is valid, and if you agree with each premise, then you believe that the
argument is sound.
You are logically obligated to believe the conclusions of sound arguments!
So, if you ever come across an argument that you think is sound, but whose
conclusion you dont believe, then either:
one (or more) of the premises is false,
or the argument is invalid (that is, there is some way for the premises to be
true yet for the conclusion to be false),
or both.
To determine whether it is valid, you must suppose for the sake of the argument
that all the premises are true, and then consider whether the conclusions logically
follow from them.
(Or: Can you imagine some way the world might be so that the premises are true
but the conclusion is false?)
Note that, in this argument, there are two conclusions: conclusions 5 and 6.
So, do you agree that conclusion 5 follows logically from premises 14

and/or that conclusion 6 follows logically from 5? If not, are there missing
premises that are needed to make the argument(s) valid? If there are, do
you agree with them (why/why not)?
Next, you must evaluate each premise. Do you agree with it? Why or why not?
Finally, do you agree with the conclusion(s)?
You might agree with a conclusion because you think that the argument is
sound;
if so, say so.
Or you might think that theres something wrong with the argument but
agree with the conclusion anyway;
if so, then try to present a better argument for the conclusion.
Or you might not agree with the conclusion(s);
if not, state why, and try to give an argument for what you do believe.
a) Your position paper should be approximately 12 typed pages, double-spaced

(that is, about 250500 words), and single-sided.
b) Please bring 5 copies to lecture on the due date.
c) At the top of the first page, please put the following information in the following
format:
Position Paper #2, 1st draft YOUR NAME

DATE DUE CLASS
d) For general assistance with writing (including my preferred method of paper

preparation and format, as well as advice on grammar), see my website, How
to Write:
As before, this doesnt have to be a beautifully written essay with an abstract. You
should just plunge in and evaluate the argument.
But you do need to give full citations to any sources that you cite.
DUE AT THE BEGINNING OF LECTURE, ONE WEEK FROM TODAY

1. When you get into your small groups, introduce yourselves quickly, and share
copies of your papers with each other.
2. Choose one paper to discuss first. (Suggestion: Go in alphabetical order by

family name.)
3. After spending about 1015 minutes on the first paper, move on to the next,
going back to step 2, above, changing roles.
Spend no more than 15 minutes per paper (because youve only got about 45
minutes at most).
Perhaps one member of the group can be a timekeeper.
4. For each paper, ask as many of the following questions as you have time for:
a) Did the author state whether and why they did or did not agree with Knuths
definition in premise 1?
Note: Knuths definition is a conjunction of 5 things: 1a & 1b & 1c &
1d & 1e.
So, in disagreeing with premise 1, an author must
a) explicitly disagree with (at least) one of 1a. . . 1e
b) and say why they disagree with that part (or those parts).
i) If the author agreed and gave reasons for agreeing, do you agree with
those reasons? Why?
ii) If the author disagreed and gave reasons for disagreeing, do you agree
with those reasons? Why?
b) Did the author state whether and why they did or did not agree with the
claim about the nature of programming languages in premise 2?
(Plus questions 4(a)i and 4(a)ii, above.)
c) Did the author state whether and why they did or did not agree with the claim
about the Turing-equivalence of programming languages in premise 3?
d) Did the author state whether and why they did or did not agree with the
claim and/or the examples in premise 4?
e) Did the author state whether and why they believe that conclusion 5 does or
does not validly follow from premises 14?
Do you agree with their evaluation?

f) If the author believes that conclusion 5 follows soundly from premises 14,
then they should state that they believe conclusion 5 for that reason.
Do they?
i) On the other hand, if the author believes that conclusion 5 does not
follow
(either because one or more of the premises is false or because the
argument is invalid),
then did the author state whether and why they did or did not agree with
the statement made in the conclusion?
ii) Note that if the author believes that the argument is unsound,
that is not a sufficient reason for disbelieving the claim!
(Thats because even a valid argument can have both false
premises and a true conclusion (or a false one),
and even an invalid argument can have a true conclusion (or a
false one).
The only thing that cant happen is to have a valid argument
both with true premises and with a false conclusion.)
g) If the author believes that conclusion 6 follows soundly from statement 5
considered as a premise along with some or all of the previous statements in
the argument (and possibly along with one or more missing premises!), then
they should state that they believe conclusion 6 for that reason. Do they?
On the other hand, if the author believes that conclusion 6 does not
follow
(either because one or more of its premises is false or because the
argument is invalid),
then did the author state whether and why they did or did not agree with
the statement made in the conclusion?

the comments made by your fellow students (that is, your peers):
Perhaps defend your claims better, or clarify statements that were misunderstood,
etc.
For help, see your instructor.
12 PAGE (250500 WORD) REVISION, 1 COPY, TYPED, DOUBLE-

SPACED, IS DUE ONE WEEK FROM TODAY.
NO LATE PAPERS WILL BE ACCEPTED!

1. Prem 1 (Knuths characterization of "algorithm")
agree? why? 0 = no answer

1 = answer, no reason
2 = answer, unclear reason
3 = answer, clear reason
------------------------------------------------------------------------
2. Prem 2 (Prog langs express/implement algorithms)
agree? why? 0,1,2,3, as above
------------------------------------------------------------------------
3. Prem 3 (Prog langs are equiv to TM prog lang)
------------------------------------------------------------------------
4. Prem 4 (Some real comp progs violate Knuths def)
------------------------------------------------------------------------
5. Conc 5 (So, such progs dont implement TMs)
1..4/..5 valid?
Why?
0 = no answer
1 = answer, no explanation
2 = answer, weak explanation
3 = answer, good explanation
agree w/ 5?
why? 0,1,2,3, as for 1
------------------------------------------------------------------------
6. Conc 6 (So, such progs are not computable)
5/..6 valid?
Why?
0 = no answer
1 = answer, no explanation
2 = answer, weak explanation
3 = answer, good explanation
agree w/6?
why? 0,1,2,3, as for 1
------------------------------------------------------------------------
7. Citation style:
used sources w/o citing -3

used sources w/ incomp & incorrect citation -2
used sources w/ incomp XOR incorrect citation -1
not applicable 0
------------------------------------------------------------------------
8. Attached draft 1 & list of peer editors

to demo that draft 2 <> draft 1? 0
Didnt -1
========================================================================
The total is 24 points, which, following my grading theory, maps into

letter grades as follows:
letter CSE484 both CSE/PHI584

A 23-24
A- 22
B+ 21
B 19-20
B- 18
C+ 17
C 14-16 9-16
C- 11-13
D+ 9-10
D 5-8
F 0-4
B.5 Position Paper #3: Is the Brain a Computer?

B.5.1 Assignment
For this position paper, I would like you to evaluate the following complex argument.
(Its complex because it consists of three subarguments, two of which treat the
conclusions of previous ones as premises.)
1. Turings Thesis: A physical object can compute if and only if it can do what a
(Universal) Turing Machine (TM) can do.
2. A computer is any physical device that can compute. (Consider this as a
(proposed) definition of computer.)
3. The human brain is a physical object that can do what a (Universal) TM can do.
4. Therefore, the human brain is a computer.
5. Microsoft Word is TM-computable. (That is, a Universal TM can execute
Microsoft Word.)
6. Therefore, any computer can execute Microsoft Word.
7. Therefore, the human brain can execute Microsoft Word.
As usual, to evaluate this argument, you must determine whether (I) it is valid and
whether (II) all the premises are true.
If both of those conditions hold, then an argument is said to be sound.
conclusion you dont believe
(by the way, do you really believe line 7 of this argument?),
then either one or more of the premises are false or it is invalid (that is,
there is some way for the premises to be true yet for the conclusion to be
false).
(I) To determine whether the argument is valid, you must suppose (or make believe)
for the sake of the argument that all the premises are true, and then consider
whether the conclusions logically follow from them. (Or: Can you imagine some
way the world might be so that the premises are true but the conclusion is false?)
Note that there are three conclusions: lines 4, 6, and 7. So, do you
agree that conclusion 4 follows logically from premises 13, and/or that
conclusion 6 follows logically from premise 5 (maybe with the help of
some of the earlier premises), and/or that conclusion 7 follows logically
from lines 4 and 6 considered as premises?
If not, are there missing premises that are needed to make the argument(s)
valid? If there are, do you agree with them (why/why not)?
B.5. POSITION PAPER #3: IS THE BRAIN A COMPUTER? 739
(II) It may be too difficult to determine whether each premise is true or false. More
realistically, you should decide whether you believe, or agree with, each premise,
and you must explain why you do or dont.
Finally, do you agree with the conclusion(s)? If you do, but you think that theres
something wrong with the argument, try to present a better one. If you dont agree with
the conclusion(s), state why, and try to give an argument for what you do believe.
a) Your position paper should be approximately

12 typed pages, double-spaced (that is, about 250500 words), and
single-sided.
b) Please bring 5 copies to lecture on the due date.
c) At the top of the first page, please put the following information in the following
format:
Position Paper #3, Draft 1 YOUR NAME

DATE DUE YOUR CLASS
For general assistance with writing (including my preferred method of paper

preparation and format, as well as advice on grammar), see my website How to
Write
As before, no abstract is needed for this position paper, but you do need to give full
citations to any sources that you cite.

d)
family name.)
going back to step 2, above, changing roles. Spend no more than 15 minutes per
paper (because youve only got about 45 minutes at most). Perhaps one member
of the group can be a timekeeper.
4. For each paper, ask as many of the following questions as you have time for:
a) Did the author state whether the argument from premises 13 to conclusion 4
was valid?
i) If they thought it was invalid, did they suggest a missing premise that
would make it valid (if thats possible)?
b) Did the author state whether the argument to conclusion 6 was valid?
i) Did they correctly identify its other premises besides premise 5? (Very
few real arguments can have only one premise.)
ii) If they thought it was invalid, did they suggest a missing premise that
would make it valid (if possible)?
c) Did the author state whether the argument to conclusion 7 was valid?
i) Did they correctly identify its other premises besides premise 6? (Note
that sentence 6 is both the conclusion of the previous argument and a
premise of this one.)
ii) If they thought it was invalid, did they suggest a missing premise that
would make it valid (if possible)?
d) For each premise, ask whether the author stated whether and why they did or
did not agree with it.
i) If the author agreed, then it is preferable (but not necessary) that they
give reasons for agreeing. If they did give such reasons, do you agree
with those reasons? Why?
ii) If the author disagreed, then it is necessary that they give reasons for
disagreeing, so do you agree with those reasons? Why?
e) For each argument, if the author thought it was unsound, did they state
whether they believed its conclusion anyway, on independent grounds (that
is, for different reasons)?
And, if so, do you agree with those reasons?
the comments made by your fellow students (that is, your peers): Perhaps
defend your claims better, or clarify statements that were misunderstood, etc.
12 PAGE (250500 WORD) REVISION, 1 COPY, TYPED, DOUBLE-

SPACED, IS DUE IN LECTURE ONE WEEK FROM TODAY. NO LATE
PAPERS WILL BE ACCEPTED!

Position Paper #3 Grading Rubric
Version: 14 Mar 10
========================================================================
I. VALIDITY OF ARGUMENTS:
1. Argument 1,2,3/.. 4: valid? + reason

B.5. POSITION PAPER #3: IS THE BRAIN A COMPUTER? 741
0 = no answer
1 = answer, no reason
OR reason that confuses definitions
of valid/invalid/sound
2 = answer, weak reason
3 = answer, good reason
------------------------------------------------------------------------
2. Argument 1,2,5/.. 6: valid? + reason

0,1,2,3 as above
------------------------------------------------------------------------
3. Argument 4,6/.. 7: valid? + reason 0,1,2,3 as above
------------------------------------------------------------------------
II. TRUTH VALUES OF STATEMENTS:
4. Prem 1: agree? + why? 0,1,2,3 as above
5. Prem 2: agree? + why? 0,1,2,3
6. Prem 3: agree? + why? 0,1,2,3
7. Conc 4: agree? + why? 0,1,2,3
8. Prem 5: agree? + why? 0,1,2,3
9. Conc 6: agree? + why? 0,1,2,3
10. Conc 7: agree? + why? 0,1,2,3
========================================================================

A 29-30
A-27-28
B+ 26
B 24-25
B-22-23
C+ 21
C 17-20 11-20
C-14-16
D+ 11-13
D 6-10
F 0-5
------------------------------------------------------------------------
As usual, on my grading scheme,
"A" means "understood the material for all practical purposes",

(here, thats 30 pts = 10 questions * 3 pts full credit)
"B" has no direct interpretation,

but comes about when averaging grades of A and C
"C" means "average",

(here, thats 20 pts = 10 * 2 pts partial credit)
"D" means "did not understand the material,

(here, thats 10 pts = 10 * 1 pt minimum credit)
"F" usually means "did not do the work" (that is, 0 pts),
but can also come about when averaging Ds and Fs.
Please see my grading website,
for the theory behind all of this, which Im happy to discuss via
UBLearns email.
B.6. POSITION PAPER #4: WHAT IS A COMPUTER PROGRAM? 743
B.6 Position Paper #4: What Is a Computer Program?

B.6.1 Assignment
B.6.1.1 The Argument
For this position paper, I would like you to evaluate the following argument:
1. A special-purpose computer (that is, a computer that does just one task) is
essentially a hardwired computer program.
2. Such a hardwired computer program is a physical machine.
3. Physical machines can be patented.
4. Therefore, such a hardwired computer program can be patented.
5. The printed text of a computer program is a literary work (that is, a piece of
writing) in the sense of the copyright law.
6. Literary works can be copyrighted.
7. Therefore, such a computer program can be copyrighted.
8. Nothing can be both patented and copyrighted.
Note: This premise is a matter of law. You must accept it as true. But you
can argue that the law should be changed.
9. There is no computational or other relevant difference between the hardwired

computer program and its textual counterpart (except for the different media in
which they are implemented, one being hardwired and the other being written
on, say, a piece of paper).
10. Therefore, computer programs can be both patented and copyrighted.
To help you evaluate this argument (which we look at in more detail in Ch. 13),
here are some extracts from some relevant websites:
a) From the official US Patent Office definition of patent:
a property right granted by the Government of the United States

of America to an inventor to exclude others from making, using,
offering for sale, or selling the invention throughout the United States
or importing the invention into the United States for a limited time
in exchange for public disclosure of the invention when the patent is
granted.
http://www.uspto.gov/main/glossary/index.html#patent
b) The Patent Office definition of invention:

any art or process (way of doing or making things), machine,

manufacture, design, or composition of matter, or any new and useful
improvement thereof, or any variety of plant, which is or may be
patentable under the patent laws of the United States.
http://www.uspto.gov/main/glossary/index.html#invention
c) The official US Copyright Office definition of copyright:
Copyright is a form of protection provided by the laws of the United

States. . . to the authors of original works of authorship, including
literary, dramatic, musical, artistic, and certain other intellectual works.
This protection is available to both published and unpublished works.
http://www.copyright.gov/circs/circ01.pdf
d) From the same website:
Copyrightable works include the following categories:

(a) literary works
(b) musical works, including any accompanying words
(c) dramatic works, including any accompanying music
(d) pantomimes and choreographic works
(e) pictorial, graphic, and sculptural works
(f) motion pictures and other audiovisual works
(g) sound recordings
(h) architectural works
These categories should be viewed broadly. For example, computer
programs and most compilations may be registered as literary
works; maps and architectural plans may be registered as pictorial,
graphic, and sculptural works.
WHAT IS NOT PROTECTED BY COPYRIGHT?
Several categories of material are generally not eligible for federal
copyright protection. These include among others:
works that have not been fixed in a tangible form of expression
(for example, choreographic works that have not been notated or
recorded, or improvisational speeches or performances that have not
been written or recorded)
titles, names, short phrases, and slogans; familiar symbols or
designs; mere variations of typographic ornamentation, lettering,
or coloring; mere listings of ingredients or contents
ideas, procedures, methods, systems, processes, concepts,
principles, discoveries, or devices, as distinguished from a
description, explanation, or illustration
works consisting entirely of information that is common property

and containing no original authorship (for example: standard
calendars, height and weight charts, tape measures and rulers,
and lists or tables taken from public documents or other common
sources)
http://www.copyright.gov/circs/circ01.pdf
To evaluate this argument, you must state whether the argument is valid and you
must state whether and why you agree or disagree with each premise. Remember:
Only single statements (like premises and conclusions) can be true or false.
For our purposes, its enough to say that a statement is true (or false) if you agree
(or disagree) with it, because Im not asking you to convince me that a statement
really is true (or false); Im only asking you to convince me that you have a good
reason for agreeing (or disagreeing) with it.
And only arguments can be valid or invalid. An argument is valid if its
impossible for all of its premises to be true while its conclusion is false (and
its invalid otherwise). For our purposes, to determine whether an argument is
valid, you must suppose (or make believe) for the sake of the argument that all
the premises are true (that is, that you agree with all of them), and then consider
whether you would have to logically agree with the conclusion. To determine
whether an argument is invalid, try to imagine some way the world might be so
that the premises are true but the conclusion is false.
Finally, only arguments can be sound or unsound. An argument is sound if
its valid and all of its premises are true (in which case, its conclusion will also
have to be true). For our purposes, well say that an argument is sound if its
valid and you really do agree with all of its premises (in which case, you really
have to agree with the conclusion).
conclusion you dont believe, then either one (or more) of the premises is false,
or it is invalid (that is, there is some way for the premises to be true yet for the
conclusion to be false), or both.
This means, of course, that you have to evaluate each premise and each (sub-
)argument, and, as usual, I also want you to evaluate the conclusion independently
of whether you think that it follows validly or doesnt follow validly from its premises.
a) For this position paper, I want to experiment with something a little bit different.
Instead of writing a first draft of your paper, I simply want you to fill in the attached
thinksheet, which will be an outline of your argument analysis.
You will write the paper after peer-editing the thinksheets.
b) Please bring 5 copies of your filled-out thinksheet to lecture on the due date.
B.6.1.2 Thinksheet
PhilCS 4/584, Spring 2010 NAME:

========================================================================
Thinksheet for Position Paper #4: What Is a Computer Program?
========================================================================
Statement |Agree?| Why?
(abbreviated versions of|(T?F?)| (use additional sheets if needed)
prems & conc of argt) | |
------------------------------------------------------------------------
1. A special-purpose | |
computer is | |
essentially a hard- | |
wired computer prog. | |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
2. Such a hardwired | |
comp.prog. is a | |
physical machine | |
| |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
3. Physical machines | |
can be patented | |
| |
| |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
4. Such a hardwired | |
comp.prog. can be | |
patented | |
| |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
1,2,3/..4 is valid | |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
5. The printed text of | |
a comp.prog. is a | |
lit.work in the | |
sense of the (c) law | |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
6. Lit.works can be (c) | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
7. The printed text of | |
a comprog can be (c) | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
5,6/..7 is valid | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
8. Nothing can be both | |
patented & (c) | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
9. Theres no compnl | |
or other diff. betw. | |
the hardwired comp. | |
prog. & its textual | |
counterpart... | |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
10. Comprogs can be both| |
patented & (c) | |
| |
| |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
4,7,9/..10 is valid | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
------------------------------------------------------------------------
Additional comments:

A) (a) When you get into your small groups, introduce yourselves quickly, and share
copies of your thinksheets with each other.
(b) Choose one thinksheet to discuss first. (Suggestion: Go in alphabetical order
by family name.)
(c) After spending about 1015 minutes on the first thinksheet, move on to the
next, going back to step Ab, above, changing roles. Spend no more than
15 minutes per paper (because youve only got about 45 minutes at most).
Perhaps one member of the group can be a timekeeper.
B) (a) Make sure each cell of the thinksheet is filled in.
i. The cells in the Agree? (T?F?) column should be filled in with
agree or disagree (or T or F).
ii. The cells in the Why? column should contain a reason why the
author agrees or disagrees with the statement, or why the author
thinks that the argument is valid or invalid.
(For a reminder about how to explain why you think that an
argument is, or isnt, valid,
584/S10/EMAIL/20100323-PP3-Comments.txt
see my earlier email (point #2).) FIX THIS.
These dont have to be complete sentences, but they should be
comprehensible.
(b) Keep a written record of the peer-editing suggestions. This will be useful to
the author, for revision.
C) The revision this time should, of course, be a correctly formatted paper, like the
ones you have been writing all semester.
(a) It should be fairly straightforward to turn the thinksheet outline into full
sentences and paragraphs (with correct citations if needed).
(b) I strongly urge you to have someone peer-edit your paper before you
produce the final version! Tell that person to make sure that you have:
i. Evaluated each statement (premise and conclusion) for (absolute
or independent) truth or falsity (see my
email (point #1) about that terminology,
FIX THIS.
and given a reason for your evaluation.
ii. Evaluated each argument for validity or invalidity
FIX THIS
(that is, evaluated each conclusion for relative truth, that is, truth
relative to the premises), and given a reason for your evaluation.
iii. Correctly used the true/false/valid/invalid terminology.
(c) Failure to correctly distinguish among true (or false) sentences,

propositions, statements, premises, or conclusions and valid (or
invalid) arguments will result in a lower grade! (After all, you need to
demonstrate that youve learned something this semester!)
D) (a) Your position paper should be approximately 12 typed pages, double-

spaced (that is, about 250500 words), and single-sided.
(b) At the top of the first page, please put the following information in the
following format:
Position Paper #4 YOUR NAME
DATE DUE YOUR CLASS
(c) Please attach the peer-edited thinksheets to your paper, as usual.
(d) For general assistance with writing (including my preferred method of paper
preparation and format, as well as advice on grammar), see my website How
to Write.
As before, no abstract is needed for this position paper, but you do need to
give full citations to any sources that you cite.
12 PAGE (250500 WORD) PAPER, 1 COPY, TYPED, SINGLE-SIDED,

DOUBLE-SPACED, IS DUE ONE WEEK FROM TODAY. NO LATE PAPERS
WILL BE ACCEPTED!

Version: 28 Mar 10
========================================================================
In general:
3 = Statement of position ("agree"/"disagree", or "valid"/"invalid")

with a clearly stated reason
2 = Statement of position, with an unclear reason
1 = Statement of position, with no reason given
0 = no response
I WILL DEDUCT 3 POINTS FROM THE TOTAL GRADE FOR THE PAPER
FOR INCORRECT USE OF THE TERMINOLOGY!
(If you are not sure of how to use the terminology, please ask me!)
AND MY OFFER TO GIVE YOU BACK ANY SUCH LOST POINTS

(OR LOST POINTS FOR INCORRECT USE OF CITATIONS)

STILL HOLDS ON ALL POSITION PAPERS.
a) Evaluation of premise 1: Agree? Why? 0,1,2,3

b) Evaluation of premise 2: Agree? Why? 0,1,2,3
c) Evaluation of premise 3: Agree? Why? 0,1,2,3
d) Evaluation of statement 4: Agree? Why? 0,1,2,3
f) Evaluation of 1,2,3/..4: Valid? why? 0,1,2,3
g) Evaluation of premise 5: Agree? Why? 0,1,2,3

h) Evaluation of premise 6: Agree? Why? 0,1,2,3
i) Evaluation of statement 7: Agree? Why? 0,1,2,3
j) Evaluation of 5,6/..7: Valid? Why? 0,1,2,3
k) Discussion of premise 8: Agree? Why? 0,1,2,3
Since conclusion 10 conflicts with the law (8),

you have two options:
Either accept the law, and reject 10...

[In that case, you must reject at least one of 1-7, & 9.
Which one, and why? (This is the "Alan Newell" solution.)]
...or else reject the law, and accept 10.

[Of course, you cant do that in real life unless maybe youre a
legislator (who can can write new laws)(see footnote (*), below)
or a Supreme Court justice (who can declare laws unconstitutional)
(see footnote (+), below.) In that case, you should propose a new
law. (This is the "David Koepsell" solution.)
l) Evaluation of premise 9: Agree? Why? 0,1,2,3

m) Evaluation of statement 10: Agree? Why? 0,1,2,3
n) Evaluation of 4,7,9/..10: Valid? Why? 0,1,2,3
========================================================================

A 37-39
A-35-36
B+ 33-34
B 31-32
B-29-30
C+ 27-28
C 21-26 14-26
C-18-21
D+ 14-17
D 7-13
F 0-6
------------------------------------------------------------------------
On my grading scheme,

(here, thats 39 pts = 13 questions * 3 pts full credit)

but comes about when averaging grades of A with Cs.

(here, thats 26 pts = 13 * 2 pts partial credit)

(here, thats 13 pts = 13 * 1 pt minimum credit)
"F" usually means "did not do the work" (that is, 0 pts),
for the theory behind all of this, which Im happy to discuss via email.
------------------------------------------------------------------------
Footnotes:
(*) Compare: Propose new axioms?
(+) Compare: Prove that a law is not a theorem of the US Constitution?
(Theres a story that the famous logician Kurt Goedel found an

inconsistency in the US Constitution when he was studying for his
American citizenship. He was going to tell the judge about it, but
Albert Einstein, who accompanied him to the ceremony, quickly changed
the subject :-) See: Goldstein, Rebecca (2006), Incompleteness: The
Proof and Paradox of Kurt Godel (Norton).)
B.7 Position Paper #5: Can Computers Think?

B.7.1 Assignment
For this position paper, I would like you to evaluate the following hypothetical debate.
Pro: If something behaves in all relevant ways as if it were cognitive, then it is

cognitive.
Con: What do you mean by being cognitive?
Pro: I mean that it:
can perceive (see, hear, etc.);
has beliefs, desires, and intentions;
can remember;
can use and understand natural language;
can reason and make rational decisions; etc.
You know, the sort of thing that AI researchers are trying to achieve by
computational means.
Con: Do you think they will succeed?
Pro: Im optimistic: I think that a computer running a suitable AI program (or maybe
a suite of programs) will eventually behave in all these ways.
Con: But that means that you think that such an AI-programmed computer will be
cognitive?
Pro: Yes.
Con: But thats crazy! Computers and computer programs are purely syntactic!
Pro: Now its my turn to ask for clarification: What do you mean by syntactic?
Con: I mean that all a computer can do is to manipulate the symbols of a formal
symbol system.7
Pro: So whats the problem?
Con: The problem is that cognition is semantic! That is, it involves the semantic
interpretation of those symbols.
Pro: Well, Im not so sure about that. But suppose youre right. What then?
Con: Well, syntax does not suffice for semantics. So, no computer executing a purely
syntactic computer program can exhibit semantic cognition, even if it behaves in
all relevant ways as if it were cognitive.
7 http://www.cse.buffalo.edu/rapaport/formalsystemsFIX THIS
B.7. POSITION PAPER #5: CAN COMPUTERS THINK? 755
It may help if you rewrite Pros and Cons arguments in terms of premises and
conclusions, and then evaluate those arguments. That is, extract each argument
from the debate and put them in the following forms:
1. Pros premise 1 1. Cons premise 1

2. Pros premise 2 2. Cons premise 2
3. (etc.) (etc.)
4. Therefore, Pros conclusion 4. Therefore, Cons conclusion
Then analyze each argument.
Keep in mind that premises and conclusions are declarative propositions (they
can be deemed to be true or false) but that some lines uttered by Pro and Con
are not declarative propositions (and thus cant be premises or conclusions). For
example, Cons first statement is a questionit is not a premise or conclusion
of anyones argumentand Pros second statment needs to be reformulated as
something like Something is cognitive means that it. . . .
1. For your peer-editing session next week, I will give you a choice: You may
either:
(a) create a thinksheet like the one for Position Paper #4 (B.6.1.2)
with one column listing the premises, conclusions, and arguments;
one column of cells to indicate your agreement or disagreement with
them;
and one column of cells to indicate your reasons for your agreement
or disagreement
(b) or write a 12 page, double-spaced (that is, about 250500 word), single-
sided, first draft.
(Of course, you might want to do option 1a for your own use before doing
option 1b! They are not mutually inconsistent.
If your document is more than 1 page long, please staple the pages together and
make sure that your name is on all pages!
2. Please bring 5 copies to class on the due date.
3. At the top of the first page, please put ALL AND ONLY the following
information in the following format:
Position Paper #5 YOUR NAME

DATE DUE YOUR CLASS
4. Failure to correctly distinguish among true (or false) sentences, propositions,

statements, premises, or conclusions and valid (or invalid) arguments
will also result in a lower grade!
5. For general assistance with writing (including my preferred method of paper

preparation and format, as well as advice on grammar), see my website How to
Write.
And dont forget to give full citations to any sources that you cite.


family name.)
going back to step 2 above, changing roles. Spend no more than 15 minutes per
paper (because youve only got about 45 minutes at most). Perhaps one member
of the group can be a timekeeper.
4. Suggestion: There are really 2 arguments in this dialogue: Pros argument and
Cons argument.
So, the first task is to present each argument. Once you have identified the
premises (including any hidden premises) and conclusion of each argument, you
can then analyze it for validity of the argument and truth of the premises.
5. For each paper in your peer-editing group, ask as many of the following questions
as you have time for:
(a) Did the author present both Pros and Cons arguments?
(b) For each argument, did the author state whether and why s/he believes the
argument to be valid?
Its possible to formulate both arguments so that they are valid!
If you do that, then ascertaining the truth value of the premises
becomes your central task.
(c) For each argument, did the author state whether and why s/he agrees with
the premises?
(d) For each argument, if the author believed either that the argument was
invalid (even with missing premises addedthat is, that there was no way
to make the argument valid) or that one or more of the premises was false,
then did the author state whether and why s/he agrees with the conclusion?
Reminder:
i. If you think an argument is sound, then you are logically obligated

to believe its conclusion (and you dont have to give any other
justification for the conclusion).
ii. If you dont believe the conclusion of an argument, then it is either
invalid or else has at least one false premise; you must identify
which, and explain why.
iii. If you think an argument is unsound (either because it is invalid
or has at least one false premise), then you might still believe the
conclusion for other reasons; in that case, you must give those
other reasons.
6. Remember!: Your revised paper must have the appropriate heading at the
top of the first page, must use the terms true, false, valid, and invalid
appropriately, and must have your peer-edited first drafts attached!
the comments made by your fellow students (that is, your peers): Perhaps
defend your claims better, or clarify statements that were misunderstood, etc.
12 PAGE (250500 WORD) REVISION, 1 COPY, TYPED, SINGLE-SIDED,

DOUBLE-SPACED, IS DUE ONE WEEK FROM TODAY. NO LATE PAPERS
WILL BE ACCEPTED!

Heres a draft of the grading rubric for Position Paper #5.
Because you need to spell out Pros and Cons arguments in

premise-conclusion form, and because this may use up space,
it will not count against the word- and page-limits.
But if your paper is > 1 page, please STAPLE the pages together
(one staple, in upper left corner)
AND
please put your NAME on ALL pages.
========================================================================
Version: 20 Apr 2010
========================================================================
a) Incorrect use of "true", "false", "valid", "invalid", "sound",

"unsound", "argument", "premise", "conclusion", etc.:
-1 pt PER ERROR!
========================================================================
b) PROS ARGUMENT
b1) List of premises & conclusion for Pros argument:
3 = * clearly stated argument,

* premises & conclusion clearly identified,
* prems & conc clearly derived from dialogue
2 = neither clearly 3 nor 1,

including: not correctly identifying
some prem or conc.
1 = arg not clearly presented

or not clearly derived from dialogue
0 = missing
------------------------------------------------------------------------
b2) Evaluation of validity of Pros argument:
3 = valid XOR invalid

+ clear explanation why
(including addition of any missing premises)
2 = valid XOR invalid, unclear explanation
1 = valid XOR invalid, no explanation
0 = no evaluation of validity
------------------------------------------------------------------------
b3) Evaluation of truth-value of Pros premises:
Note: Because each of you might have slightly different premises,

I cant assign points to each one in any equally fair way,
so I will grade you on your overall evaluation of the truth-
values of the premises that you have explicitly identified.
6 = for EACH premise:

truth-value clearly stated
& good reasons given
4 = for SOME (but not all) premises:

truth-value not stated
OR no or weak reason given
2 = for MOST (UPDATED: or all) premises:

truth-value not stated
OR no or weak reasons given
0 = no evaluation of truth-values of premises
[Note: Because of my "quantum" scheme of grading,

it is not possible to get 1,3, or 5 points!]
------------------------------------------------------------------------
b4) Evaluation of truth-value of Pros conclusion:
3 = if arg is sound,
then that is your reason for believing conc,
-- say so!
else (if arg is not sound, then)
say whether you believe conc
& give clear reason why
2 = neither clearly 1 nor 3
1 = you think arg is not sound (which is fine),

but you give no clear statement of truth-value of conc
& no or weak reason given
0 = no evaluation of conclusion
========================================================================
c) CONS ARGUMENT (to be graded similarly, namely:)
c1) List of premises & conclusion for Cons argument:
0,1,2,3 pts as above
c2) Evaluation of validity of Cons argument:
c3) Evaluation of truth-value of Cons premises:

c4) Evaluation of truth-value of Cons conclusion:
========================================================================

A 29-30
A-27-28
B+ 26
B 24-25
B-22-23
C+ 21
C 17-20 11-20
C-14-16
D+ 11-13
D 6-10
F 0-5
------------------------------------------------------------------------
On my "quantum-triage" grading scheme,

(here, thats 30 pts = 6 questions * 3 pts full credit
+ 2 questions * 6 pts full credit)

but comes about when averaging grades of A and C

(here, thats 20 pts = 6 * 2 pts partial credit
+ 2 * 4 pts partial credit)

(here, thats 10 pts = 6 * 1 pt minimum credit
+ 2 * 2 pts minimum credit)
"F" usually means "did not do the work" (i.e., 0 pts),

for the theory behind all of this, which Im happy to discuss.

B.8 Optional Position Paper: A Competition

Some of you have told us that you would like to come up with your own arguments
instead of merely analyzing ones that we give you.
Heres your opportunity!
No later than two weeks from today, try your hand at creating an argument relevant
to one of the topics of this course. It could be on a topic that weve already discussed,
on a topic that were going to discuss, or on some other topic in the philosophy of
computer science (for ideas, take a look at the Further Readings at the ends of each
chapter).
Your argument should have at least two premises and at least one conclusion. Try
to make it a valid argument!
The winner (if there is onewe reserve the right to decide not to choose one) will
have the honor of her or his argument being used as the argument to be analyzed for
the next Position Paper. (To make it interesting and fair, for his or her position-paper
assignment, the winner may be asked to refute the argument!)
You may submit the argument on paper (in lecture or by email). We also reserve the
right to slightly modify the argument, if we feel that would make it more interesting.
Appendix C
Term Paper Topics
Version of 23 February 2013

To the instructor: Discuss this at approximately 3 weeks into a 15-week term.
Here is a list of some possible term-paper topics:
1. Further discussion of any topic covered in class, e.g.:
(a) A critical examination of (someone elses) published answer to one of the

questions listed on the syllabus.
(b) Your answer to one of the questions listed on the syllabus, including a
defense of your answer.
2. A critical examination of any of the required or recommended (or any other

approved and relevant) readings.
3. A critical study of any monograph (i.e., single-topic book) or anthology

(including special issues of journals) on the philosophy of computer science.
4. A critical, but general, survey article on the philosophy of computer science that
would be appropriate for an encyclopedia of philosophy or an encyclopedia of
computer science.
5. A presentation and well-argued defense of your philosophy of computer

science, that is, your answers to all (or most) of our questions, together with
supporting reasons.
6. Other ideas of your own, approved by me in advance, including, but not limited
to, such topics that we have mentioned but have not investigated as:
Is CS an empirical or a mathematical science?

Is CS an artificial or a natural science?
How does (the philosophy of CS) fit in with Alfred North Whiteheads
process philosophy?
763
764 APPENDIX C. TERM PAPER TOPICS
For general assistance with writing (including my required method of paper

preparation and format, as well as advice on grammar), see my website How to Write
http://www.cse.buffalo.edu/
For specific assistance on writing a philosophy paper, see:
Mendelovici, Angela (2011), A Sample Philosophy Paper, online at:
http://tinyurl.com/SamplePhilPaper (accessed 13 February 2013)
Wolff, Robert Paul (1975), A Simple Foolproof Method for Writing Philosophy
Papers, in his About Philosophy (Prentice-Hall); online at:
http://www.amyscott.com/Philosophy%20Paper.pdf (accessed 13 February
2013)
Or do an Internet search for how to write a philosophy paper.

The paper should be a maximum of 10 double-spaced, single-sided pages (that is,
about 2500 words) (not counting the bibliography).
Deadlines:
1. Two weeks from today: Proposal and reading list due.

Your proposal and reading list must be approved by me before you begin
your research and writing.
Because the term paper is optional, you do not need to commit yourself to
it even if you turn in a proposal.
2. Final paper will be due on the last day of the course.

Appendix D
Final Exam
D.1 To the Instructor

At least one of the goals of philosophy education ought to be the fostering of the
students development of analytical and critical thinking skills. In order to explain
the nature of the sample final exam to be presented in the next section, I want to
call attention to a theory of cognitive development of college students, to discuss its
implications for critical thinking, and to show how it can apply to the development of
writing exercises such as this final exam.1
William G. Perrys scheme of cognitive development [Perry, 1970], [Perry, 1981]
is a descriptive theory of stages that represent students changing attitudes towards
knowledge and values as they progress through their education. There are nine stages,
which fall into four groups.2
I. Dualism Position 1. Basic Duality: Students at this stage believe that there are
correct answers to all questions, that the instructor is the authority figure
who has access to golden tablets in the sky containing all correct answers,
and that their (the students) job is to learn the correct answers so that they
can be repeated back to the instructor when asked. A Basic Duality student
who offers a wrong answer to a question hears the authority figure say, You
are wrong.
Position 2. Dualism: Students move to this stage when faced with alternative
opinions or disagreements among different authority figures (different
instructors), such as when one literature teacher says that Huckleberry Finn
is the best American novel, but another says that it is the worst. Dualistic
1 This section is adapted from [Rapaport, 1984].
2 My descriptions
are culled from [Perry, 1981], [Cornfeld and Knefelkamp, 1979], and [Goldberger, 1979]. These are three
essential readings for anyone concerned with implications and applications of Perry theory in the classroom.
765
766 APPENDIX D. FINAL EXAM
students infer that one of those literature teachers view of the golden
tablets is obscured. Consequently, they see the purpose of education as
learning to find the correct answers.
Dualistic students prefer structured classes, which they see as providing the
correct answers, and subjects such as math, which they see as having clear
answers (all math teachers agree that the golden tablets say that 2 + 2 =
4). Conflict between instructor and text, or between two instructors, is seen
threateningly as conflicts among authority figures.
II. Multiplicity Position 3. Early Multiplicity: Here, the student has moved from the
narrow Dualism of seeing all questions as having correct or else incorrect
answers to a wider dualism of classifying questions into two kinds: those
where instructors know the correct answers and those where they dont
know the correct answers yet. Concerning the latter, the instructors role
is seen by Early Multiplistic students as providing methods for finding the
correct answers, rather than as giving the correct answers directly.
Position 4. Late Multiplicity: As students move along, chronologically or
cognitively, they begin to see the second kind of question as being the more
common one. Because, it is felt, most questions are such that instructors
dont have the correct answers for them, everyone has a right to his
own opinion; no one is wrong! [Perry, 1981, 79]. The instructors task,
therefore, is seen as either teaching how to think or, worse, being irrelevant
(after all, everyone has a right to their own opinion, including instructors
but its just their opinion).
III. Contextual Relativism Position 5. Here, students have begun to see that
instructors arent always asking for correct answers but, rather, for
supported answers. The second category of questions has become the
only category (except, of course, in mathematics and science!), but,
although there can be many answers for each question, some are better
(more adequate, more appropriate, etc.) than others. Answers are now seen
as being better or worse relative to their supporting context (hence the name
of this position).
IV. Commitment within Relativism Positions 69. These stages
characterize students as they begin to see the need for making their own
decisions (making commitments, affirming values), as they balance their
differing commitments, and as they realize the never-ending nature of this
process.
Finally, there is evidence that a student as Position x will not understandwil literally
not be able to make any sense out ofinstruction aimed at Position x + 2 (or beyond).
Conversely, students at higher levels are bored by instruction aimed at lower levels.3
3 Perrys theory is far, far richer than I have portrayed it here, and the interested reader is urged to follow
up the suggested readings. A useful survey of criticisms, together with a discussion of the relevance of the
theory to mathematics education and to the history of matheamtics, is [Copes, 1980]. The relevance of the
theory to philosophy is discussed in [Rapaport, 1982].
D.1. TO THE INSTRUCTOR 767
Here is a useful anecdote (adapted from Perry) for illustrating the scheme: Suppose
that a teacher presents three theories for explaining the purpose of science (for example,
is science merely descriptive, merely predictive, or explanatory?), or three different
algorithms for solving some computational problem. The Dualistic student will wonder
which is the correct one (and why the teacher bothered to talk about the incorrect
ones); The Multiplistic student will think, Only 3? Heck, I can come up with 6!; the
Contextual Relativist will wonder what advantages and disadvantages each theory has;
and the Commitment-oriented student will be wondering about how to decide which is
most appropriate or useful in a given situation.
Data from several studies indicate that most entering college freshmen are at
Positions 2 or 3 [Perry, 1970], [Perry, 1981], [Copes, 1980]. Courses designed to teach
critical thinking skills to students through the first two years of college are thus dealing
with Dualists or (early) Multiplists, and this can result in several problems that the
instructor should be aware of in order to foster the students development along the
Perry scheme.
First, Dualists want to be told the correct answers. But critical-thinking courses
are largely involved with criticism and argument analysis. Accordingly, the entire
activity may appear to them as incomprehensible at worst and pointless at best, or may
simply result in the students learning the sport of dumping on bad or incorrect
arguments. Hence, such courses, including the present one, must be more than mere
criticism courses; they must make a serious attempt to teach ways of constructing
arguments, solving problems, or making decisions. In this way, they can offer an
appropriate challenge to Dualistic students, especially if couched in a context of
adequate support.4
Second, The highly logical argument that, since everybody has a right to their
own opinion, there is no basis for rational choice is very typical of Multiplistic
students [Goldberger, 1979, 7]. But a goal of critical-thinking courses should be
precisely to provide bases for rational choice: logical validity, inductive strength, etc.
Accordingly, Multiplistic students either will not comprehend this goal or will view it
as pointless. Again, such a course can be appropriately challenging to the students, but
the instructor must be aware of how the students are likely to perceive itto hear
students negative comments not as marks of pseudo-sophistication or worse, but as
marks of viewing the world Multiplistically.
Finally, consider the concept of logical validity. Larry Copes (personal
conversation) points out that it is a relativistic concept: A valid conclusion is
one that is true relative to the truth of the premises. Dualistic students searching for
absolutes and Multiplistic students feeling that anything goes may not accept, like,
or understand validity. This may explain why so many students believ that arguments
with true conclusions are valid or that valid arguments require true premiseseven
after having dutifully memorized the definition of validity!
How can an instructor simultaneously challenge students in order to help them
move to a higher-numbered stage, yet not threaten them, especially when a given class
might have students at widely varying stages? One suggestion, based on work by Lee
4 For details and specific advice, see [Sanford, 1967, Ch. 6, esp. pp. 5152],
[Cornfeld and Knefelkamp, 1979].
Knefelkamp, is to create assignments that can appeal to students at several levels.

The suggested final exam in the next section is one such assignment. Students are
offered five questions and asked to answer any three of them. Question 1 is similar
to, but not exactly the same as, the argument in Position Paper #1, so it is a little
bit challenging to the Dualistic student, but is supportive in that they have already
practiced giving a response to it. Question 2 is a brand-new argument for analysis,
but one that students could have predicted that it would have appeared on a final
exam, because it covers a topic that was explicitly discussed in the course yet was
not the subject of a position paper; consequently, it is challenging, because it is new,
but supportive, because it covers material that should be familiarit should appeal to
both Dualistic and Multiplistic students. Questions 3 and 4 appear to be open-ended
questions that should appeal to Multiplistic students, though they can be understood as
questions that might appeal to Dualistic students, too; after all, they are topics that were
covered in the course, and the students have been given tools for evaluating answers
to such questionsthe challenge here is to construct arguments for answers to the
questions. Here, the students choice of which question (3 or 4) to answer (if any) is
a function of their personal interests or familiarity with the issues. Question 5 is the
most challenging, because the student must come up with a question and then answer
it; it should appeal to Multiplistic as well as Contextual Relativistic students.
If left to their own choices, students will choose the least challenging question
commensurate with their current stage. Thus, students need not be threatened by
a question that they perceive as being to difficult or even unintelligible. But each
question is just a bit more challenging than the previous one, and, because the students
must answer three questions, they are offered a choice that includes at least one fully
supportive question and at least one more-challenging question. (If such an exam is
offered as a mid-term exam, then the final exam could begin with a least-challenging
question that is more challenging than the least-challenging one on the mid-term.) In
this way, students are encouraged to strive for a bit more, thus, hopefully, beginning
the move to the next stage of cognitive development.
It is imperative for those of us who teach such courses to learn how to challenge
our students appropriately in order to foster their intellectual growth; we must hear
how our students inevitably make their own meaning out of what we say to them; and
we must be ready to support them in the ego-threatening process of development.
D.2 A Final Exam

Do any 3 of the following. Write about 250500 words for each answer.
This is a closed-book, closed-notes, closed-neighbor, open-mind exam.
No books, notebooks, food, beverages, or electronic devices of any kind are
permitted in the exam room.
1. Evaluate the following argument (note that it is similar to, but not exactly the
same as, the argument in Position Paper #1):
Natural science is the
systematic observation, description, experimental investigation, and
D.2. A FINAL EXAM 769
theoretical explanation of natural phenomena. Computer science is

the study of computers and computing. Therefore, computer science
is not a natural science.
2. Evaluate the following argument:
Suppose that computers running certain computer programs can

make rational decisions (at least in the sense of outputting values of
functions that serve as a basis for decision making). That is, suppose
that they can determine the validity of arguments and ascertain the
probable truth-values of the premises of the arguments, and that they
can consider the relative advantages and disadvantages of different
courses of action, in order to determine the best possible choices.
(For example, there are computers and computer programs that can
diagnose certain diseases and (presumably) recommend appropriate
medical treatments; there are computers and computer programs that
can prove and verify proofs of mathematical theorems; and there
are computers and computer programs that can play winning chess.)
Suppose for the sake of argument that some of these computers and
computer programs can make decisions (and recommendations) on
certain important matters concerning human welfare. Suppose further
that they can regularly make better recommendations than human
experts on these matters. Therefore, these computers should make
decisions on these important matters concerning human welfare.
3. What is computer science?
4. Can computers think?
5. Choose either 5a or 5b:
(a) In your opinion, what is the most fundamental or important question in the
philosophy of computer science?
(b) What is a question that interests you in the philosophy of computer science
that we did not discuss this semester?
Pose the question, explain why you think it is important or interesting, and
present your answer to it.
D.3 NOTES FOR NEXT DRAFT

On the Perry Scheme:
See http://www.askphilosophers.org/question/5563
Bibliography
[Aaronson, 2008] Aaronson, S. (2008). The limits of quantum computers. Scientific

American, pages 6269. Accessed 30 June 2014 from:
http://www.cs.virginia.edu/robins/The Limits of Quantum Computers.pdf.
[Aaronson, 2011] Aaronson, S. (2011). Quantum computing promises new insights,
not just supermachines. New York Times, page D5. Accessed 16 July 2014 from:
http://www.nytimes.com/2011/12/06/science/scott-aaronson-quantum-computing-
promises-new-insights.html.
[Aaronson, 2013] Aaronson, S. (2013). Why philosophers should care about
computational complexity. In Copeland, B. J., Posy, C. J., and Shagrir, O.,
editors, Computability: Turing, Godel, Church, and Beyond, pages 261327. MIT
Press, Cambridge, MA. http://www.scottaaronson.com/papers/philos.pdf (accessed
16 December 2013).
[Aaronson, 2014] Aaronson, S. (2014). Quantum randomness. American Scientist,
102(4):266271. Accessed 8 September 2014 from:
http://www.americanscientist.org/issues/pub/quantum-randomness.
[Abrahams, 1987] Abrahams, P. (1987). What is computer science? Communications
of the ACM, 30(6):472472.
[Abramson, 2014] Abramson, D. (2014). The philosophical legacy of alan turing.
http://www.thecritique.com/articles/the-philosophical-legacy-of-alan-turing/.
[Acocella, 2009] Acocella, J. (2009). Betrayal: Should we hate judas iscariot? The
New Yorker, pages 6873.
[Adami, 2012] Adami, C. (2012). Boldly going beyond mathematics. Science,
338:14211422. review of [Holland, 2012].
[Adleman, 1998] Adleman, L. M. (1998). Computing with dna. Scientific American,
pages 5461. http://www.usc.edu/dept/molecular-science/papers/fp-sciam98.pdf
[Aizawa, 2010] Aizawa, K. (2010). Computation in cognitive science: It is not all
about Turing-equivalent computation. Studies in History and Philosophy of Science,
41(3):227236. Reply to [Piccinini, 2004a].
771
772 BIBLIOGRAPHY
[Allen, 2001] Allen, L. G. (2001). Teaching mathematical induction: An alternative

approach. Mathematics Teacher, 94(6):500504.
[Allen, 1989] Allen, S. (1989). Meeting of Minds. Prometheus Books.
[Anderson and Belnap, 1975] Anderson, A. R. and Belnap, Jr., N. D., editors (1975).
Entailment: The Logic of Relevance and Necessity, volume I. Princeton University
Press, Princeton, NJ.
[Anderson et al., 1975] Anderson, A. R., Belnap Jr., N. D., and Dunn, J. M., editors
(1975). Entailment: The Logic of Relevance and Necessity, volume II. Princeton
University Press, Princeton, NJ.
[Anderson and Anderson, 2006] Anderson, M. and Anderson, S. L., editors (2006).
Special Issue on Machine Ethics. IEEE Intelligent Systems. Vol. 21, No. 4
(July/August).
[Angier, 2010] Angier, N. (2010). Abtract thoughts? the body takes them literally.
New York Times/Science Times, page D2. Accessed 31 January 2014 from:
http://www.nytimes.com/2010/02/02/science/02angier.html.
[Anonymous, 1853] Anonymous (1853). Review of j.h. wythes, the microscopist; or a
complete manual on the use of the microscope. Quarterly Journal of Microscopical
Science, 1(1):5153.
[Anthes, 2006] Anthes, G. (1 May 2006). Computer science looks for a remake. In
Computerworld, accessed 2 July 2013 from
http://www.computerworld.com/s/article/110959/Computer Science Looks for a Remake.
[Antoy and Hanus, 2010] Antoy, S. and Hanus, M. (2010). Functional logic
programming. Communications of the ACM, 53(4):7485.
[Apostel, 1961] Apostel, L. (1961). Towards the formal study of models in the non-
formal sciences. In Freudenthal, H., editor, The Concept and the Role of the Model
in Mathematics and Natural and Social Sciences: Proceedings of the Colloquium
Sponsored by the Division of Philosophy of Sciences of the International Union of
History and Philosophy of Sciences Organized at Utrecht, January 1960, pages 1
37. D. Reidel, Dordrecht, Holland.
[Appiah, 2007] Appiah, K. A. (2007). The new new philosophy. New York Times
Magazine, pages 3436.
[Appiah, 2008] Appiah, K. A. (2008). Experimental philosophy. Proceedings and
Addresses of the American Philosophical Association, 82(2):722.
https://member.apaonline.org/V82 2 experimentalphilosophy.aspx.
[Ardis et al., 1989] Ardis, M., Basili, V., Gerhart, S., Good, D., Gries, D., Kemmerer,
R., Leveson, N., Musser, D., Neumann, P., and von Henke, F. (1989). Editorial
process verification. Communications of the ACM, 32(3):287290. ACM Forum
letter to the editor, with replies by James H. Fetzer and Peter J. Denning.
BIBLIOGRAPHY 773
[Aspray, 1990] Aspray, W., editor (1990). Computing before Computers. Iowa State
University Press, Ames, IA. online at [http://ed-thelen.org/comp-hist/CBC.html].
[Austin, 1983] Austin, A. K. (1983). An elementary approach to np-completeness.

American Mathematical Monthly, 90:398399.
[Ayer, 1956] Ayer, A. (1956). The Problem of Knowledge. Penguin, Baltimore.
[Baars, 1997] Baars, B. J. (1997). Contrastive phenomenology: A thoroughly

empirical approach to consciousness. In Block, N., Flanagan, O., and Guzeldere,
G., editors, The Nature of Consciousness: Philosophical Debates, pages 187201.
MIT Press, Cambridge, MA.
[Bader, 2013] Bader, R. M. (2013). Towards a hyperintensional theory of intrinsicality.

Journal of Philosophy, 110(10):525563. Preprint accessed 13 February 2014 from
http://users.ox.ac.uk/sfop0426/Intrinsicality%20(R.%20Bader).pdf.
[Bajcsy et al., 1992] Bajcsy, R. K., Borodin, A. B., Liskov, B. H., and Ullman,
J. D. (1992). Computer science statewide review draft preface. Technical report,
Computer Science Rating Committee. Confidential Report and Recommendations
to the Commissioner of Education of the State of New York (unpublished),
http://www.cse.buffalo.edu/rapaport/Papers/Papers.by.Others/bajcsyetal92.pdf,
accessed 30 September 2011.
[Ballard, 1997] Ballard, D. H. (1997). An Introduction to Natural Computation. MIT

Press, Cambridge, MA.
[Barr, 1985] Barr, A. (1985). Systems that know that they dont understand.
http://www.stanford.edu/group/scip/avsgt/cognitiva85.pdf (accessed 21 July 2013).
[Barwise, 1989] Barwise, J. (1989). For whom the bell rings and cursor blinks. Notices
of the American Mathematical Society, 36(4):386388.
[Baum, 1900] Baum, L. F. (1900). The Wizard of Oz. PUB, CITY.
[Beebe, 2011] Beebe, J. R. (2011). Experimental epistemology research group.

http://eerg.buffalo.edu/. Accessed 9 September 2011.
[Benenson, 2013] Benenson, Y. (2013). Recombinatorial logic. Science,

340(6132):554555. Accessed 31 October 2014 from:
http://www.rle.mit.edu/sbg/wp-content/uploads/2013/05/Science-2013-Benenson-
554-5.pdf.
[Berners-Lee et al., 2006] Berners-Lee, T., Hall, W., Hendler, J., Shadbolt, N., and
Weitzner, D. J. (2006). Creating a science of the web. Science, 313:769771.
10.1126/science.1126902.
[Blessing, 2013] Blessing, K. (2013). I re-read, therefore i understand. Philosophy

Now, 94:17. Accessed 20 February 2013 from
http://philosophynow.org/issues/94/I Re-Read Therefore I Understand.
774 BIBLIOGRAPHY
[Boden, 1977] Boden, M. A. (1977). Artificial Intelligence and Natural Man. Basic
Books, New York.
[Boden, 2006] Boden, M. A. (2006). Mind as Machine: A History of Cognitive
Science. Oxford University Press, Oxford.
[Boey, 2014] Boey, G. (2014). On my interview with peter unger, and the value of
philosophy. 3 Quarks Daily. Accessed 2 July from:
http://www.3quarksdaily.com/3quarksdaily/2014/06/on-my-interview-with-peter-
unger-and-the-value-of-philosophy.html.
[Bohm and Jacopini, 1966] Bohm, C. and Jacopini, G. (1966). Flow diagrams, turing
machines and languages with only two formation rules. Communications of the
ACM, 9(5):366371. Accessed 19 November 2013 from:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.119.9119&represent=rep1&type=pdf.
[Boole, 2009] Boole, G. (2009). An Investigation of the Laws of Thought: On Which

Are Founded the Mathematical Theories of Logic and Probabilities. Cambridge
University Press, Cambridge, UK.
[Boolos and Jeffrey, 1974] Boolos, G. S. and Jeffrey, R. C. (1974). Computability and
Logic. Cambridge University Press, Cambridge, UK.
[Boorstin, 1983] Boorstin, D. (1983). The Discoverers. Random House, New York.
Ch. 49: The Microscope of Nature.
[Borbely, 2005] Borbely, R. (2005). Letter to the editor. Scientific American, pages
12, 15.
[Borges, 1981] Borges, J. L. (1981). Partia enchantments of the quixote. In Monegal,
E. R. and Reid, A., editors, Borges, A Reader; A Selection from the Writings of Jorge
Luis Borges, pages 232235. E.P. Dutton, New York.
[Bostrom, 2003] Bostrom, N. (2003). Are you living in a computer simulation?
Philosophical Quarterly, 53(211):243255. Available online, along with many
related items, at http://www.simulation-argument.com/.
[Bowie, 1973] Bowie, G. L. (1973). An argument against churchs thesis. Journal of
Philosophy, 70(3):6676.
[Brenner, 2012] Brenner, S. (2012). The revolution in the life sciences. Science,
338:14271428.
[Bringsjord, 2006] Bringsjord, S. (11 August 2006). unpublished ms.
. http://www.cse.buffalo.edu/rapaport/Papers/Papers.by.Others/bringsjord06-
constraints on philofcompsci.pdf.
[Bringsjord and Arkoudas, 2004] Bringsjord, S. and Arkoudas, K. (2004). The
modal argumenbt for hypercomputing minds. Theoretical Computer Science,
317:167190. doi:10.1016/j.tcs.2003.12.010; accessed 5 February 2014 from
http://kryten.mm.rpi.edu/modal.hypercomputing.pdf.
BIBLIOGRAPHY 775
[Brooks, 1996] Brooks, Jr., F. P. (1996). The computer scientist as toolsmith ii.
Communications of the ACM, 39(3):6168. Revised and extended verson of
Brooks Jr., Frederick P. (1977), The Computer Scientist as ToolsmithStudies
in Interactive Computer Graphics, in B. Gilchrist (ed.), Information Processing 77,
Proceedings of IFIP Congress 77 (Toronto) (Amsterdam: North-Holland): 625
634; accessed 6 June 2013 from
http://www.cs.unc.edu/techreports/88-041.pdf.
[Brooks, 1975] Brooks, Frederick P., J. (1975). The Mythical Man-Month. Addison-
Wesley, Reading, MA.
[Brooks, 1991] Brooks, R. A. (1991). Intelligence without representation. Artificial

Intelligence, 47:139159.
[Brown, 1992] Brown,

R. (4 August 1992). contribution to newsgroup discussion of cyc. Article 13305
of comp.ai. http://www.cse.buffalo.edu/rapaport/definitions.of.ai.html.
[Buchanan, 2006] Buchanan, B. G. (2006). What do we know about knowledge? AI

Magazine, pages 3546.
[Bundy, 1983] Bundy, A. (1983). The nature of ai: A reply to schank. AI Magazine,
4(4):2931.
[Bunge, 1974] Bunge, M. (1974). Toward a philosophy of technology. In Michalos,

A. C., editor, Philosophical Problems of Science and Technology, pages 2847.
Allyn & Bacon, Boston.
[Burks, 1970] Burks, A. W., editor (1970). Essays on Cellular Automata. University
of Illinois Press, Urbana, IL.
[Buxton, 2012] Buxton, W. A. (2012). Innovative interaction: From concept to the

wild. Communications of the ACM, 55(9):90.
[Campos, 2011] Campos, D. G. (2011). On the distinction between peirces abduction

and liptons inference to the best explanation. Synthese, 180:419442.
[Cane, 2014] Cane, S. (2014). Interview: David chalmers and andy clark. New
Philosopher, 2:mind. Accessed 13 March 2014 from:
http://www.newphilosopher.com/articles/interview-david-chalmers-and-andy-
clark/.
[Cannon, 2013] Cannon, P. (2013). Kant at the bar: Transcendental idealism in daily
life. Philosophy Now, Issue 95:1517. Accessed 15 November 2013 from
http://philosophynow.org/issues/95/Kant at the Bar Transcendental Idealism in Daily Life.
[Cardoso, 1997] Cardoso, S. H. (1997). Specialized functions of the cerebral cortex.

http://www.cerebromente.org.br/n01/arquitet/cortex i.htm.
776 BIBLIOGRAPHY
[Carhart, 1956] Carhart, R. R. (1956). The systems approach to reliability. In

Proceedings, 2nd National Symposium on Quality Control and Reliabillity in
Electronics, January 910, pages 149155. IRE Professional Group on Reliability
and Quality Control, American Society for Quality Control, Electronics Technical
Committee, Institute of Radio Engineers, Washington, DC.
[Carleton, 1984] Carleton, L. R. (1984). Programs, language understanding, and

searle. Synthese, 59:219230.
[Carlson et al., 1996]

Carlson, B., Burgess, A., and Miller, C. (1996). Timeline of computing history.
http://www.computer.org/cms/Computer.org/Publications/timeline.pdf.
[Carnap, 1956] Carnap, R. (1956). Meaning and Necessity: A Study in Semantics and
Modal Logic, Second Edition. University of Chicago Press, Chicago.
[Carpenter and Doran, 1977] Carpenter, B. and Doran, R. (1977). The other turing
machine. The Computer Journal, 20(3):269279. Accessed 6 November 2013 from
http://comjnl.oxfordjournals.org/content/20/3/269.full.pdf+html.
[Carroll, 1871] Carroll, L. (1871). Through the looking-glass. online at:

http://www.gutenberg.org/files/12/12-h/12-h.htm. (accessed 20 July 2013).
[Carroll, 1893] Carroll, L. (1893). Sylvie and Bruno Concluded. Macmillan, London.
[Carroll, 1895] Carroll, L. (1895). What the tortoise said to achilles. Mind, 4(14):278
280. online at http://www.ditext.com/carroll/tortoise.html, accessed 7 January 2013.
[Carroll, 1848] Carroll, L. (ca. 1848). The two clocks. In The Rectory Umbrella.
(unknown). Accessed 4 February 2014 from: http://etc.usf.edu/lit2go/112/poems-
puzzles-and-stories-of-lewis-carroll/4953/the-two-clocks/.
[Cathcart and Klein, 2007] Cathcart, T. and Klein, D. (2007). Plato and a Platypus
Walk into a Bar: Understanding Philosophy through Jokes. Abrams Image, New
York.
[Cerf, 2012] Cerf, V. G. (2012). Computer science revisited. Communications of the

ACM, 55(12):7.
[Cerf, 2015] Cerf, V. G. (2015). There is nothing new under the sun. Communications
of the ACM, 58(2):7.
[Ceruzzi, 1989] Ceruzzi, P. (1989). Electronics technology and computer science,

19401975: A coevolution. Annals of the History of Computing, 10(4):257275.
[Chaitin, 1987] Chaitin, G. J. (1987). Computing the busy beaver function. In Cover,
T. and Gopinath, B., editors, Open Problems in Communication and Computation,
pages 108112. Springer. http://www.cs.auckland.ac.nz/chaitin/bellcom.pdf
BIBLIOGRAPHY 777
[Chalmers, 1995] Chalmers, D. J. (1995). On implementing a computation. Minds

and Machines, 4:391402.
[Chalmers, 1996a] Chalmers, D. J. (1996a). The Conscious Mind: In
Search of a Fundamental Theory. Oxford University Press, New York.
http://tinyurl.com/plv877.
[Chalmers, 1996b] Chalmers, D. J. (1996b). Does a rock implement every finite-state
automaton? Synthese, 108:309333. 1993 preprint accessed 23 May 2012 from
http://consc.net/papers/rock.html; Chalmers corrects an error in my arguments in
this paper in [Chalmers, 2012, 236238].
[Chalmers, 2011] Chalmers, D. J. (2011). A computational foundation for the study of
cognition. Journal of Cognitive Science (South Korea), 12(4):323357. Originally
written in 1993; slightly revised version published 2011; accessed 3 November 2014
from:
http://j-cs.org/gnuboard/bbs/board.php?bo table= vol012i4&wr id=2.
[Chalmers, 2012] Chalmers, D. J. (2012). The varieties of computation: A
reply. Journal of Cognitive Science (South Korea), 13(3):211248. http://j-
cs.org/gnuboard/bbs/board.php?bo table= vol013i3&wr id=1
[Changizi et al., 2008] Changizi, M. A., Hsieh, A., Nijhawan, R., Kanai, R., and
Shimojo, S.
(2008). Perceiving the present and a systematization of illusions. Cognitive Science,
32:459503. http://onlinelibrary.wiley.com/doi/10.1080/03640210802035191/epdf
[Chase, 1980] Chase, G. C. (1980). History of mechanical computing machinery.
Annals of the History of Computing, 2(3):198226.
[Chater and Oaksford, 2013] Chater, N. and Oaksford, M. (2013). Programs as causal
models: Speculations on mental programs and mental representation. Cognitive
Science, 37(6):11711191.
[Chetty, 2013] Chetty, R. (2013). Yes, economics is a science. New York Times,
page A21. online at http://www.nytimes.com/2013/10/21/opinion/yes-economics-
is-a-science.html (accessed 4 November 2013).
[Chisum, 1986] Chisum, D. S. (1985-1986). The patentability of algorithms.
University of Pittsburgh Law Review, 47:9591022.
[Chopra, 2014] Chopra, S. (2014). Computer programs are people,
too. http://www.thenation.com/article/180047/computer-programs-are-people-too.
Accessed 13 June 2014.
[Church, 1936a] Church, A. (1936a). A note on the entscheidungsproblem. Journal
of Symbolic Logic, 1(1):4041. See also Correction to A Note on the
Entscheidungsproblem, Journal of Symbolic Logic 1((3) (September): 101102.
778 BIBLIOGRAPHY
[Church, 1936b] Church, A. (1936b). An unsolvable problem of elementary number

theory. American Journal of Mathematics, 58(2):345363. Accessed 4 February
2014 from: http://phil415.pbworks.com/f/Church.pdf.
[Church, 1937] Church, A. (1937). Review of [Turing, 1936]. Journal of Symbolic
Logic, 2(1):4243. For commentary on this review, see [Hodges, 2013].
[Church, 1956] Church, A. (1956). Introduction to Mathematical Logic. Princeton

[Churchland and Sejnowski, 1992] Churchland, P. S. and Sejnowski, T. J. (1992). The
Computational Brain. MIT Press, Cambridge, MA.
[Clarke, 1953] Clarke, A. C. (1953). Childhoods End. PUB, CITY.

[Cleland, 1993] Cleland, C. E. (1993). Is the Church-Turing thesis true? Minds and
Machines, 3(3):283312.
[Cleland, 1995] Cleland, C. E. (1995). Effective procedures and computable functions.
Minds and Machines, 5(1):923.
[Cleland, 2001] Cleland, C. E. (2001). Recipes, algorithms, and programs. Minds and
Machines, 11(2):219237.
[Cleland, 2002] Cleland, C. E. (2002). On effective procedures. Minds and Machines,
12(2):159179.
[Cleland, 2004] Cleland, C. E. (2004). The concept of computability. Theoretical

Computer Science, 317:209225.
[Cockshott and Michaelson, 2007] Cockshott, P. and Michaelson, G. (2007). Are there
new models of computation? reply to wegner and eberbach. The Computer Journal,
50(2):232247. Preprint accessed 30 June 2014 from:
http://www.dcs.gla.ac.uk/wpc/reports/wegner25aug.pdf.
[Coffa, 1991] Coffa, J. (1991). The Semantic Tradition from Kant to Carnap: To the
Vienna Station. Cambridge University Press, Cambridge, UK.
[Cohen), 1919] Cohen), P. M. R. (1919). A slackers apology. The New Republic,
pages 1920.
[Colburn and Shute, 2011] Colburn, T. and Shute, G. (2011). Decoupling as a
fundamental value of computer science. Minds and Machines, 21(2):241259.
http://www.d.umn.edu/tcolburn/papers/Decoupling.pdf.
[Colburn, 1999] Colburn, T. R. (1999). Software, abstraction, and ontology. The

Monist, 82(1):319.
[Colburn, 2000] Colburn, T. R. (2000). Philosophy and Computer Science.
M.E. Sharpe, Armonk, NY.
BIBLIOGRAPHY 779
[Colburn et al., 1993] Colburn, T. R., Fetzer, J. H., and Rankin, T. L., editors (1993).
Program Verification: Fundamental Issues in Computer Science. Kluwer Academic
Publishers, Dordrecht, The Netherlands.
[Collins and Quillian, 1972] Collins, A. M. and Quillian, M. R. (1972). How to make a
language user. In Tulving, E. and Donaldson, W., editors, Organization of Memory,
pages 309351. Academic Press, New York.
[Computer Science, 2003] Computer Science, G. W. U. D. o. (2003). Computer
science careers. http://www.seas.gwu.edu/simhaweb/misc/cscareers.html.
[Cooper, 2012] Cooper, S. B. (2012). Turings titanic machine? Communications
of the ACM, 55(3):7483. http://cacm.acm.org/magazines/2012/3/146259-turings-
titanic-machine/fulltext (accessed 26 February 2014).
[Copeland, 1996] Copeland, B. J. (1996). What is computation? Synthese, 108:335
359. Preprint (accessed 18 March 2014) at
http://www.alanturing.net/turing archive/pages/pub/what/what.pdf.
[Copeland, 1997] Copeland, B. J. (1997). The broad conception of computation.
American Behavioral Scientist, 40(6):690716.
[Copeland, 1998] Copeland, B. J. (1998). Even turing machines can compute
uncomputable functions. In Calude, C., Casti, J., and Dinneen, M. J., editors,
Unconventional Models of Computtion, pages 150164. Springer-Verlag. Preprint
http://www.alanturing.net/turing archive/pages/pub/even/even.pdf.
[Copeland, 2000] Copeland, B. J. (2000). The church-turing thesis.
http://www.alanturing.net/turing archive/pages/Reference%20Articles/The%20Turing-
Church%20Thesis.html. Accessed 24 September 2014.
[Copeland, 2002] Copeland, B. J. (2002). Hypercomputation. Minds and Machines,
12(4):461502.
[Copeland and Proudfoot, 2010] Copeland, B. J. and Proudfoot, D. (2010). Deviant
encodings and turings analysis of computability. Studies in History and Philosophy
of Science, 41(3):247252.
[Copeland and Shagrir, 2011] Copeland, B. J. and Shagrir, O. (2011). Do accelerating
Turing machines compute the uncomputable? Minds and Machines, 21(2):221239.
[Copeland and Sylvan, 1999] Copeland, B. J. and Sylvan, R. (1999). Beyond the
universal turing machine. Australasian Journal of Philosophy, 77(1):4667.
Preprint accessed 27 May 2014 from:
http://www.alanturing.net/turing archive/pages/pub/beyond/beyond.pdf.
[Copeland and Sylvan, 2000] Copeland, B. J. and Sylvan, R. (2000). Computability
is logic-relative. In Hyde, D. and Priest, G., editors, Sociative Logics and Their
Applications: Essays by the Late Richard Sylvan, pages 189199. Ashgate.
780 BIBLIOGRAPHY
[Copeland, 1999] Copeland, B. J. e. (1999). The turing-wilkinson lecture series on

the automatic computing engine. In Furukawa, K., Michi, D., and Muggleton,
S., editors, Machine Intelligence 15: Intelligent Agents, pages 381444. Oxford
University Press, Oxford.
[Copes, 1980] Copes, L. (1980). College teaching, mathematical knowledge, and the
perry development scheme. (St. Paul, MN: Institute for Studies in Educational
Mathematics).
[Cornfeld and Knefelkamp, 1979] Cornfeld,
J. and Knefelkamp, L. (1979). Combining student stage and type in the design of
learning environments: An integration of perry stages and holland typologies. Paper
presented at the American College Personnel Association, Los Angeles, March.
[Covert, 2014] Covert, M. W. (2014). Simulating a living cell. Scientific American,
310(1):4451.
[Crossley and Henry, 1990] Crossley, J. N. and Henry, A. S. (1990). Thus spake al-
khwarizmi: A translation of the text of cambridge university library ms. ii.vi.5.

Historia Mathematica, 17:103131.
[Crowcroft, 2005] Crowcroft, J. (2005). On the nature of computing. Communications
of the ACM, 48(2):1920.
[Darwin, 1872] Darwin, C. (1872). The Origin of Species. Signet Classics, 1958, New
York.
[DATA-LINK, 1958] DATA-LINK, E. o. (1958). Whats in a name? Communications
of the ACM, 1(4):6.
[David, 2009] David, M. (2009). The correspondence theory of truth. In
Zalta, E. N., editor, Stanford Encyclopedia of Philosophy (Fall 2009 Edition).
http://plato.stanford.edu/archives/fall2009/entries/truth-correspondence/.
[Davidson, 2006] Davidson, J. (2006). Measure for measure: Exploring the mysteries
of conducting. The New Yorker, pages 6069.
[Davies, 1999] Davies, D. W. (1999). Repairs to turings universal computing
machine. In Furukawa, K., Michie, D., and Muggleton, S., editors, Machine
Intelligence 15: Intelligent Agents, pages 477488. Oxford University Press,
Oxford. http://tinyurl.com/lqero7l (accessed 21 January 2014).
[Davis, 1995a] Davis, M. (1995a). An historical preface to engineering ethics. Science
and Engineering Ethics, 1(1):3348.
[Davis, 1995b] Davis, M. (1995b). Mathematical logic and the origin of modern
computers. In Herken, R., editor, The Universal Turing Machine: A Half-Century
Survey, Second Edition, pages 135158. Springer-Verlag, Vienna.
[Davis, 1996] Davis, M. (1996). Defining engineer: how to do it and why it matters.
Journal of Engineering Education, 85(2):97101.
BIBLIOGRAPHY 781
[Davis, 1998] Davis, M. (1998). Thinking Like an Engineer: Studies in the Ethics of a
Profession. Oxford University Press, New York.
[Davis, 2000] Davis, M. (2000). Overheard in the park. American Scientist, 88:366
367. Accessed 19 March 2014 from
http://www.americanscientist.org/bookshelf/pub/overheard-in-the-park.
[Davis, 2004] Davis, M. (2004). The myth of hypercomputation. In Teuscher, C.,

editor, Alan Turing: The Life and Legacy of a Great Thinker, pages 195202.
Springer, Berlin.
[Davis, 2006a] Davis, M. (2006a). The church-turing thesis: Consensus and

opposition. In Beckmann, A., Berger, U., Lowe, B., and Tucker, J., editors, Logical
Approaches to Computational Barriers: Second Conference on Computability in
Europe, CiE 2006, Swansea, UK, June 30July 5, pages 125132. Springer-Verlag
Lecture Notes in Computer Science 3988, Berlin.
[Davis, 2006b] Davis, M. (2006b). What is Turing reducibility? Notices of the AMS,
53(10):12181219. http://www.ams.org/notices/200610/whatis-davis.pdf (accessed
30 May 2014).
[Davis, 2006c] Davis, M. (2006c). Why there is no such discipline as

hypercomputation. Applied Mathematics and Computation, 178:47.
[Davis, 2011] Davis, M. (2011). Will software engineering ever be engineering?

Communications of the ACM, 54(11):3234.
[Davis, 2012] Davis, M. (2012). The Universal Computer: The Road from Leibniz
to Turing; Turing Centenary Edition. CRC Press/Taylor & Francis Group, Boca
Raton, FL. also published as Engines of Logic: Mathematicians and the Origin of
the Computer (New York: W.W. Norton, 2000).
[Davis and Weyuker, 1983] Davis, M. D. and Weyuker, E. J. (1983). Computability,

Complexity and Langauges. Academic Press, New York.
[Davis et al., 1996] Davis, R., Samuelson, P., Kapor, M., and Reichman, J. (1996).
A new view of intellectual property and software. Communications of the ACM,
people.csail.mit.edu/davis/cacm96.ps. A summary version of
[Samuelson et al., 1994].
[Davis, 1977] Davis, R. M. (1977). Evolution of computers and computing. Science,

195:10961102.
[Daylight, 2013] Daylight, E. G. (2013). Towards a historical notion of Turing

the father of computer science.
http://www.dijkstrascry.com/sites/default/files/papers/Daylightpaper91.pdf.
782 BIBLIOGRAPHY
[De Millo et al., 1979] De Millo, R. A., Lipton, R. J., and Perlis, A. J. (1979). Social
processes and proofs of theorems and programs. Communications of the ACM,
22(5):271280.
[Dedekind, 1890] Dedekind, R. (1890). Letter to keferstein. In van Heijenoort, J.,
editor, From Frege to Godel: A Source Book in Mathematical Logic, 18791931,
pages 98103. Harvard University Press, Cambridge, MA.
[DeJohn and Dietrich, 2003] DeJohn, J. and Dietrich, E. (2003). Editorial: Subvert
the dominant paradigm! a review of computationalism: new directions, edited by
Matthias Scheutz. Journal of Experimental and Theoretical Artificial Intelligence,
15(4):375382.
[Dennett, 2009] Dennett, D. (2009). Darwins strange inversion of reasoning.

Proceedings of the National Academy of Science, 106, suppl. 1:1006110065.
http://www.pnas.org/cgi/doi/10.1073/pnas.0904433106. See also [Dennett, 2013].
[Dennett, 2013] Dennett, D. (2013). Turings strange inversion of reasoning. In
Cooper, S. B. and van Leeuwen, J., editors, Alan Turing: His Work and Impact,
pages 569573. Elsevier, Amsterdam. See also [Dennett, 2009].
[Dennett, 1971] Dennett, D. C. (1971). Intentional systems. Journal of Philosophy,
68:87106. Reprinted in Daniel C. Dennett, Brainstorms (Montgomery, VT:
Bradford Books): 322.
[Dennett, 1982] Dennett, D. C. (1982). Notes on prosthetic imagination. Boston

Review, 7(3):37.
[Dennett, 2012] Dennett, D. C. (2012). Sakes and dints: And other definitions
that philosophers really need not seek. Times Literary Supplement, pages 1214.
http://ase.tufts.edu/cogstud/papers/TLS2012.pdf (accessed 21 November 2013).
[Denning, 1985] Denning, P. J. (1985). What is computer science? American Scientist,

73:1619.
[Denning, 1999] Denning, P. J. (1999). Computer science: The discipline. In Ralston,
A., Reilly, E. D., and Hemmendinger, D., editors, Encyclopedia of Computer
Science, Fourth Edition. Groves Dictionaries, 2000, New York. preprint (dated
1999) available online at:
http://cs.gmu.edu/cne/pjd/PUBS/ENC/cs99.pdf (accessed 11 June 2013); page
references are to the preprint.
[Denning, 2003] Denning, P. J. (2003). Great principles of computing.
[Denning, 2005] Denning, P. J. (2005). Is computer science science? Communications

of the ACM, 48(4):2731.
[Denning, 2013] Denning, P. J. (2013). The science in computer science.
BIBLIOGRAPHY 783
[Denning et al., 1989] Denning, P. J., Comer, D. E., Gries, D., Mulder, M. C.,
Tucker, A., Turner, A. J., and Young, P. R. (1989). Computing as a discipline.
[Denning and Freeman, 2009] Denning, P. J. and Freeman, P. A. (2009). Computings

paradigm. Communications of the ACM, 52(12):2830.
[Denning and Martell, 2015] Denning, P. J. and Martell, C. H. (2015). Great

Principles of Computing. MIT Press, Cambridge, MA.
[Dershowitz and Gurevich, 2008] Dershowitz, N. and Gurevich, Y. (2008). A natural

axiomatization of computability and proof of churchs thesis. Bulletin of Symbolic
Logic, 14(3):299350. Accessed 30 June 2014 from:
http://research.microsoft.com/pubs/70459/tr-2007-85.pdf.
[Deutsch, 1985] Deutsch, D. (1985). Quantum theory, the church-turing principle and
the universal computer. Proceedings of the Royal Society of London A, 400:97117.
Preprint accessed 8 April 2014 from:
http://www.cs.berkeley.edu/christos/classics/Deutsch quantum theory.pdf;page
references are to the preprint.
[Deutsch, 2009] Deutsch, M. (2009). Experimental philosophy and the theory of

reference. Mind & Language, 24(4):445466.
[Dewdney, 1989] Dewdney, A. (1989). The Turing Omnibus: 61 Excursions

in Computer Science. Computer Science Press, Rockville, MD. esp.
Chs. 1 (Algorithms: Cooking Up Programs), 28 (Turing Machines: The Simplest
Computers), and 48 (Universal Turing Machines: Computers as Programs).
[Dewey, 1993] Dewey, J. (1993). How We Think: A Restatement of Reflective Thinking

to the Educative Process, revised ed. Heath, Boston.
[Dick, 1968] Dick, P. K. (1968). Do Androids Dream of Electric Sheep? Doubleday,

New York. Reprinted by Random House/Del Rey/Ballantine, 1996.
[Dietrich, 2001] Dietrich, E. (2001). Homo sapiens 2.0: Why we should build the
better robots of our nature. Journal of Experimental and Theoretical Artificial
Intelligence, 13(4):323328.
[Dietrich, 2007] Dietrich, E. (2007). After the humans are gone. Journal of
Experimental and Theoretical Artificial Intelligence, 19(1):5567.
[Dijkstra, 1968] Dijkstra, E. W. (1968). Go to statement considered harmful.

Communications of the ACM, 11(3):147148. Preprint accessed 27 July 2014 from:
https://www.cs.utexas.edu/EWD/transcriptions/EWD03xx/EWD361.html.
[Dijkstra, 1974] Dijkstra, E. W. (1974). Programming as a discipline of mathematical

nature. American Mathematical Monthly, ?(?):608612.
784 BIBLIOGRAPHY
[Dijkstra, 1975] Dijkstra, E. W. (1975). EWD 512: Comments at a symposium. In

Selected Writings on Computing: A Personal Perspective, pages 161164. Springer-
Verlag, New York. Accessed 19 April 2013 from
http://www.cs.utexas.edu/EWD/ewd05xx/EWD512.PDF.
[Dijkstra, 1976] Dijkstra, E. W. (1976). EWD 611: On the fact that the
atlantic ocean has two sides. In Selected Writings on Computing: A
Personal Perspective, pages 268276. Springer-Verlag, New York. online at
http://www.cs.utexas.edu/EWD/transcriptions/EWD06xx/EWD611.html
[Dijkstra, 1986] Dijkstra, E. W. (1986). Mathematicians and computing scientists:
The cultural gap. Mathematical Intelligencer, 8(1):4852.
[Dipert, 1993] Dipert, R. R. (1993). Artifacts, Art Works, and Agency. Temple
University Press, Philadelphia.
[Doxiadis et al., 2009] Doxiadis, A., Papadimitriou, C. H., Papadatos, A., and
Di Donna, A. (2009). Logicomix: An Epic Search for Truth. Bloomsbury USA,
New York.
[Dresner, 2003] Dresner, E. (2003). Effective memory and Turings model of mind.
Journal of Experimental & Theoretical Artificial Intelligence, 15(1):113123.
[Dresner, 2012] Dresner, E. (2012). Turing, Matthews and Millikan: Effective
memory, dispositionalism and pushmepullyou states. International Journal of
Philosophical Studies, 20(4):461472.
[Dretske, 1981] Dretske, F. (1981). Knowledge and the Flow of Information.
Blackwell, Oxford.
[Dublon and Paradiso, 2014] Dublon, G. and Paradiso, J. A. (2014). Extra sensory
perception. Scientific American, 311(1):3641.
[Dumas, 1844] Dumas, A. p. (1844). The Count of Monte Cristo. Penguin Books,
2003, London.
[Dunning, 2007] Dunning, B. (16 May 2007). The importance of teaching critical
thinking. http://skeptoid.com/episodes/4045 (accessed 13 February 2013).
[Dyson, 2004] Dyson, F. (2004). The world on a string. New York Review of Books,
pages 1619.
[Dyson, 2011a] Dyson, F. (2011a). The case for far-out possibilities. New York Review
of Books, pages 2627.
[Dyson, 2011b] Dyson, F. (2011b). How we know. New York Review of Books, pages
8, 10, 12.
[Dyson, 2012a] Dyson, G. (2012a). Turing centenary: The dawn of computing.
Nature, 482(7386):459460. doi:10.1038/482459a.
BIBLIOGRAPHY 785
[Dyson, 2012b] Dyson, G. (2012b). Turings Cathedral. PUB, CITY.

[Easton, 2006] Easton, T. A. (2006). Beyond the algorithmization of the sciences.
[Eco, 1982] Eco, U. (1982). On the impossibility of drawing a map of the empire on
a scale of 1 to 1. In Hot to Travel with a Salmon and Other Essays, pages 95106.
Harcourt Brace, New York.
[Edelman, 2008] Edelman, S. (2008). Computing the Mind. Oxford University Press,
New York.
[Eden, 2007] Eden, A. H. (2007). Three paradigms of computer science. Minds and
Machines, 17(2):135167. online at:
http://www.ic.unicamp.br/wainer/cursos/2s2006/epistemico/filosofia-cs.pdf
[Eden and Turner, 2007] Eden, A. H. and Turner, R. (2007). Problems in the ontology
of computer programs. Applied Ontology, 2(1):1336. Preprint accessed 23 October
2014 from:
http://www.eden-study.org/articles/2007/problems-ontology-programs ao.pdf.
[Editorial, 2006] Editorial (2006). Bush restrained. New York Times, page WK13.
http://www.nytimes.com/2006/09/17/opinion/17sun1.html (accessed 11 December
2013).
[Edmonds and Warburton, 2010] Edmonds, D. and Warburton, N. (2010). Philosophy

Bites. Oxford University Press, Oxford.
[Egan, 2010] Egan, F. (2010). Computational models: A modest role for content.
Studies in History and Philosophy of Science, 41(3):253259.
[Egan, 2012] Egan, F. (2012). Metaphysics and computational cognitive science: Lets
not let the tail wag the dog. Journal of Cognitive Science (South Korea), 13(1):39
49. http://j-cs.org/gnuboard/bbs/board.php?bo table= vol013i1&wr id=2
[Egan, 2014] Egan, F. (2014). How to think about mental content. Philosophical
Studies, 170(1):115135. Preprint at
https://www.academia.edu/4160744/How to think about Mental Content
(accessed 23 February 2015); video at https://vimeo.com/60800468 (accessed
23 February 2015).
[Einstein, 1940] Einstein, A. (1940). Considerations concerning the fundaments of
theoretical physics. Science, 91(2369):487492.
[Ekdahl, 1999] Ekdahl, B. (1999). Interactive computing does not supersede churchs
thesis. The Association of Management and the International Association of
Management, 17th Annual International Conference, San Diego, CA, August 68,
Proceedings Computer Science, 17(2):261265.
786 BIBLIOGRAPHY
[Elser, 2012] Elser, V. (2012). In a class by itself. American Scientist, 100:418420.

http://www.americanscientist.org/bookshelf/pub/in-a-class-by-itself,
http://www.americanscientist.org/authors/detail/veit-elser (both accessed
16 December 2013).
[Ensmenger, 2003] Ensmenger, N. (2003). Bits of history: Review of a.r. burkss who
invented the computer? the legal battle that changed computing history. American
Scientist, 91:467468. http://www.americanscientist.org/bookshelf/pub/bits-of-
history
[Everett, 2012] Everett, M. (2012). Answer to whats the most important question,
and why?. Philosophy Now, 92:3841.
[Eysenck, 1990] Eysenck, M. W. (1990). Artificial intelligence. In Eysenck, M.,

editor, The Blackwell Dictionary of Cognitive Psychology, page 22. Basil Blackwell,
Oxford.
[Feferman, 1992] Feferman, S. (1992). Turings oracle: From absolute to relative

computabilityand back. In Echeverria, J., Ibarra, A., and Mormann, T.,
editors, The Space of Mathematics: Philosophical, Epistemological, and Historical
Exporations, pages 314348. Walter de Gruyter, Berlin.
[Feigenbaum, 2003] Feigenbaum, E. A. (2003). Some challenges and grand

challenges for computational intelligence. Journal of the ACM, 50(1):3240.
[Fekete and Edelman, 2011] Fekete, T. and Edelman, S. (2011). Towards a

computational theory of experience. Consciousness and Cognition, 20(3):807827.
preprint at
http://kybele.psych.cornell.edu/edelman/Fekete-Edelman-ConCog11-in-
press.pdf
(accessed 26 August).
[Fetzer, 1988] Fetzer, J. H. (1988). Program verification: The very idea.

[Feyerabend, 1975] Feyerabend, P. (1975). Against Method: Outline of an Anarchistic

Theory of Knowledge. Verso, London.
[Fisher, 1989] Fisher, L. M. (1989).

Xerox sues apple computer over macintosh copyright. New York Times. Accessed
8 February 2014 from: http://www.nytimes.com/1989/12/15/business/company-
news-xerox-sues-apple-computer-over-macintosh-copyright.html.
[Fitch et al., 2005] Fitch, W. T., Hauser, M. D., and Chomsky, N. (2005). The
evolution of the language faculty: Clarifications and implications. Cognition,
97:179210. Accessed 14 July 2014 from:
http://dash.harvard.edu/bitstream/handle/1/3117935/Hauser EvolutionLanguageFaculty.pdf.
BIBLIOGRAPHY 787
[Flanagan, 2012] Flanagan, O. (2012). Buddhism without the hocus-pocus. The

Chronicle [of Higher Education] Review, 58(19):B4B5.
[Fletcher, 1972] Fletcher, J. (1972). Indicators of humanhood: A tentative profile of

man. Hastings Center Report, 2(5):14.
[Fletcher, 2007] Fletcher, L. R. (2007). Slow reading: The affirmation of authorial

intent. http://www.freelance-academy.org/slowread.htm. Accessed 18 February
2014.
[Floridi, 2010] Floridi, L. (2010). Information: A Very Short Introduction. Oxford

[Florman, 1994] Florman, S. C. (1994). The Existential Pleasures of Engineering,

Second Edition. St. Martins Press, New York.
[Fodor, 1968] Fodor, J. A. (1968). Psychological Explanation: An Introduction to the

Philosophy of Psychology. Random House, New York.
[Fodor, 1974] Fodor, J. A. (1974). Special sciences (or: The disunity of science as a
working hypothesis). Synthese, 28(2):97115.
[Fodor, 1978] Fodor, J. A. (1978). Tom Swift and his procedural grandmother.
Cognition, 6:229247. Accessed 11 December 2013 from:
http://www.nyu.edu/gsas/dept/philo/courses/mindsandmachines/Papers/tomswift.pdf.
[Fodor, 1980] Fodor, J. A. (1980). Methodological solipsism considered as a research

strategy in cognitive psychology. Behavioral and Brain Sciences, 3:63109.
[Fodor, 1981] Fodor, J. A. (1981). The mind-body problem. Scientific American,

244(1):114123.
[Foley, 2002] Foley, J. (2002). Computing > computer science. Computing Research
News, 14(4):6. revised version online at:
http://archive.cra.org/reports/computing/index.html
(accessed 18 June 2013); response by Robert L. Glass online at:
http://archive.cra.org/reports/computing/glass.html
[Forster, 1909] Forster, E. (1909). The machine stops. In ED, editor, TITLE. PUB,
ADDR. http://archive.ncsa.illinois.edu/prajlich/forster.html.
[Forsythe, 1967a] Forsythe, G. E. (1967a). A universitys educational program in

computer science. Communications of the ACM, 10(1):38.
[Forsythe, 1967b] Forsythe, G. E. (1967b). What to do till the computer scientist

comes. American Mathematical Monthly, 75(5):454462.
[Forsythe, 1968] Forsythe, G. E. (1968). Computer science and education.

Proceedings IFIP 68 Cong, pages 92106.
788 BIBLIOGRAPHY
[Fortnow, 2009] Fortnow, L. (2009). The status of the P versus NP problem.

Communications of the ACM, 52(9):7886. Accessed 7 April 2014 from:
http://cacm.acm.org/magazines/2009/9/38904-the-status-of-the-p-versus-np-
problem/fulltext.
[Fortnow, 2010] Fortnow, L. (2010). What is computation? Ubiquity, pages 15.

Article 5 of DOI:10.1145/1895419.1921573. Accessed 28 June 2017 from:
http://people.cs.uchicago.edu/fortnow/papers/compute.pdf.
[Fortnow, 2013] Fortnow, L. (2013). The Golden Ticket: P, NP, and the Search for the
Impossible. Princeton University Press, Princeton, NJ.
[Fox, 2013] Fox, M. (2013). Janos starker, master of the cello, dies at 88. New York
Times, page B16. http://www.nytimes.com/2013/04/30/arts/music/janos-starker-
master-cellist-dies-at-88.html (accessed 12 May 2013).
[Franzen, 2005] Franzen, T. (2005). Godels Theorem: An Incomplete Guide to Its

Use and Abuse. A K Peters, Wellesley, MA.
[Frazer, 1915] Frazer, J. G. (19111915). The Golden Bough: A Study in Magic and
Religion, 3rd ed. Macmillan, London.
[Freeman, 1994] Freeman, P. A. (1994). Effective computer science. ACM Computing

Surveys, 27(1):2729.
[Freeth, 2006] Freeth, T., e. a. (2006). Decoding the ancient greek astronomical
calculator known as the antikythera mechanism. Nature, 444:587591.
[Freeth, 2009] Freeth, T. (2009). Decoding an ancient computer. Scientific American,

301(6):7683.
[French, 2000] French, R. M. (2000). The turing test: The first fifty
years. Trends in Cognitive Sciences, 4(3):115121. http://leadserv.u-
bourgogne.fr/rfrench/french/TICS turing.pdf (accessed 11 October 2015).
[Frenkel, 2013] Frenkel, E. (2013). Love and Math: The Heart of Hidden Reality.
Basic Books, New York.
[Frenkel, 1993] Frenkel, K. (1993). An interview with Robin Milner. Communications

of the ACM, 36(1):9097. Accessed 26 May 2014 from:
http://delivery.acm.org/10.1145/160000/151241/a1991-frenkel.pdf.
[Friedman and Kahn, 1997] Friedman, B. and Kahn, Peter H., J. (1997). People are
responsible, computers are not. In Ermann, M. D., Williams, M. B., and Shauf,
M. S., editors, Computers, Ethics, and Society, Second Edition, pages 303314.
Oxford University Press, New York. Excerpt from their Human Agency and
Responsible Computing: Implications for Computer System Design, Journal of
Systems and Software (1992): 714.
BIBLIOGRAPHY 789
[Frigg and Hartmann, 2012] Frigg, R. and Hartmann, S. (2012). Models in science.
In Zalta, E. N., editor, Stanford Encyclopedia of Philosophy (Fall 2012 Edition).
online. http://plato.stanford.edu/archives/fall2012/entries/models-science/.
[Gal-Ezer and Harel, 1998] Gal-Ezer, J. and Harel, D. (1998). What (xelse) shoud cs
educators know? Communications of the ACM, 41(9):7784.
[Galbi, 1971] Galbi, E. W. (1971). Software and patents: A status report.
Communications of the ACM, 14(4):274280. Accessed 3 September 2014 from:
http://simson.net/ref/2007/cs3610/ref/p274-galbi.pdf.
[Gandy, 1980] Gandy, R. (1980). Churchs thesis and principles for mechanisms. In
Barwise, J., Keisler, H., and Kunen, K., editors, The Kleene Symposium, pages 123
148. North-Holland.
[Gandy, 1988] Gandy, R. (1988). The confluence of ideas in 1936. In Herken, R.,
editor, The Universal Turing Machine: A Half-Century Survey, Second Edition,
pages 51102. Springer-Verlag, Vienna.
[Gazzaniga, 2010] Gazzaniga, M. S. (2010). Neuroscience and the correct level
of explanation for understanding mind: An extraterrestrial roams through some
neuroscience laboratories and concludes earthlings are not grasping how best to
understand the mind-brain interface. Trends in Cognitive Sciences, 14(7):291292.
[Gettier, 1963] Gettier, E. L. (1963). Is justified true belief knowledge? Analysis,
23:121123. Reprinted in A.P. Griffiths (ed.), Knowledge and Belief (Oxford:
Oxford University Press, 1967);
http://www.ditext.com/gettier/gettier.html (accessed 19 September 2011).
[Gigerenzer and Goldstein, 1996] Gigerenzer, G. and Goldstein, D. G. (1996). Mind
as computer: Birth of a metaphor. Creativity Research Journal, 9(23):131
144. http://web.ebscohost.com/ehost/pdfviewer/pdfviewer?sid=9fd0749e-892b-
4f80-8663-7a539038aa82%40sessionmgr198&vid=1&hid=114
[Ginsberg, 1987] Ginsberg, M. L., editor (1987). Readings in Nonmonotonic
Reasoning. Morgan Kaufmann, Los Altos, CA.
[Gladwell, 2011] Gladwell, M. (2011). Creation myth. New Yorker, pages 44,
46, 4850, 52A53A. slightly different version accessed 8 February 2014 from:
http://www.newyorker.com/reporting/2011/05/16/110516fa fact gladwell.
[Gleick, 2008] Gleick, J. (2008). if shakespeare had been able to google. . . . New
York Review of Books, 55(20):7779; highlighted paragraph on p. 78, cols. 23.
[Gleiser, 2014] Gleiser, M. (2014). Does consciousness compute? The Chronicle
Review, 61(2):B13B15.
[Godel, 930s] Godel, K. (ca 1930s). Undecidable diophantine propositions. In
Feferman, S. et al., editors, Collected Works, Vol. III, pages 164175. Oxford
790 BIBLIOGRAPHY
[Gold, 1967] Gold, E. M. (1967). Language identification in the limit. Information

and Control, 10:447474.
[Goldberger, 1979] Goldberger, N. R. (1979). Developmental assumptions underlying

models of general education. Paper presented at the Conference on General
Education, William Patterson College.
[Goldin et al., 2006] Goldin, D., Smolka, S. A., and Wegner, P., editors (2006).
Interactive Computation: The New Paradigm. Springer-Verlag, Berlin. Some
chapters available online at:
http://www.springer.com/computer/theoretical+computer+science/book/978-3-
540-34666-1.
[Goldin and Wegner, 2005] Goldin, D. and Wegner, P. (2005). The church-turing
thesis: Breaking the myth. In Cooper, S. and Lowe, B., editors, CiE: New
Computational Paradigms, pages 152168. Springer LNCS 3526. Accessed
26 February 2014 from: http://www.cse.uconn.edu/dqg/papers/cie05.pdf.
[Goldstein, 2014] Goldstein, R. N. (2014). Plato at the Googleplex: Why Philosophy

Wont Go Away. Pantheon Books, New York.
[Goodman, 1987] Goodman, N. D. (1987). Intensions, churchs thesis, and the

formalization of mathematics. Notre Dame Journal of Formal Logic, 28(4):473
489. Accessed 10 April 2014 from:
http://projecteuclid.org/download/pdf 1/euclid.ndjfl/1093637644.
[Google, 2012] Google, I. (2012). Googles 4/12/12 copyright liability trial

brief, case no. 3:10-cv 03561-wha, document 897. http://www.wired.com/wp-
content/uploads/blogs/wiredenterprise/wp-content/uploads/2012/04/20120412-
Googles-Copyright-Stance.pdf. Accessed 26 September 2014.
[Gopnik, 1996] Gopnik, A. (1996). The scientist as child. Philosophy of Science,

63(4):485
514. http://courses.media.mit.edu/2002fall/mas962/MAS962/gopnik.pdf (accessed
13 May 2015).
[Gopnik, 2009a] Gopnik, A. (2009a). The Philosophical Baby: What Childrens

Minds Tell Us about Truth, Love, and the Meaning of Life. Farrar, Straus and Giroux,
New York.
[Gopnik, 2009b] Gopnik, A. (2009b). Whats the recipe? New Yorker, pages
106112. Accessed 12 May 2014 from:
http://www.newyorker.com/arts/critics/atlarge/2009/11/23/091123crat atlarge gopnik.
[Gopnik, 2013] Gopnik, A. (2013). Moon man: What galileo saw. New Yorker, pages
103109. online at:
http://www.newyorker.com/arts/critics/atlarge/2013/02/11/130211crat atlarge gopnik
BIBLIOGRAPHY 791
[Gries, 1981] Gries, D. (1981). The Science of Programming, 3rd printing. Springer-
Verlag, 1985, New York.
[Grobart, 2011] Grobart, S. (2011). Spoiled by the all-in-one gadget. New York Times,
page WK3. http://www.nytimes.com/2011/03/27/weekinreview/27grobart.html
[Grover, 1999] Grover, L. K. (1999). Quantum computing. The Sciences, pages 24
30. Accessed 9 December 2013 from:
http://cryptome.org/qc-grover.htm.
[Guernsey, 2009] Guernsey, L. (2009). Computers track the elusive metaphor.
Chronicle of Higher Education, page A11.
[Gurevich, 1999] Gurevich, Y. (1999). The sequential ASM thesis. Bulletin of the
European Association for Theoretical Computer Science,
67:93124. http://research.microsoft.com/en-us/um/people/gurevich/Opera/136.pdf
[Gurevich, 2000] Gurevich, Y. (2000). Sequential abstract-state machines capture
sequential algorithms. ACM Transactions on Computational Logic, 1(1):77111.
http://research.microsoft.com/en-us/um/people/gurevich/opera/141.pdf (accessed
17 December 2013).
[Gurevich, 2011] Gurevich, Y. (2011). What is an algorithm? In SOFSEM 2012:
Theory and Practice of Computer Science, pages 3142. Springer Lecture Notes in
Computer Science 7147. http://research.microsoft.com/pubs/155608/209-3.pdf
[Gurevich, 2012] Gurevich, Y. (2012). Foundational analyses of computation.
Technical Report MSR-TR-2012-14, Microsoft Research, Redmond, WA.
http://research.microsoft.com/pubs/158617/210.pdf (accessed 21 January 2014).
[Gurevich, 2013] Gurevich, Y. ((accessed 24 December 2013)). transcript of aaai talk.
http://msrvideo.vo.msecnd.net/rmcvideos/120362/120362.doc. see also:
https://www.facebook.com/events/47127123902/ (accessed 24 December 2013).
[Gutting, 2012] Gutting, G. (2012). Philosophywhats the use? New York Times.
online at http://opinionator.blogs.nytimes.com/2012/01/25/philosophy-whats-the-
use/ (accessed 29 May 2013).
[Guzdial and Kay, 2010] Guzdial, M. and Kay, A. (2010). The core of computer
science:
Alan kays triple whammy. http://computinged.wordpress.com/2010/05/24/the-
core-of-computer-science-alan-kays-triple-whammy/. Part of Guzdials Computing
Education Blog; accessed 28 October 2014.
[Haberman, 2009] Haberman, C. (2009). The story of a landing. New York Times
Book Review. http://www.nytimes.com/2009/11/29/books/review/Haberman-t.html
792 BIBLIOGRAPHY
[Habib, 1983] Habib, S. (1983). Emulation. In Ralston, A. and Edwin D.Reilly, J.,
editors, Encyclopedia of Computer Science and Engineering, 2nd edition, pages
602603. Van Nostrand Reinhold, New York.
[Haigh, 2013] Haigh, T. (2013). stored program concept considered harmful:

History and historiography. In Bonizzoni, P., Brattka, V., and Lowe, B., editors, CiE
2013, pages 241251. Springer-Verlag Lecture Notes in Computer Science 7921,
Berlin.
[Halmos, 1973] Halmos, P. R. (1973). The

legend of John von Neumann. American Mathematical Monthly, pages 382394.
http://poncelet.math.nthu.edu.tw/disk5/js/biography/v-n.pdf,
http://stepanov.lk.net/mnemo/legende.html (accessed 7 April 2014).
[Halpern et al., 2001] Halpern, J. Y., Harper, R., Immerman, N., Kolaitis, P. G., Vardi,
M. Y., and Vianu, V. (2001). On the unusual effectiveness of logic in computer
science. Bulletin of Symbolic Logic, 7(2):213236. online at:
http://www.lsi.upc.edu/roberto/EffectivenessOfLogic.pdf (accessed 17 June
2013).
[Hamming, 1968] Hamming, R. (1968). One mans view of computer science. Journal
of the Association for Computing Machinery, 16(1):312.
[Hamming, 1980] Hamming, R. (1980). The unreasonable effectiveness of

mathematics. American Mathematical Monthly, 87(2).
[Hamming, 1998] Hamming, R. (1998). Mathematics on a distant planet. American

Mathematical Monthly, 105(7):640650.
[Hammond, 2003] Hammond, T. A. (6 May 2003).

What is computer science? myths vs. truths.
http://web.archive.org/web/20030506091438/http://www.columbia.edu/tah10/cs1001/whatcs.html.
[Harman, 1965] Harman, G. (1965). The inference to the best explanation. The
Philosophical Review, 74(1):8895.
[Hartmanis, 1993] Hartmanis, J. (1993). Some observations about the nature of

computer science. In Shyamasundar, R., editor, Foundations of Software Technology
and Theoretical Computer Science, volume 761 of Lecture Notes in Computer
Science, pages 112. Springer Berlin / Heidelberg. 10.1007/3-540-57529-4 39.
[Hartmanis, 1995a] Hartmanis, J. (1995a). On computational complexity and the

nature of computer science. ACM Computing Surveys, 27(1):716. reprinted from
Communications of the ACM 37(10) (October 1994): 3743.
[Hartmanis, 1995b] Hartmanis, J. (1995b). Response to the essays on computational

complexity and the nature of computer science. ACM Computing Surveys,
27(1):5961.
BIBLIOGRAPHY 793
[Hartmanis and Lin, 1992] Hartmanis, J. and Lin, H. (1992). What is computer
science and engineering? In Hartmanis, J. and Lin, H., editors, Computing the
Future: A Broader Agenda for Computer Science and Engineering, pages 163216.
National Academy Press, Washington, DC. Ch. 6.
[Hartmanis and Stearns, 1965] Hartmanis, J. and Stearns, R. (1965). On the

computational complexity of algorithms. Transactions of the American
Mathematical Society, 117:285306.
[Hartmanis and Stearns, 1967] Hartmanis, J. and Stearns, R. (1967). Sets of numbers
defined by finite automata. American Mathematical Monthly, pages 539542.
[Hartree, 1949] Hartree, D. (1949). Calculating Instruments and Machines. University

of Illinois Press, Urbana, IL.
[Haugeland, 1981] Haugeland, J. (1981). Analog and analog. Philosophical Topics,

12(1):213225.
[Hauser et al., 2002] Hauser, M. D., Chomsky, N., and Fitch, W. T. (2002). The faculty
of language: What is it, who has it, and how did it evolve. Science, 298:15691579.
http://www.chomsky.info/articles/20021122.pdf.
[Hayes, 1990] Hayes, B. (1990). No forwarding address. The Sciences, pages 8

11. Accessed 18 March 2014 from http://bit-player.org/wp-content/extras/bph-
publications/Sciences-1990-09-Hayes- forwarding.pdf.
[Hayes, 1995] Hayes, B. (1995). The square root of not. American Scientist,
83:304308. http://www.americanscientist.org/issues/num2/the-square-root-of-not/
[Hayes, 2007a] Hayes, B. (2007a). Calculating the weather. American Scientist, 95(3).
online at:
http://www.americanscientist.org/bookshelf/pub/calculating-the-weather
[Hayes, 2007b] Hayes,

B. (2007b). Trains of thought. American Scientist, 95(2). Accessed 31 January
2014 from: http://www.americanscientist.org/issues/pub/trains-of-thought.
[Hayes, 2014a] Hayes, B. (2014a). Pencil, paper, and pi. American Scientist,
102(1):342345. Accessed 22 October 2014 from:
http://www.americanscientist.org/libraries/documents/20148513205110590-2014-
09CompSciHayes.pdf.
[Hayes, 2014b] Hayes, B. (2014b). Programming your quantum computer. American

Scientist, 102(1):2225. Accessed 19 January 2014 from:
https://www.americanscientist.org/issues/pub/programming-your-quantum-
computer/.
794 BIBLIOGRAPHY
[Hayes, 2015] Hayes, B. (2015). The 100-billion-body problem. American Scientist,

103(2):9093. http://www.americanscientist.org/issues/pub/the-100-billion-body-
problem (accessed 13 May 2015).
[Hayes, 1997] Hayes, P. J. (1997). What is a computer? Monist, 80(3):389404.
[Hearst and Hirsh, 2000] Hearst, M. and Hirsh, H. (2000). Ais greatest trends and
controversies. IEEE Intelligent Systems, 15(1):817. Accessed 2 July 2014 from:
http://www.cs.cornell.edu/courses/cs472/2002fa/handouts/challenges-ai.pdf.
[Hedger, 1998] Hedger, L. (1998). Analog computation: Everything old is new again.
Indiana University Research & Creative Activity, 21(2). Accessed 31 January 2014
from:
http://www.indiana.edu/rcapub/v21n2/p24.html.
[Heng, 2014] Heng, K. (2014). The nature of scientific proof in the age of simulations.
American Scientist, 102(3):174177. Accessed 6 May 2014 from:
http://www.americanscientist.org/issues/pub/2014/3/the-nature-of-scientific-proof-
in-the-age-of-simulations.
[Herman, 1983] Herman, G. (1983). Algorithms, theory of. In Ralston, A. S. and

Riley, E. D., editors, Encyclopedia of Computer Science, 3rd edition, pages 3739.
Van Nostrand Reinhold, New York.
[Hidalgo, 2015] Hidalgo, C. (2015). Planet hard drive. Scientific American,

313(2):7275.
[Higginbothan, 2014] Higginbothan, A. (2014). The disillusionist. New York Times

Magazine, pages 4853, 6061. Online as The Unbelievable Skepticism of the
Amazing Randi, accessed 15 December 2014 from:
http://www.nytimes.com/2014/11/09/magazine/the-unbelievable-skepticism-of-
the-amazing-randi.html.
[Hilbert, 1900] Hilbert, D. (1900). Mathematical problems: Lecture delivered before

the international congress of mathematicians at paris in 1900. Bulletin of the
American Mathematical Society, 8(10):437479. trans. by Mary Winston Newson;
first published in Gottinger Nachrichten (1900): 253297.
[Hilbert and Ackermann, 1928] Hilbert, D. and Ackermann, W. (1928). Principles of

Mathematical Logic. Chelsea, New York (1950). Robert E. Luce (ed.); Lewis M.
Hammon, George G. Leckie, & F. Steinhardt (trans.); based on the 1938 second
edition of Grundzuge der Theoretischen Logik.
[Hill, 2008] Hill, R. (25 September 2008). Empire, regime,

and perspective change in our creative activities. Confer-
ence on ReVisioning the (W)hole II: Curious Intersections.
http://web.archive.org/web/20080925064006/http://www.newhumanities.org/events/revisioning.html
BIBLIOGRAPHY 795
[Hills, 2007] Hills, D. (November 2007). professional website

. http://www.stanford.edu/dhills/cv.html (accessed 12 February 2013).
[Hilpinen, 2011] Hilpinen, R. (2011). Artifact. In Zalta, E. N., editor, Stanford

Encyclopedia of Philosophy. online. http://plato.stanford.edu/entries/artifact/
[Hintikka and Mutanen, 1997] Hintikka, J. and Mutanen, A. (1997). An alternative

concept of computability. In Hintikka, J., editor, Language, Truth, and Logic in
Mathematics, pages 174188. Springer, Dordrecht, The Netherlands.
[Hirst, 1991] Hirst, G. (1991). Existence assumptions in knowledge representation.

Artificial Intelligence, 49:199242.
[Hoare, 1969] Hoare, C. (1969). An axiomatic basis for computer programming.

Communications of the ACM, 12(10):576580, 583.
[Hoare, 1986] Hoare, C. (1986). Mathematics of programming. Byte, page 115ff.

Reprinted in Timothy R. Colburn, James H. Fetzer, & Terry L. Rankin (eds.),
Program Verification: Fundamental Issues in Computer Science (Dordrecht,
Holland: Kluwer Academic Publishers, 1993): 135154.
[Hodges, 2013] Hodges, A. (2013). Churchs review of computable numbers. In

pages 117118. Elsevier, Amsterdam.
[Hoffmann, 2010] Hoffmann, L. (2010). Amir Pnueli: Ahead of his time.

Communications of the ACM, 53(1):2223. http://cacm.acm.org/magazines/2010/1/55750-
amir-pnueli-ahead-of-his-time/fulltext (accessed 4 March 2015).
[Hofstadter, 1979] Hofstadter, D. (1979). Godel, Escher, Bach: An Eternal Golden

Braid. Basic Books, New York.
[Hofstadter, 2007] Hofstadter, D. (2007). I Am a Strange Loop. Basic Books, New

York.
[Hofstadter and Sander, 2013] Hofstadter, D. and Sander, E. (2013). Surfaces and
Essences: Analogy as the Fuel and Fire of Thinking. Basic Books, New York.
[Hollan et al., 2000] Hollan, J., Hutchins, E., and Kirsh, D. (2000). Distributed
cognition: Toward a new foundation for human-computer interaction research. ACM
Transactions on Computer-Human Interaction, 7(2):174196.
[Holland, 2012] Holland, J. H. (2012). Signals and Boundaries: Building Blocks for
Complex Adaptive Systems. MIT Press, Cambridge, MA.
[Holst, 2000] Holst, P. A. (2000). Analog computer. In Ralston, A., Reilly, E. D., and
Hemmendinger, D., editors, Encyclopedia of Computer Science, Fourth Edition,
pages 5359. Groves Dictionaries, New York.
796 BIBLIOGRAPHY
[Holt, 2009] Holt, J. (2009). Death: Bad? New York Times Book Review, page BR27.
http://www.nytimes.com/2009/02/15/books/review/Holt-t.html.
[Hopcroft and Ullman, 1969] Hopcroft, J. E. and Ullman, J. D. (1969). Formal
Languages and Their Relation to Automata. Addison-Wesley, Reading, MA.
[Hopper, 1981] Hopper, G. M. (1981). The first bug. Annals of the History of
Computing, 3(3):285286. online at:
http://ieeexplore.ieee.org/stamp/stamp.jsp?reload=true&tp=&arnumber=4640691
[Horwich, 2013] Horwich, P. (2013). Was wittgenstein right? New York Times.
online at http://opinionator.blogs.nytimes.com/2013/03/03/was-wittgenstein-right/
[Huber, 1966] Huber, H. G. (1966). Algorithm and formula. Communications of the
ACM, 9(9):653654.
[Huemer, 2010] Huemer, W. (2010). Franz brentano. In Zalta, E. N., editor, Stanford
Encyclopedia of Philosophy. online. http://plato.stanford.edu/entries/brentano/.
[Hugo, 1862] Hugo, V. (1862). Les Miserables. Signet Classics, 1987, New York.
[Humphreys, 1990] Humphreys, P. (1990). Computer simulations. PSA: Proceedings
of the [1990] Biennial Meeting of the Philosophy of Science Association, 2:497
506.
[Humphreys, 2002] Humphreys, P. (2002). Computational models. Philosophy of
Science, 69:S1S11. http://ist.uap.asia/jpira/compthink/17232634.pdf (accessed
12 January 2015).
[Hutchins, 1995a] Hutchins, E. (1995a). Cognition in the Wild. MIT Press,
Cambridge, MA.
[Hutchins, 1995b] Hutchins, E. (1995b). How a cockpit remembers its speeds.
Cognitive Science, 19:265288.
[Huxley, 1932] Huxley, A. (1932). Brave New World. online.
http://www.huxley.net/bnw/.
[Irmak, 2012] Irmak, N. (2012). Software is an abstract artifact. Grazer
Philosophische Studien, 86:5572. Accessed 4 August 2014 from:
http://philpapers.org/archive/IRMSIA.pdf.
[Jackendoff, 2012] Jackendoff, R. (2012). A Users Guide to Thought and Meaning.
Oxford University Press, Oxford.
[Jackendoff and Pinker, 2005] Jackendoff, R. and Pinker, S. (2005). The nature of the
language faculty and its implications for evolution of language. Cognition, 97:211
225. Accessed 14 July 2014 from:
http://pinker.wjh.harvard.edu/articles/papers/2005 09 Jackendoff Pinker.pdf.
BIBLIOGRAPHY 797
[Jackson, 1960] Jackson, A. S. (1960). Analog Computation. McGraw-Hill, New

York.
[Jacob, 1999] Jacob, C. (1999). What is computer science? Chapter 3;
http://pages.cpsc.ucalgary.ca/jacob/Courses/Fall99/CPSC231/03-
ComputerScience.pdf, accessed 20 September 2011.
[Jacob, 2010] Jacob, P. (2010). Intentionality. In Zalta, E. N., editor, Stanford
Encyclopedia of Philosophy. online. http://plato.stanford.edu/entries/intentionality/.
[James, 1892] James, W. (1892). Psychology: Briefer course. In Myers, G. E., editor,
William James: Writings 18781899, pages 1443. Library of America, New York.
Original edition accessed 16 September 2014 from:
https://archive.org/details/psychologybriefe00willuoft.
[James, 1897] James, W. (1897). The will to believe. In Myers, G. E., editor,
William James: Writings 18781899, pages 457479. Library of America, New
York. Project Gutenberg version accessed 16 September 2014 from:
http://www.gutenberg.org/files/26659/26659-h/26659-h.htm.
[Johnson, 2001] Johnson, G. (2001). All science is computer science. New York Times.
online at:
http://www.nytimes.com/2001/03/25/weekinreview/the-
world-in-silica-fertilization-all-science-is-computer-science.html (accessed 11 June
2013).
[Johnson, 2002] Johnson, G. (2002). To err is human. New York times. online at:
http://www.nytimes.com/2002/07/14/weekinreview/deus-ex-machina-to-err-is-
human.html (accessed 11 June 2013).
[Johnson-Laird, 1981] Johnson-Laird, P. N. (1981). Mental models in cognitive
science. In Norman, D. A., editor, Perspectives on Cognitive Science, chapter 7,
pages 147191. Ablex, Norwood, NJ.
[Johnson-Laird, 1988] Johnson-Laird, P. N. (1988). The Computer and the Mind: An
Introduction to Cognitive Science. Harvard University Press, Cambridge, MA. Ch.
3 (Computability and Mental Processes), pp. 3753.
[Kahneman, 2011] Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Strauss
and Giroux, New York.
[Kaiser, 2012] Kaiser, D. (2012). Paradoxial roots of social construction. Science,
335:658659.
[Kalmar, 1959] Kalmar, L. (1959). An argument against the plausibility of churchs
thesis. In Heiting, A., editor, Constructivity in Mathematics, pages 7280. North-
Holland, Amsterdam.
[Kanat-Alexander, 2008] Kanat-Alexander, M. (10 October 2008). What is a
computer? http://www.codesimplicity.com/post/what-is-a-computer/. Accessed
29 March 2014 from Code Simplicity blog.
798 BIBLIOGRAPHY
[Kant, 1787] Kant, I. (1781/1787). Critique of Pure Reason. St. Martins Press, New
York. translation published 1929.
[Kay, 2010] Kay, M. (2010). Introduction. In Flickinger, D. and Oepen, S., editors,
Collected Papers of Martin Kay: A Half Century of Computational Linguistics,
pages 118. CSLI Studies in Computational Linguistics, Stanford, CA.
[Kearns, 1997] Kearns, J. (1997). Thinking machines: Some fundamental confusions.

Minds and Machines, 7(2):269287.
[Keith, 2014] Keith, T. (2014). The letter that kicked off a radio career. NPR.org.
http://www.npr.org/2014/07/05/328512614/the-letter-that-kicked-off-a-radio-
career.
[Kemeny, 1959] Kemeny, J. G. (1959). A Philosopher Looks at Science.

D. van Nostrand, Princeton, NJ.
[Khalil and Levy, 1978] Khalil, H. and Levy, L. S. (1978). The academic image of
computer science. ACM SIGCSE Bulletin, 10(2):3133.
[Kirk, 1974] Kirk, R. (1974). Sentience and behaviour. Mind, 83(329):4360.
[Kleene, 1936a] Kleene, S. C. (1935-1936a). General recursive functions of natural

numbers. Mathematische Annalen, 112:727742.
[Kleene, 1936b] Kleene, S. C. (1936b). -definability and recursiveness. Duke

Mathematical Journal, 2:340353.
[Kleene, 1952] Kleene, S. C. (1952). Introduction to Metamathematics. D. Van

Nostrand, Princeton, NJ.
[Kleene, 1967] Kleene, S. C. (1967). Mathematical Logic. Wiley, New York.
[Kleene, 1987] Kleene, S. C. (1987). Reflections on churchs thesis. Notre Dame

Journal of Formal Logic, 28(4):490498. Accessed 9 April 2014 from:
http://projecteuclid.org/download/pdf 1/euclid.ndjfl/1093637645.
[Kleene, 1995] Kleene, S. C. (1995). Turings analysis of computability, and major

applications of it. In Herken, R., editor, The Universal Turing Machine: A Half-
Century Survey, Second Edition, pages 1549. Springer-Verlag, Vienna.
[Kling et al., 1993] Kling, R., Wegner, P., Rice, J. R., and Weiss, E. A. (1993).
Broadening computer science. Communications of the ACM, 36(2):1519.
[Knobe, 2009] Knobe, J. (2008/2009). Can a robot, an insect, or god be aware?

Scientific American Mind, 19(6):6871.
[Knuth, 1966] Knuth, D. E. (1966). Algorithm and program: Information and data.
BIBLIOGRAPHY 799
[Knuth, 972a] Knuth, D. E. (1972a). Ancient babylonian algorithms. Communications

of the ACM, 15(7):671677.
[Knuth, 972b] Knuth, D. E. (1972b). George forsythe and the development of
computer science. Communications of the ACM, 15(8):721727.
[Knuth, 1973] Knuth, D. E. (1973). The Art of Computer Programming, Second

Edition. Addison-Wesley, Reading, MA.
[Knuth, 1974a] Knuth, D. E. (1974a). Computer programming as an art.
[Knuth, 1974b] Knuth, D. E. (1974b). Computer science and its relation to
mathematics. American Mathematical Monthly, 81(4):323343.
[Knuth, 1985] Knuth, D. E. (1985). Algorithmic thinking and mathematical thinking.
American Mathematical Monthly, 92(3):170181.
[Knuth, 2001] Knuth, D. E. (2001). Things a Computer Scientist Rarely Talks About.
CSLI Publications, Stanford, CA. CSLI Lecture Notes Number 136.
[Knuth, 2014] Knuth, D. E. (2014). Twenty questions for donald knuth. Informit.com.
Accessed 3 June from:
http://www.informit.com/articles/article.aspx?p=2213858&WT.mc id=Author Knuth 20Questions.
[Koen, 1988] Koen,
B. V. (1988). Toward a definition of the engineering method. European Journal
of Engineering Education, 13(3):307315. reprinted from Engineering Education
(December 1984): 150155; http://dx.doi.org/10.1080/03043798808939429.
[Koepsell, 1998] Koepsell, D. R. (1998). Review of smith 1996. Philosophical
Psychology, 11(3):389391.
[Koepsell, 2000] Koepsell, D. R. (2000). The Ontology of Cyberspace: Philosophy,

Law, and the Future of Intellectual Property. Open Court, Chicago.
[Kolata, 2004] Kolata, G. (2004). New studies question value of opening arteries. New
York Times, pages A1, A21. http://www.nytimes.com/2004/03/21/us/new-heart-
studies-question-the-value-of-opening-arteries.html (accessed 14 November 2012).
[Kornblith, 2013] Kornblith, H. (2013). Naturalism vs. the first-person perspective.

Proceedings & Addresses of the American Philosophical Association, 87:122141.
[Korsmeyer, 2012] Korsmeyer, C. (2012). Touch and the experience of the genuine.
British Journal of Aesthetics, 52(4):365377.
[Kramer, 2007] Kramer, J. (2007). Is abstracton the key to computing?

[Krantz, 1984] Krantz, S. G. (1984). Letter to the editor. American Mathematical
Monthly, 91(9):598600.
800 BIBLIOGRAPHY
[Krebs and Thomas, 1981] Krebs, A. and Thomas, Jr., R. M. (1981). Historic moth.
New York Times. online at:
http://www.nytimes.com/1981/08/07/nyregion/notes-on-people-historic-moth.html
[Kripke, 2013] Kripke, S. A. (2013). The church-turing thesis as a special corollary
of godels completeness theorem. In Copeland, B. J., Posy, C. J., and Shagrir, O.,
editors, Computability: Turing, Godel, Church, and Beyond, pages 77104. MIT
[Kugel, 2002] Kugel, P. (2002). Computing machines cant be intelligent (. . . and
turing said so). Minds and Machines, 12(4):56359?
[Kuhn, 1957] Kuhn, T. S. (1957). The Copernican Revolution: Planetary Astronomy
in the Development of Western Thought. Harvard University Press, Cambridge, MA.
[Kuhn, 1962] Kuhn, T. S. (1962). The Structure of Scientific Revolutions. University
of Chicago Press, Chicago.
[Kumar, 1994] Kumar, D. (1994). From beliefs and goals to intentions and actions:
An amalgamated model of inference and acting. http://www.cse.buffalo.edu/tech-
reports/94-04.ps. Unpublished PhD dissertation, Department of Computer Science,
SUNY Buffalo.
[Kumar, 1996] Kumar, D. (1996). The SNePS BDI architecture. Decision Support
Systems, 16(1):319. http://www.cse.buffalo.edu/sneps/Bibliography/kum96.pdf
[LaChat, 1986] LaChat, M. R. (1986). Artificial intelligence and ethics: An exercise
in the moral imagination. AI Magazine, 7(2):7079.
[Ladyman, 2009] Ladyman, J. (2009). What does it mean to say that a physical system
implements a computation? Theoretical Computer Science, 410(45):376383.
http://ac.els-cdn.com/S0304397508007238/1-s2.0-S0304397508007238-main.pdf? tid=04248730-5f6c-
11e4-8b44-00000aab0f26&acdnat=1414588006 ff4ead0d50f1774e6d0779cf705e19e6
[Lakoff, 1987] Lakoff, G. (1987). Women, Fire, and Dangerous Things: What
Categories Reveal about the Mind. University of Chicago Press, Chicago.
[Lakoff and Johnson, 1980a] Lakoff, G. and Johnson, M. (1980a). Conceptual
metaphor in everyday language. Journal of Philosophy, 77(8):453486.
[Lakoff and Johnson, 1980b] Lakoff, G. and Johnson, M. (1980b). Meaphors We Live
By. University of Chicago Press, Chicago.
[Lamport, 2011]
Lamport, L. (2011). Euclid writes an algorithm: A fairytale. International Journal
of Software and Informatics, 5(1-2, Part 1):720. http://research.microsoft.com/en-
us/um/people/lamport/pubs/euclid.pdf (accessed 23 April 2015), page references to
PDF version.
BIBLIOGRAPHY 801
[Lamport, 2012] Lamport, L. (2012). How to write a 21st century proof. Journal of
Fixed Point Theory and Applications, 11(1):4363. Accessed 6 October 2014 from:
http://research.microsoft.com/en-us/um/people/lamport/pubs/proof.pdf.
[Lamport, 2015] Lamport, L. (2015). Who builds a house without drawing blueprints?
Communications of the
ACM, 58(4):3841. http://cacm.acm.org/magazines/2015/4/184705-who-builds-a-
house-without-drawing-blueprints/fulltext (accessed 18 April 2015).
[Langewiesche, 2009] Langewiesche, W. (2009). Fly by Wire: The Geese, the Glide,
the Miracle on the Hudson. Farrar, Strauss & Giroux, New York?
[Lanier, 2005] Lanier, J. (2005). Early computings long, strange trip. American
Scientist, 93(4):364365. http://www.americanscientist.org/bookshelf/pub/early-
computings-long-strange-trip (accessed 12 November 2013).
[Lazowska, 2014] Lazowska, E. (2014). Letter to the editor. New York Times, page
A26. Accessed 26 May 2014 from:
http://www.nytimes.com/2014/05/16/opinion/should-young-children-learn-
coding.html.
[Leibniz, 1685] Leibniz, G. W. (16831685). Introduction to a secret encyclopedia. In

Dascal, M., editor, G.W. Leibniz: The Art of Controversies, pages 219224. Springer
(2008), Dordrecht, The Netherlands. http://tinyurl.com/Leibniz1683 (accessed
6 December 2013).
[Leibniz, 1714] Leibniz, G. W. (1714). The principles of philosophy known as

monadology. http://www.earlymoderntexts.com/pdfs/leibniz1714b.pdf. Trans. by
Jonathan Bennett (July 2007). Accessed 29 March 2014 from Some Texts from Early
Modern Philosopny.
[Leiter, 2005] Leiter, B. (12 October 2005). Why is there a nobel prize in economics?
Leiter Reports: A Philosophy
Blog. http://leiterreports.typepad.com/blog/2005/10/why is there a .html (accessed
4 November 2013).
[Leiter, 2009] Leiter, B.

(20 September 2009). Alex rosenberg on cochrane and economics. Leiter Reports:
A Philosophy Blog. http://leiterreports.typepad.com/blog/2009/09/alex-rosenberg-
on-cochrane-and-economics.html (accessed 4 November 2013).
[Leiter, 2004] Leiter, B. (8 October 2004). Is economics a science? Leiter Reports:

A Philosophy Blog.
http://leiterreports.typepad.com/blog/2004/10/is economics a .html (accessed
4 November 2013).
[Leler, 1985] Leler, W. (3 December 1985). newsgroup message. net.ai. contribution

to discussion of definition of AI.
802 BIBLIOGRAPHY
[Lem, 1971] Lem, S. (1971). Non serviam. In A Perfect Vacuum. Harcourt Brace
Jovanovich, New York. 1979.
[Lemonick, 2015] Lemonick, M.
(2015). The pluto wars revisited. http://www.newyorker.com/tech/elements/nasa-
dawn-ceres-pluto-dwarf-planets. The New Yorker online.
[Lessing, 1778] Lessing, G. E. (1778). Anti-goetze: Eine duplik. In
Gopfert, H., editor, Werke, pages Vol. 8, pp. 3233. (published online).
http://harpers.org/blog/2007/11/lessings-search-for-truth/ (accessed 5 February
2013).
[Lewis, 1953] Lewis, W. (1953). Electronic computers and telephone switching.
Proceedings of the Institute of Radio Engineers, 41(10):12421244.
[Lewontin, 2014] Lewontin, R. (2014). The new synthetic biology: Who gains? New
York Review of Books, 61(8):2223. Accessed 19 May 2014 from:
http://www.nybooks.com/articles/archives/2014/may/08/new-synthetic-biology-
who-gains/.
[Licklider and Taylor, 1968] Licklider, J. and Taylor, R. W. (1968). The
computer as a communication device. Science and Technology. online at
http://memex.org/licklider.pdf (accessed 22 May 2013).
[Lightman, 2012] Lightman, A. (2012). Mr g. Pantheon, New York.
[Lindell, 2001] Lindell, S. (24 January 2001). Computer science as a liberal art: The
convergence of technology and reason. talk presented at Haverford College. online
at:
http://www.haverford.edu/cmsc/slindell/Presentations/Computer%20Science%20as%
20a%20Liberal%20Art.pdf (accessed 1 July 2013).
[Lipton, 2004] Lipton, P. (2004). Inference to the Best Explanation. Routledge, 2nd
edition edition.
[Lloyd and Ng, 2004] Lloyd, S. and Ng, Y. J. (2004). Black hole computers. Scientific
American, 291(5):5261.
[Lohr, 2006] Lohr, S. (2006). Group of university researchers to make web science a
field of study. New York Times, page C6. Accessed 3 July 2013 from:
http://www.nytimes.com/2006/11/02/technology/02compute.html.
[Lohr, 2008] Lohr, S. (2008). Does computing add up in the classroom? New York
Times Bits (blog). Accessed 8 July 2013 from:
http://bits.blogs.nytimes.com/2008/04/01/does-computing-add-up-in-the-
classroom/.
[Lohr, 2013] Lohr, S. (2013). Algorithms get a human hand in steering web. New York
Times. Accessed 18 July 2014 from:
http://www.nytimes.com/2013/03/11/technology/computer-algorithms-rely-
increasingly-on-human-helpers.html.
BIBLIOGRAPHY 803
[Loui, 1987] Loui, M. C. (1987). Computer science is an engineering discipline.

Engineering Education, 78(3):175178.
[Loui, 1995] Loui, M. C. (1995). Computer science is a new engineering discipline.

ACM Computing Surveys, 27(1):3132.
[Loui, 1996] Loui, M. C. (1996). Computational com-

plexity theory. ACM Computing Surveys, 28(1):4749.
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.15.9179&represent=rep1&type=pdf
[Loui, 1998] Loui, R. (1998). Review of smith 1996. Artificial Intelligence, 106:353
358.
[Lycan, 1990] Lycan, W. G. (1990). The continuity of levels of nature. In Lycan,

W. G., editor, Mind and Cognition: A Reader, pages 7796. Basil Blackwell,
Cambridge, MA. Accessed 6 October 2014 from:
http://mugwump.pitzer.edu/bkeeley/CLASS/FoNS/lycan.pdf.
[Lynch, 2013]
Lynch, M. P. (2013). Of flies and philosophers: Wittgenstein and philosophy. New
York Times. online at http://opinionator.blogs.nytimes.com/2013/03/05/of-flies-and-
philosophers-wittgenstein-and-philosophy/ (accessed 29 May 2013).
[Macari, 2012] Macari, M. (2012). Oracle thinks you can copyright a programming
language, google disagrees. The Verge. Accessed 26 September 2014 from:
http://www.theverge.com/2012/4/13/2944440/google-oracle-lawsuit-programming-
language-copyright.
[MacFarlane, 2013] MacFarlane, A. (2013). Information, knowledge and intelligence.

Philosophy Now, 98:18
20. http://philosophynow.org/issues/98/Information Knowledge and Intelligence
[Machery, 2012] Machery, E. (2012). Why i stopped worrying about the definition
of life. . . and why you should as well. Synthese, 185:145164. Preprint accessed
2 May 2014 from:
http://www.pitt.edu/machery/papers/Definition
[MacKenzie, 1992] MacKenzie, D. (1992). Computers, formal proofs, and the law
courts. Notices of the American Mathematical Society, 39(9):10661069. Also see
introduction by Keith Devlin, same issue, pp. 10651066.
[Madigan, 2014] Madigan, T. (2014). A mind is a wonderful thing to

meet. Philosophy Now, 100:4647. Accessed 26 February 2014 from:
http://philosophynow.org/issues/100/A Mind is a Wonderful Thing to Meet.
[Mahoney and Haigh (ed.), 2011] Mahoney, M. S. and Haigh (ed.), T. (2011).
Histories of Computing. Harvard University Press, Cambridge, MA.
804 BIBLIOGRAPHY
[Maida and Shapiro, 1982] Maida, A. S. and Shapiro, S. C. (1982). Intensional

concepts in propositional semantic networks. Cognitive Science, 6:291330.
reprinted in Ronald J. Brachman & Hector J. Levesque (eds.), Readings in
Knowledge Representation (Los Altos, CA: Morgan Kaufmann, 1985): 169189.
[Malcolm, 2001] Malcolm, N. (2001). Ludwig Wittgenstein: A Memoir, with a

Biographical Sketch by G.H. von Wright, Second Edition, with Wittgensteins Letters
to Malcolm. Clarendon Press, Oxford. Originally published in 1984.
[Malpas, 2000] Malpas, R. (2000). The universe of engineering: A uk

perspective. Technical report, Royal Academy of Engineering, London.
http://www.engc.org.uk/ecukdocuments/internet/document%20library/The%20Universe%20of%20Engineering%20Report%20(The%20Malpas%20Report).pdf,
accessed 5 October 2011.
[Mander, 2007] Mander, K. (February 2007). Demise of computer science

exaggerated. BCS: The Chartered Institute for IT; Features, Press and Policy.
http://www.bcs.org/content/ConWebDoc/10138.
[Manovich, 2013] Manovich, L. (2013). The algorithms of our lives. The Chronicle
[of Higher Education] Review, 60(16):B10B13. Accessed 7 April 2014 from:
https://chronicle.com/article/The-Algorithms-of-Our-Lives-/143557/.
[Manzano, 1997] Manzano, M. (1997). Alonzo church: His life, his work and some
of his miracles. History and Philosophy of Logic, 18:211232.
[Markov, 1954] Markov, A. (1954). Theory of algorithms. Tr. Mat. Inst. Steklov,
42:114. trans. by Edwin Hewitt, in American Mathematical Society Translations,
Series 2, Vol. 15 (1960).
[Marr, 1982] Marr, D. (1982). Vision: A Computational Investigation into the Human
Representation and Processing of Visual Information. W.H. Freeman, New York.
[Martin, 2015] Martin, C. (2015). She seconds that emotion. The Chronicle [of
Higher Education] Review, 61(33):B16. http://chronicle.com/article/Choosing-
Love/229569/ (accessed 14 June 2015).
[Marx, 1845] Marx, K. (1845). Theses on Feuerbach.

https://www.marxists.org/archive/marx/works/1845/theses/theses.htm
[McAllister, 2012] McAllister, N. (2012). Oracle vs. google: Who owns the java apis?
InfoWorld. Accessed 26 September 2014 from:
http://www.infoworld.com/article/2617268/java/oracle-vs--google--who-owns-
the-java-apis-.html.
[McBride, 2007] McBride, N. (22 January 2007). The death of computing. BCS:
The Chartered Institute for IT; Features,
Press and Policy. http://www.bcs.org/content/ConWebDoc/9662 (accessed 3 July
2013).
BIBLIOGRAPHY 805
[McCarthy, 1988] McCarthy, J. (1 November 1988). contribution to newsgroup

discussion of ai as cs and the scientific epistemology of the commone sense world.
Article 1818 of comp.ai.digest. http://www.cse.buffalo.edu/rapaport/mccarthy.txt.
[McCarthy, 2007] McCarthy, J. (12 November 2007). What is ai? http://www-
formal.stanford.edu/jmc/whatisai.html.
[McCarthy, 1959] McCarthy, J. (1959). Programs with common sense. In Blake,
D. and Uttley, A., editors, Proceedings of the [Teddington] Symposium on
Mechanization of Thought Processes. HM Stationary Office, London. Original
version accessed 7 July 2014 from:
http://aitopics.org/sites/default/files/classic/TeddingtonConference/Teddington-1.3-
McCarthy.pdf
Version accessed 7 July 2014 from McCarthys archive at:
http://www-formal.stanford.edu/jmc/mcc59/mcc59.html.
[McCarthy, 1963] McCarthy, J. (1963). A basis for a mathematical theory of
computation. In Braffort, P. and Hirshberg, D., editors, Computer Programming and
Formal Systems. North-Holland. http://www-formal.stanford.edu/jmc/basis.html
(accessed 13 May 2015), page references to online PDF version.
[McCarthy, 1974] McCarthy, J. (1974). Review of artificial intelligence: A general
survey. Artificial Intelligence, 5:317322.
[McCarthy et al., 1955] McCarthy, J., Minsky, M., Rochester, N., and Shannon, C.
(31 August 1955). A proposal for the dartmouth summer research project on
artificial intelligence. http://www-formal.stanford.edu/jmc/history/dartmouth.html.
[McDermott, 2001] McDermott, D. (2001). Mind and Mechanism. MIT Press,
Cambridge, MA.
[McDermott, 2014] McDermott, D. (2014). On the claim that a table-lookup program
could pass the turing test. Minds and Machines, 24(2):143188. Accessed 29 April
2014 from:
http://cs-www.cs.yale.edu/homes/dvm/papers/humongous.pdf
with supplementary material at:
http://link.springer.com/content/esm/art:10.1007/s11023-013-9333-
3/file/MediaObjects/11023 2013 9333 MOESM1 ESM.pdf.
[McGinn, 2015a] McGinn, C. (2015a). Philosophy of Language: The Classics
Explained. MIT Press, Cambridge, MA.
[McGinn, 2015b] McGinn, C. (2015b). The science of philosophy. Metaphilosophy,
46(1):84103.
http://onlinelibrary.wiley.com/store/10.1111/meta.12116/asset/meta12116.pdf?v=1&t=i5dvrsm5&s=210194c1272c25e8bac66889cc9db416aac0340d
(accessed 26 January 2015); video online at

https://www.youtube.com/watch?v=TEkTbq5EE M
(accessed 26 January 2015); preprint online at
https://docs.google.com/file/d/0BzokFqaWjk4JZ0NPMHFiUnNnZWs/edit
806 BIBLIOGRAPHY
[McGinn, 2012b] McGinn, C. (4 March 2012b). Philosophy by another name. The

Stone/The New York
Times Opinionator. http://opinionator.blogs.nytimes.com/2012/03/04/philosophy-
by-another-name/ (accessed 12 February 2013).
[McGinn, 2012a] McGinn, C. (9 March 2012a). Name calling: Philosophy as ontical
science. The Stone/The New
York Times Opinionator. http://opinionator.blogs.nytimes.com/2012/03/09/name-
calling-philosophy-as-ontical-science/ (accessed 12 February 2013).
[McMillan, 2013] McMillan, R. (7 July 2013). The end of digital tyranny: Why the
future of computing
is analog. Wired. http://www.wired.com/wiredenterprise/2013/07/analogfuture/
[McSherry, 2014] McSherry, C. (2014).
Dangerous decision in oracle v. google: Federal circuit reverses sensible court ruling
on apis. https://www.eff.org/deeplinks/2014/05/dangerous-ruling-oracle-v-google-
federal-circuit-reverses-sensible-lower-court. Accessed 26 September 2014 from
Electronic Frontier Foundation.
[Melville, 2002] Melville, H. (1851/2002). Moby-Dick (Norton Critical Edition,
second edition). W.W. Norton, New York. Hershel Parker & Harrison Hayford
(eds.).
[Menabrea, 1843] Menabrea, L. F. (1843). English translation of notions sur la
machine analytique de m. charles babbage. [Richard Taylors] Scientific Memoirs,
3:666ff. translation by Augusta Ada Lovelace.
[Mendelson, 1990] Mendelson, E. (1990). Second thoughts about churchs thesis and
mathematical proofs. Journal of Philosophy, 87(5):225233.
[Mertens, 2004] Mertens, S. (2004). The revolution will be digitized. American
Scientist, 92:195196.
[Mervis and Rosch, 1981] Mervis, C. B. and Rosch, E. (1981). Categorization of
natural objects. Annual Review of Psychology, 32:89115.
[Micali, 2015]
Micali, S. (2015). What it means to receive the turing award. Communications
of the ACM, 58(1):5253. http://cacm.acm.org/magazines/2015/1/181611-what-it-
means-to-receive-the-turing-award/fulltext
[Michie, 1971] Michie, D. (1971). Formation and execution of plans by machine.
In Findler, N. and Meltzer, B., editors, Artificial Intelligence and Heuristic
Programming, pages 101124. American Elsevier, New York.
[Mili et al., 1986] Mili, A., Desharnais, J., and Gagne, J. R. (1986). Formal models of
stepwise refinement of programs. ACM Computing Surveys, 18(3):231276.
BIBLIOGRAPHY 807
[Miller et al., 1960] Miller, G. A., Galanter, E., and Pribram, K. H. (1960). Plans and
the Structure of Behavior. Henry Holt, New York.
[Milner, 1993] Milner, R. (1993). Elements of interaction: Turing award lecture.
Communications of the ACM, 36(1):7889. Accessed 26 May 2014 from:
http://delivery.acm.org/10.1145/160000/151240/a1991-milner.pdf.
[Minsky, 1967] Minsky, M. (1967). Computation: Finite and Infinite Machines.
Prentice-Hall, Englewood Cliffs, NJ.
[Minsky, 1968] Minsky, M. (1968). Preface. In Minsky, M., editor, Semantic
Information Processing, page v. MIT Press, Cambridge, MA.
[Minsky, 1979] Minsky, M. (1979). Computer science and the representation of
knowledge. In Dertouzos, L. and Moses, J., editors, The Computer Age: A Twenty
Year View, pages 392421. MIT Press, Cambridge, MA.
[Mish, 1983] Mish, F. C., editor (1983). Websters Ninth New Collegiate Dictionary.
Merriam-Webster, Springfield, MA.
[Mitcham, 1994] Mitcham, C. (1994). Thinking through Techology: The Path between
Engineering and Philosophy. University of Chicago Press, Chicago.
[Mizoguchi and Kitamura, 2009] Mizoguchi, R. and Kitamura, Y. (2009). A
functional ontology of artifacts. The Monist, 92(3):387402.
[Monroe and Wineland, 2008] Monroe, C. R. and Wineland, D. J. (2008). Quantum
computing with ions. Scientific American, pages 6471. Accessed 29 March 2014
from http://www.cs.virginia.edu/robins/Quantum Computing with Ions.pdf.
[Montague, 1960] Montague, R. (1960). Towards a general theory of computability.
Synthese, 12(4):429438.
[Montague, 1970] Montague, R. (1970). English as a formal language. In Thomason,
R. H., editor, Formal Philosophy: Selected Papers of Richard Montague, pages 192
221. Yale University Press, 1974, New Haven, CT. Accessed 10 February 2014
from: http://strangebeautiful.com/uwo/metaphys/montague-formal-philosophy.pdf.
[Mooers, 1975] Mooers, C. N. (1975). Computer software and copyright. Computing
Surveys, 7(1):4572.
[Moor, 1978] Moor, J. H. (1978). Three myths of computer science. British Journal
for the Philosophy of Science, 29(3):213222.
[Moor, 1979] Moor, J. H. (1979). Are there decisions computers should never make?
Nature and System, 1:217229.
[Moor, 1985]
Moor, J. H. (1985). What is computer ethics? Metaphilosophy, 16(4):266
275. http://web.cs.ucdavis.edu/rogaway/classes/188/spring06/papers/moor.html
808 BIBLIOGRAPHY
[Moor, 2003] Moor, J. H., editor (2003). The Turing Test: The Elusive Standard of
Artificial Intelligence. Kluwer Academic, Dordrecht, The Netherlands.
[Moravec, 1998] Moravec, H. (1998). When will

computer hardware match the human brain? Journal of Evolution and Technology,
1(1). Accessed 24 March 2014 from http://www.jetpress.org/volume1/moravec.htm
and http://www.jetpress.org/volume1/moravec.pdf.
[Morris, 1938] Morris, C. (1938). Foundations of the Theory of Signs. University of

Chicago Press, Chicago.
[Morris and Reilly, 2000] Morris, G. J. and Reilly, E. D. (2000). Digital computer.
In Ralston, A., Reilly, E. D., and Hemmendinger, D., editors, Encyclopedia of
Computer Science, Fourth Edition, pages 539545. Groves Dictionaries, New York.
[Mycielski, 1983] Mycielski, J. (1983). The meaning of the conjecture p 6= np for

mathematical logic. American Mathematical Monthly, 90:129130.
[Nagel et al., 2001] Nagel, E., Newman, J. R., and Hofstadter, D. R. (2001). Godels
Proof, Revised Edition. New York University Press, New York.
[Nagel, 1987] Nagel, T. (1987). What Does It All Mean? A Very Short Introduction to
Philosophy. Oxford University Press, New York.
[Nahmias et al., 2006] Nahmias, E., Morris, S. G., Nadelhoffer, T., and Turner, J.
(2006). Is incompatibilism intuitive? Philosophy and Phenomenological Research,
73(1):2853.
[Natarajan, 2014] Natarajan, P. (2014). What scientists really do. New York Review of
Books, 61(16):6466. Accessed 5 November 2014 from:
http://www.nybooks.com/articles/archives/2014/oct/23/what-scientists-really-do/.
[Naur, 2007] Naur, P. (2007). Computing versus human thinking. Communications of

the ACM, 50(1):8594.
[Nayebi, 2014] Nayebi, A. (2014). Practical intractability: A critique of the

hypercomputation movement. Minds and Machines, 24(3):275305.
[Neuberger, 2012] Neuberger, J. (2012). Benefits of a teacher and coach collaboration:

A case study. Journal of Mathematical Behavior, 31(2):290311.
[Newcombe et al., 2015] Newcombe, C., Rath, T., Zhang, F., Munteanu, B., Brooker,
M., and Deardeuff, M. (2015). How Amazon web services uses formal methods.
http://delivery.acm.org/10.1145/2700000/2699417/p66-newcombe.pdf
(accessed 18 April 2015)
and http://m.cacm.acm.org/magazines/2015/4/184701-how-amazon-web-services-
uses-formal-methods/fulltext
BIBLIOGRAPHY 809
[Newell, 1980] Newell, A. (1980). Physical symbol systems. Cognitive Science,

4:135183. Accessed 15 March 2014 from http://tinyurl.com/Newell1980.
[Newell, 1986] Newell, A. (1985-1986). Response: The models are broken, the
models are broken. University of Pittsburgh Law Review, 47:10231031.
[Newell et al., 1967] Newell, A., Perlis, A. J., and Simon, H. A. (1967). Computer
science. Science, 157(3795):13731374.
[Newell et al., 1958] Newell, A., Shaw, J., and Simon, H. A. (1958). Elements of a
theory of human problem solving. Psychological Review, 65(3):151166.
[Newell and Simon, 1976] Newell, A. and Simon, H. A. (1976). Computer science as
empirical inquiry symbols and search. Communications of the ACM, 19(3):113126.
[Nichols, 2011] Nichols, S. (2011). Experimental philosophy and the problem of free
will. Science, 331(6023):14011403.
[Nilsson, 1971] Nilsson, N. J. (1971). Problem-Solving Methods in Artificial
Intelligence. McGraw-Hill, New York.
[Nilsson, 1983] Nilsson, N. J. (1983). Artificial intelligence prepares for 2001. AI
Magazine, 4(4):714.
[OConnor and Robertson, 2005] OConnor, J. and Robertson, E. (October 2005). The
function concept. The MacTutor History of Mathematics Archive. http://www-
history.mcs.st-andrews.ac.uk/HistTopics/Functions.html.
[Ohlsson et al., 2015] Ohlsson, S., Sloan, R. H., Turan, G., and Urasky, A. (2015).
Measuring an artificial intelligence systems
performance on a verbal iq test for young children. http://arxiv.org/abs/1509.03390.
http://arxiv.org/pdf/1509.03390v1.pdf (accessed 11 October 2015).
[ONeill, 2013] ONeill, O. (2013). Interpreting the world, changing the world.
Philosophy Now, Issue 95:89. online at:
http://philosophynow.org/issues/95/Interpreting The World Changing The World
[ONeill, 2015] ONeill, S. (2015). The human race is a computer. New Scientist,
227(3029):2627. Interview with Cesar Hidalgo.
[Oommen and Rueda, 2005] Oommen, B. J. and Rueda, L. G. (2005). A formal
analysis of why heuristic functions work. Artificial Intelligence, 164(1-2):122.
[Oracle America, 2012] Oracle America, I. (2012).
Oracles april 12, 2012 brief regarding copyright issues, case no. 3:10-cv 10-03561-
wha, document 900. http://assets.sbnation.com/assets/1057275/Oracle s Brief.pdf.
Accessed 26 September 2014.
[Orr, 2013] Orr, H. A. (2013). Awaiting a new darwin. New York Review of Books,
60(2):2628.
810 BIBLIOGRAPHY
[Pandya, 2013] Pandya, H. (2013). Shakuntala devi, human computer who bested
the machines, dies at 83. New York Times. Accessed 26 February 2014 from:
http://www.nytimes.com/2013/04/24/world/asia/shakuntala-devi-human-computer-
dies-in-india-at-83.html.
[Parlante, 2005] Parlante, N. (2005). What is computer science? InroadsThe
SIGSCE Bulletin, 37(2):2425.
[Parnas, 1985] Parnas, D. L. (1985). Software aspects of strategic defense systems.
American Scientist, 73(5):432440. Reprinted in Communications of the ACM
28(12) (December 1985): 13261335.
[Parnas, 1998] Parnas, D. L. (1998). Software engineering programmes are not

computer science programs. Annals of Software Engineering, 6:1937. pre-print
online at:
http://www.cas.mcmaster.ca/serg/papers/crl361.pdf (accessed 10 June 2013; page
references are to the online pre-print); a paper with the same title (but American
spelling as program) appeared in IEEE Software 16(6) (1990): 1930.
[Parsons, 2015] Parsons, K. M. (2015). What is the public value of philosophy?

Huffington Post. http://www.huffingtonpost.com/keith-m-parsons/what-is-the-
public-value-of-philosophy b 7018022.html (accessed 13 May 2015).
[Pawley, 2009] Pawley, A. L. (2009). Universalized narratives: Patterns in how faculty
members define engineering. Journal of Engineering Education, 98(4):309319.
[Peacocke, 1995] Peacocke, C. (1995). Content, computation and externalism.

Philosophical Issues, 6:227264.
[Peacocke, 1999] Peacocke, C. (1999). Computation as involving content: A response
to Egan. Mind & Language, 14(2):195202.
[Peano, 1889] Peano, G. (1889). The principles of arithmetic, presented by a

new method. In van Heijenoort, J., editor, From Frege to Godel: A Source
Book in Mathematical Logic, 18791931, pages 8397. Harvard University Press,
Cambridge, MA.
[Pearl, 2000] Pearl, J. (2000). Causality: Models, Reasoning, and Inference.
Cambridge University Press, Cambridge, UK.
[Pelletier, 1999] Pelletier, F. J. (1999). A history of natural deduction and elementary
logic textbooks. History and Philosophy of Logic, 20:131. Accessed 17 February
2014 from: http://www.sfu.ca/jeffpell/papers/NDHistory.pdf.
[Pennisi, 2013] Pennisi, E. (2013). Nervous system may have evolved twice. Science,
339(6118):391.
[Perlis, 1962] Perlis, A. (1962). The computer in the university. In Greenberger, M.,
editor, Management and the Computer of the Future, pages 181217. MIT Press,
Cambridge, MA.
BIBLIOGRAPHY 811
[Perruchet and Vinter, 2002] Perruchet, P. and Vinter, A. (2002). The self-organizing
consciousness. Behavioral and Brain Sciences, 25(3):297388.
[Perry, 1981] Perry, William G., J. (1981). Cognitive and ethical growth: The making
of meaning. In Chickering, A. and Associates, editors, The Modern American
College, pages 76116. Jossey-Bass, San Francisco.
[Perry, 1970] Perry, Jr., W. G. (1970). Forms of Intellectual and Ethical Development
in the College Years: A Scheme. Holt, Rinehart and Winston, New York.
[Peschl and Scheutz, 2001a] Peschl, M. and Scheutz, M. (2001a). Some thoughts
on computation and simulation in cognitive science. In Proceedings of the 6th
Congress of the Austrian Philosophical Society, pages 534540. publisher unknown.
http://hrilab.tufts.edu/publications/scheutzpeschl00linz.pdf
(accessed 5 January 2015; citations to the online version).
[Peschl and Scheutz, 2001b] Peschl, M. F. and Scheutz, M. (2001b). Ex-
plicating the epistemological role of simulation in the development
of theories of cognition. In Proceedings of the 7th Colloquium on
Cognitive Science (ICCS-01), pages 274280. publisher unknown.
http://www.academia.edu/719102/Explicating the epistemological role of simulation in the development of theories of cognition
[Petersen, 2007] Petersen, S. (2007). The ethics of robot servitude. Journal of
Experimental and Theoretical Artificial Intelligence, 19(1):4354.
[Petroski, 2003] Petroski, H. (2003). Early education. American Scientist, 91:206

209. online at:
http://childrensengineering.org/resources/Petroski.pdf (accessed 21 October 2013).
[Petroski, 2005] Petroski, H. (2005). Technology and the humanities. American
Scientist, 93:304307.
[Petroski, 2006] Petroski, H. (2006). A great profession. American Scientist, 94:304

307. online at:
http://www.wsechicago.org/2006-07Petroski.pdf (accessed 21 October 2013).
[Petzold, 2008] Petzold, C. (2008). The Annotated Turing: A Guided Tour through
Alan Turings Historic Paper on Computability and the Turing Machine. Wiley,
Indianapolis.
[Picard, 1997] Picard, R. (1997). Affective Computing. MIT Press, Cambridge, MA.
[Piccinini, 2000] Piccinini, G. (2000). Turings rules for the imitation game. Minds
and Machines, 10(4):573582.
[Piccinini, 2003] Piccinini, G. (2003). Alan turing and the mathematical objection.
Minds and Machines, 13:2348.
812 BIBLIOGRAPHY
[Piccinini, 2004a] Piccinini, G. (2004a). The first computational theory of mind and
brain: A close look at McCulloch and Pittss Logical calculus of ideas immanent
in nervous activity. Synthese, 141:175215. For a reply, see [Aizawa, 2010].
[Piccinini, 2004b] Piccinini, G. (2004b). Functionalism, computationalism, and
mental contents. Canadian Journal of Philosophy, 34(3):375410. Accessed
22 April 2014 from:
http://www.umsl.edu/piccininig/Functionalism Computationalism and Mental Contents.pdf.
[Piccinini, 2005] Piccinini, G. (2005). Symbols, strings, and spikes: The empirical
refutation of computationalism. Accessed 24 March 2011 from
[http://www.umsl.edu/piccininig/Symbols
[Piccinini, 2006a] Piccinini, G. (2006a). Computation without representation.

Philosophical Studies, 137(2):204241. Accessed 29 April 2014 from:
http://www.umsl.edu/piccininig/Computation without Representation.pdf.
[Piccinini, 2006b] Piccinini, G. (2006b). Computational explanation in neu-
roscience. Synthese, 153(3):343353. Accessed 31 January 2014 from:
http://www.umsl.edu/piccininig/Computational%20Explanation%20in%20Neuroscience.pdf.
[Piccinini, 2007a] Piccinini, G. (2007a). Computational explanation and mechanistic
explanation of mind. In Marraffa, M., Caro, M. D., and Ferretti, F., editors,
Cartographies of the Mind: Philosophy and Psychology in Intersection, pages 23
36. Springer, Dordrecht, The Netherlands.
[Piccinini, 2007b] Piccinini, G. (2007b). Computational modelling vs. computational
explanation: Is everything a turing machine, and does it matter to the philosophy
of mind? Australasian Journal of Philosophy, 85(1):93115. Accessed 31 January
2014 from: http://www.umsl.edu/piccininig/Is Everything a TM.pdf.
[Piccinini, 2007c] Piccinini, G. (2007c). Computing mechanisms. Philosophy of

Science, 74(4):501526.
[Piccinini, 2008] Piccinini, G. (2008). Computers. Pacific Philosophical Quarterly,
89:3273.
[Piccinini, 2009] Piccinini, G. (2009). Computationalism in the philosophy of mind.
Philosophy Compass, 4(3):515532. 10.1111/j.1747-9991.2009.00215.x.
[Piccinini, 2010a] Piccinini, G. (2010a). The mind as neural software? understanding
functionalism, computationalism, and computational functionalism. Philosophy and
Phenomenological Research, 81(2):269311.
[Piccinini, 2010b] Piccinini, G. (2010b). The resilience of computionalism.

Philosophy of Science, 77(5):852861.
[Piccinini, 2011] Piccinini, G. (2011). The physical Church-Turing thesis: Modest or
bold? British Journal for the Philosophy of Science, 62:733769.
BIBLIOGRAPHY 813
[Piccinini, 2012] Piccinini, G. (2012). Computationalism. In Margolis, E., Samuels,

R., and Stich, S. P., editors, The Oxford Handbook of Philosophy of Cognitive
Science. Oxford University Press. 10.1093/oxfordhb/9780195309799.001.0001.
[Piccinini and Craver, 2011] Piccinini, G. and Craver, C. (2011). Integrating

psychology and neuroscience: Functional analyses as mechanism sketches.
Synthese, 183(3):283311. preprint online at:
http://philosophy.artsci.wustl.edu/files/philosophy/imce/integrating psychology and neuroscience functional analyses as mechanism sketches 0.pdf
[Piccinini and Scarantino, 2011] Piccinini, G. and Scarantino, A. (2011). Information

processing, computation, and cognition. Journal of Biological Physics, 37:138.
[Pincock, 2011] Pincock, C. (2011). Philosophy of mathematics. In Saatsi, J. and

French, S., editors, Companion to the Philosophy of Science, pages 314333.
Continuum. Pre-print accessed 31 March 2014 from:
http://pincock-yilmazer.com/chris/pincock%20philosophy%20of%20mathematics%20final%20draft.pdf.
[Pinker and Jackendoff, 2005] Pinker, S. and Jackendoff, R. (2005). The faculty of
language: Whats special about it? Cognition, 95:201236. Accessed 14 July 2014
from:
http://pinker.wjh.harvard.edu/articles/papers/2005 03 Pinker Jackendoff.pdf.
[Plato, 1961a] Plato (1961a). Phaedrus. In Hamilton, E. and Cairns, H., editors,
The Collected Dialogues of Plato, including the Letters. Princeton University Press,
Princeton, NJ.
[Plato, 1961b] Plato (1961b). Republic. In Hamilton, E. and Cairns, H., editors,
The Collected Dialogues of Plato, including the Letters, pages 575844. Princeton
[Polger, 2011] Polger, T. W. (2011). Are sensations still brain processes?

Philosophical Psychology, 24(1):121.
[Pollan, 2013] Pollan, M. (2013). The

intelligent plant. The New Yorker, pages 92105. Accessed 31 January 2014 from:
http://www.newyorker.com/reporting/2013/12/23/131223fa fact pollan.
[PolR, 2009] PolR (11 November 2009). An explanation of computation theory for
lawyers. http://www.groklaw.net/article.php?story=20091111151305785. also at:
http://www.groklaw.net/staticpages/index.php?page=20091110152507492
and at:
http://www.groklaw.net/pdf2/ComputationalTheoryforLawyers.pdf (accessed
16 December 2013).
[Popova, 2012] Popova, M. (2012). What is philosophy? an omnibus of definitions

from
prominent philosophers. http://www.brainpickings.org/index.php/2012/04/09/what-
is-philosophy/. (accessed 6 February 2013).
814 BIBLIOGRAPHY
[Popper, 1978]
Popper, K. (1978). Three worlds. http://tannerlectures.utah.edu/ documents/a-to-
z/p/popper80.pdf. Accessed 20 February 2015.
[Post, 1941] Post, E. L. (1941). Absolutely unsolvable problems and relatively

undecidable propositions: Account of an anticipation. In Davis, M., editor,
Solvability, Provability, Definability: The Collected Works of Emil L. Post, pages
375441. Birkhaauser, Boston.
[Post, 1943] Post, E. L. (1943). Formal reductions of the general combinatorial

decision problem. American Journal of Mathematics, 65:197215.
[Pour-El, 1974] Pour-El, M. B. (1974). Abstract computability and its relation to

the general purpose analog computer (some connections between logic, differential
equations and analog computers). Transactions of the American Mathematical
Society, 199:128.
[Powers, 1995] Powers, R. (1995). Galatea 2.2: A Novel. Far-

rar Straus & Giroux, New York. Excerpts are on at
http://www.amazon.com/gp/reader/0312423136/ref=sib dp pt/104-4726624-
5955928#reader-link
and https://play.google.com/store/books/details/Richard Powers Galatea 2 2?id=9xCw4QPsy88C.
[Prasse and Rittgen, 1998] Prasse, M. and Rittgen, P. (1998). Why churchs thesis
still holds. some notes on peter wegners tracts on interaction and computability.
The Computer Journal, 41(6):357362. Accessed 3 June 2014 from:
http://research.cs.queensu.ca/akl/cisc879/papers/SELECTED PAPERS FROM VARIOUS SOURCES/Prasse.pdf.
[Preston, 2006] Preston, B. (2006). The centralized control model of production. from
The Stuff of Life: Towards a Philosophy of Material Culture, Ch. 1.
[Preston, 2012] Preston, J. (2012). Paul feyerabend. In Zalta, E. N., editor, Stanford
Encyclopedia of
Philosophy. online. http://plato.stanford.edu/archives/win2012/entries/feyerabend/
[Proudfoot and Copeland, 2012] Proudfoot, D. and Copeland, B. J. (2012). Artificial

intelligence. In Margolis, E., Samuels, R., and Stich, S. P., editors, The Oxford
Handbook of Philosophy of Cognitive Science. Oxford University Press.
[Putnam, 1960] Putnam, H. (1960). Minds and machines. In Hook, S., editor,
Dimensions of Mind: A Symposium, pages 148179. New York University Press,
New York.
[Putnam, 1965] Putnam, H. (1965). Trial and error predicates and the solution
to a problem of mostowski. Journal of Symbolic Logic, 30(1):4957.
Accessed 5 February 2014 from: http://www.ninagierasimczuk.com/flt2013/wp-
content/uploads/2013/01/Putnam 1965.pdf.
BIBLIOGRAPHY 815
[Putnam, 1975] Putnam, H. (1975). The meaning of meaning. In Gunderson,

K., editor, Minnesota Studies in the Philosophy of Science, Vol. 7: Language,
Mind, and Knowledge, pages 131193. University of Minnesota Press, Minneapolis.
reprinted in Hilary Putnam, Mind, Language and Reality (Cambridge, UK:
Cambridge University Press): 215271; accessed 13 February 2014 from:
http://mcps.umn.edu/assets/pdf/7.3 Putnam.pdf.
[Putnam, 1994] Putnam, H. (1994). Sense, nonsense, and the senses: An inquiry into
the powers of the human mind. Journal of Philosophy, 91(9):445517.
[Putnam, 2001] Putnam, H. (2001). Skepticism, stroud and the contextuality

of knowledge. Philosophical Explorations: An International Journal for the
Philosophy of Mind and Action, 4(1):216.
[Pylyshyn, 1984] Pylyshyn, Z. W. (1984). Computation and Cognition: Towards a

Foundation for Cognitive Science. MIT Press, Cambridge, MA. Ch. 3 (The
Relevance of Computation), pp. 4886, esp. the section The Role of Computer
Implementation (pp. 7478).
[Pylyshyn, 1992] Pylyshyn, Z. W. (1992). Computers and the symbolization of

knowledge. In Morelli, R., Brown, W. M., Anselmi, D., Haberlandt, K., and Lloyd,
D., editors, Minds, Brains & Computers: Perspectives in Cognitive Science and
Artificial Intelligence, pages 8294. Ablex, Norwood, NJ. Preprint (1987) accessed
9 April 2014 from:
(page references are to the preprint).
[Qian and Winfree, 2011] Qian, L. and Winfree, E. (2011). Scaling up digital circuit
computation with dna strand displacement cascades. Science, 332:11961201. See
review in [Reif, 2011].
[Quillian, 1994] Quillian, M. R. (1994). A content-independent explanation of

sciences effectivenss. Philosohy of Science, 61(3):429448.
[Quine, 1948] Quine, W. v. O. (1948). On what there is. Review of Metaphysics,

2(5):2138. reprinted in W.V.O. Quine, From a Logical Point of View:
9 Logico-Philosophical Essays, Second Edition, revised (Cambridge, MA: Harvard
University Press, 1980): 119.
[Quine, 1951] Quine, W. v. O. (1951). Two dogmas of empiricism. Philosophical

Review, 60:2043.
[Quine, 1988] Quine, W. v. O. (1988). Letter from quine to chris hookway. In Hilary
Putnams Sardonic Comment blog. Accessed 14 July 2014 from:
http://putnamphil.blogspot.com/2014/07/a-letter-1988-from-quine-to-chris.html.
[Rado, 1962] Rado, T. (1962). On non-computable functions. The Bell

System Technical Journal, pages 877884. http://alcatel-lucent.com/bstj/vol41-
1962/articles/bstj41-3-877.pdf (accessed 16 December 2013).
816 BIBLIOGRAPHY
[Ralston, 1999] Ralston, A. (1999). Lets abolish pencil-and-paper arithmetic. Journal

of Computers in Mathematics and Science Teaching, 18(2):173194. Preprint
accessed 17 April 2014 from:
http://tonyralston.com/papers/abolpub.htm.
[Ramsey, 1929] Ramsey, F. P. (1929). Theories. In Mellor, D., editor, Foundations:
Essays in Philosophy, Logic, Mathematics and Economics; revised edition, pages
101125. Humanities Press, 1978, Atlantic Highlands, NJ.
[Randell, 1994] Randell, B. (1994). The origins of computer programming. IEEE
Annals of the History of Computing, 16(4):614. online at:
https://www.researchgate.net/publication/3330487 The origins of computer programming/
[Rapaport, 1978] Rapaport, W. J. (1978). Meinongian theories and a russellian
13 (1979): 125.
paradox. Nous, 12(2):153180. Errata, No(u)s
[Rapaport, 1982] Rapaport, W. J. (1982). Unsolvable problems and philosophical
progress. American Philosophical Quarterly, 19:289298.
[Rapaport, 1984] Rapaport, W. J. (1984). Critical thinking and cognitive development.

American Philosophical Associtation Newsletter on Pre-College Instruction in
Philosophy, 1:45. Reprinted in Proceedings and Addresses of the American
Philosophical Association 57(5) (May 1984): 610615;
http://www.cse.buffalo.edu/rapaport/Papers/rapaport84-perryAPA.pdf
[Rapaport, 1986a] Rapaport, W. J. (1986a). Logical foundations for belief
representation. Cognitive Science, 10:371422.
[Rapaport, 1986b] Rapaport, W. J. (1986b). Philosophy of artificial intelligence: A
course outline. Teaching Philosophy, 9(2):103120.
http://www.cse.buffalo.edu/rapaport/Papers/teachphil1986.pdf
[Rapaport, 1988] Rapaport, W. J. (1988). Syntactic semantics: Foundations of
computational natural-language understanding. In Fetzer, J. H., editor, Aspects of
Artificial Intelligence, pages 81131. Kluwer Academic Publishers, Dordrecht, The
Netherlands. Reprinted with numerous errors in Eric Dietrich (ed.) (1994), Thinking
Machines and Virtual Persons: Essays on the Intentionality of Machines (San Diego:
Academic Press): 225273.
[Rapaport, 1995] Rapaport, W. J. (1995). Understanding understanding: Syntactic
semantics and computational cognition. In Tomberlin, J. E., editor,
AI, Connectionism, and Philosophical Psychology, pages 4988. Ridgeview,
Atascadero, CA. Philosophical Perspectives, Vol. 9; reprinted in Toribio, Josefa,
& Clark, Andy (eds.) (1998), Language and Meaning in Cognitive Science:
Cognitive Issues and Semantic Theory, Artificial Intelligence and Cognitive Science:
Conceptual Issues, Vol. 4 (New York: Garland): 7388.
BIBLIOGRAPHY 817
[Rapaport, 1998] Rapaport, W. J. (1998). How minds can be computational systems.

Journal of Experimental and Theoretical Artificial Intelligence, 10:403419.
[Rapaport, 1999] Rapaport, W. J. (1999). Implementation is semantic interpretation.

The Monist, 82:109130.
[Rapaport, 2000a] Rapaport, W. J. (2000a). Cognitive science. In Ralston, A., Reilly,

E. D., and Hemmendinger, D., editors, Encyclopedia of Computer Science, 4th
edition, pages 227233. Groves Dictionaries, New York.
[Rapaport, 2000b] Rapaport, W. J. (2000b). How to pass a Turing test: Syntactic

semantics, natural-language understanding, and first-person cognition. Journal of
Logic, Language, and Information, 9(4):467490. Reprinted in [Moor, 2003, 161
184].
[Rapaport, 2003] Rapaport, W. J. (2003). What did you mean by that?

Misunderstanding, negotiation, and syntactic semantics. Minds and Machines,
13(3):397427.
[Rapaport, 2005a] Rapaport, W. J. (2005a). Castaneda, hector-neri. In Shook, J. R.,

editor, The Dictionary of Modern American Philosophers, 18601960, pages 452
412. Thoemmes Press, Bristol, UK.
[Rapaport, 2005b] Rapaport, W. J. (2005b). Implemention is semantic interpretation:

Further thoughts. Journal of Experimental and Theoretical Artificial Intelligence,
17(4):385417.
[Rapaport, 2006a] Rapaport, W. J. (2006a). How Helen Keller used syntactic

semantics to escape from a Chinese room. Minds and Machines, 16:381436. See
reply to comments, in [Rapaport, 2011b].
[Rapaport, 2006b] Rapaport, W. J. (2006b). The turing test. In Brown, K., editor,
Encyclopedia of Language and Linguistics, 2nd Edition, pages Vol. 13, pp. 151
159. Elsevier, Oxford. http://www.cse.buffalo.edu/ rapaport/Papers/rapaport06-
turingELL2.pdf
[Rapaport, 2007] Rapaport, W. J. (2007). Searle on brains as computers. American

Philosophical Association Newsletter on Philosophy and Computers, 6(2):49.
Accessed 31 January 2014 from:
http://c.ymcdn.com/sites/www.apaonline.org/resource/collection/EADE8D52-
8D02-4136-9A2A-729368501E43/v06n2Computers.pdf.
[Rapaport, 2009] Rapaport, W. J. (2009). Intensionality vs. intentionality.

http://www.cse.buffalo.edu/rapaport/intensional.html, accessed 30 September
2011.
[Rapaport, 2011a] Rapaport, W. J. (2011a). A triage theory of grading: The good, the
bad, and the middling. Teaching Philosophy, 34(4):347372.
818 BIBLIOGRAPHY
[Rapaport, 2011b] Rapaport, W. J. (2011b). Yes, she was! Reply to Fords Helen
Keller was never in a Chinese room. Minds and Machines, 21(1):317.
[Rapaport, 2012a] Rapaport, W. J. (2012a). Intensionality vs. intentionality.
http://www.cse.buffalo.edu/rapaport/intensional.html.
[Rapaport, 2012b] Rapaport, W. J. (2012b). Semiotic systems, computers, and the
mind: How cognition could be computing. International Journal of Signs and
Semiotic Systems, 2(1):3271.
[Rapaport, 2013] Rapaport, W. J. (2013). How to write. Accessed 4 February 2014
from: http://www.cse.buffalo.edu/rapaport/howtowrite.html.
[Rapaport, 2014] Rapaport, W. J. (2014). How to study: A brief guide.

http://www.cse.buffalo.edu/rapaport/howtostudy.html. Accessed 19 January 2014.
[Rapaport, 2015] Rapaport, W. J. (2015). On the relation of computing to the world.
Current draft in progress at
http://www.cse.buffalo.edu/rapaport/Papers/covey.pdf.
[Rapaport and Kibby, 2010] Rapaport, W. J. and Kibby, M. W. (2010). Contextual

vocabulary acquisition: From algorithm to curriculum.
http://www.cse.buffalo.edu/rapaport/CVA/reading4CgSJnl.pdf (accessed 13 April
2013).
[Rapaport et al., 1997] Rapaport, W. J., Shapiro, S. C., and Wiebe, J. M. (1997).
Quasi-indexicals and knowledge reports. Cognitive Science, 21:63107.
[Reese, 2014a] Reese, H. (18 February 2014a). The joy of teaching computer
science in the age of facebook. The Atlantic. Accessed 13 March 2014
from: http://www.theatlantic.com/education/archive/2014/02/the-joy-of-teaching-
computer-science-in-the-age-of-facebook/283879/.
[Reese, 2014b] Reese, H. (27 February 2014b). Why study philosophy? to

challenge your own point of view. The Atlantic. Accessed 12 March
2014 from: http://www.theatlantic.com/education/archive/2014/02/why-study-
philosophy-to-challenge-your-own-point-of-view/283954/.
[Reif, 2011] Reif, J. H. (2011). Scaling up dna computation. Science, 332:11561157.
Review of [Qian and Winfree, 2011].
[Rescorla, 2007] Rescorla, M. (2007). Churchs thesis and the conceptual analysis
of computability. Notre Dame Journal of Formal Logic, 48(2):253280. Preprint
(accessed 29 April 2014) at
http://www.philosophy.ucsb.edu/people/profiles/faculty/cvs/papers/church2.pdf.
[Rescorla, 2012a] Rescorla, M. (2012a). Are computational transitions sensitive

to semantics? Australian Journal of Philosophy, 90(4):703721. Preprint at
http://www.philosophy.ucsb.edu/docs/faculty/papers/formal.pdf
BIBLIOGRAPHY 819
[Rescorla, 2012b] Rescorla,

M. (2012b). How to integrate representation into computational modeling, and
why we should. Journal of Cognitive Science (South Korea), 13(1):138. http://j-
cs.org/gnuboard/bbs/board.php?bo table= vol013i1&wr id=1
[Rescorla, 2013] Rescorla, M. (2013). Against structuralist theories of computational
implementation. British Journal for the Philosophy of Science, 64(4):681707.
Preprint (accessed 31 October 2014) at
http://philosophy.ucsb.edu/docs/faculty/papers/against.pdf.
[Rescorla, 2014a] Rescorla, M. (2014a). The causal relevance of content to
computation. Philosophy and Phenomenological Research, 88(1):173208. Preprint
at
http://www.philosophy.ucsb.edu/people/profiles/faculty/cvs/papers/causalfinal.pdf
[Rescorla, 2014b] Rescorla, M. (2014b). A theory of computational implementation.
Synthese, 191:12771307. Preprint accessed 3 November 2014 from:
http://philosophy.ucsb.edu/docs/faculty/papers/implementationfinal.pdf.
[Rescorla, 2015] Rescorla, M. (2015). The computational theory of mind. In Zalta,
E. N., editor, The Stanford Encyclopedia
of Philosophy. http://plato.stanford.edu/archives/win2015/entries/computational-
mind/, winter 2015 edition.
[Richards, 1988] Richards, W. (1988). Natural Computation. MIT Press, Cambridge,
MA.
[Roberts, 1996] Roberts, E. (1996). Computer
science users guide. http://web.archive.org/web/19970609072402/http://www-
cse.stanford.edu/resources/users guide/1.html. (accessed 21 June 2013).
[Roberts, 2006] Roberts, E. S. (2006). The Art and Science of Java. Stanford
University,
Stanford, CA. http://people.reed.edu/jerry/121/materials/artsciencejava.pdf,
[Robertson, 2003] Robertson, D. S. (2003). Phase Change: The Computer Revolution
in Science and Mathematics. Oxford University Press.
[Robertson, 1979] Robertson, J. I. (1979). How to do arithmetic. American
Mathematical Monthly, 86:431439.
[Robinson, 1994] Robinson, J. (1994). Logic, computers, turing, and von
neumann. In Furukawa, K., Michie, D., and Muggleton, S., editors,
Machine Intelligence 13: Machine Intelligence and Inductive Learning,
pages 135. Clarendon Press, Oxford. Accessed 19 February 2014 from:
http://staffweb.worc.ac.uk/DrC/Courses%202013-
14/COMP3202/Reading%20Materials/ robinson94.pdf.
820 BIBLIOGRAPHY
[Rohlf, 2010] Rohlf, M. (2010). Immanuel kant.

In Zalta, E. N., editor, Stanford Encyclopedia of Philosophy (Fall 2010 Edition).
online. http://plato.stanford.edu/archives/fall2010/entries/kant/
[Rosch, 1978] Rosch, E. (1978). Principles of categorization. In Rosch, E. and Lloyd,

B. B., editors, Cognition and Categorization, pages 2748. Lawrence Erlbaum
Associates, Hillsdale, NJ.
[Rosch and Mervis, 1975] Rosch, E. and Mervis, C. B. (1975). Family resemblances:
Studies in the internal structure of categories. Cognitive Psychology, 7:573605.
[Rosenberg, 1994] Rosenberg, A. (1994). If economics isnt science, what is it? In

Hausman, D. M., editor, The Philosophy of Economics: An Anthology, Second
Edition, pages 376394. Cambridge University Press, New York. online at
http://philoscience.unibe.ch/documents/kursarchiv/SS07/Rosenberg1994.pdf
[Rosenblueth and Wiener, 1945] Rosenblueth, A. and Wiener, N. (1945). The role of
models in science. Philosophy of Science, 12:316321.
[Ross, 1974] Ross, D. (1974). Churchs thesis: What its difficulties are and are not.
Journal of Philosophy, 71(15):515525.
[Rosser, 1939] Rosser, J. B. (1939). An informal exposition of proofs of godels

theorems and churchs theorem. Journal of Symbolic Logic, 4(2):5360.
[Rosser, 1978] Rosser, J. B. (1978). Logic for Mathematicians: Second Edition. Dover
Publications, Mineola, NY. First edition (1953) online at:
https://archive.org/details/logicformathemat00ross
[Roth, 1983] Roth, P. (1983). Simulation. In Ralston, A. and Edwin D.Reilly, J.,
editors, Encyclopedia of Computer Science and Engineering, 2nd edition, pages
13271341. Van Nostrand Reinhold, New York.
[Royce, 1900] Royce, J. (1900). The World and the Individual. Macmillan, London.
[Rubinoff, 1953] Rubinoff, M. (1953). Analogue vs. digital computersa

comparison. Proceedings of the IRE, 41(10):12541262.
[Rupp, 2003] Rupp, N. A. (9 August 2003). Computer science is philosophy.

https://weblogs.java.net/blog/n alex/archive/2003/08/computer scienc.html.
[Russell, 1912] Russell, B. (1912). The Problems of Philosophy. Oxford University

Press (1959), London.
[Russell, 1917] Russell, B. (1917). Mysticism and Logic and Other Essays. George
Allen & Unwin, London. http://archive.org/details/mysticism00russuoft.
BIBLIOGRAPHY 821
[Russell, 1936] Russell, B. (1936). The limits of empiricism. Proceedings of the

Aristotelian Society, New Series, 36:131150.
[Russell, 1995] Russell, S. (1995). Rationality and intelligence. Artificial Intelligence,
94:5777. online at:
http://www.cs.berkeley.edu/russell/papers/aij-cnt.pdf
(accessed 18 September 2013).
[Russell and Norvig, 2003] Russell, S. J. and Norvig, P. (2003). Artificial Intelligence:
A Modern Approach; Second Edition. Pearson Education, Upper Saddle River, NJ.
[Russo, 1986] Russo, J. (1986). Saturns rings: What gms saturn project
is really about. Cornell University Labor Research Review, 1(9).
http://digitalcommons.ilr.cornell.edu/cgi/viewcontent.cgi?article=1084&context=lrr
(accessed 31 August).
[Sacks, 1999] Sacks, O. (1999). Migraine: Revised and Expanded. Random
House/Vintage Books, New York.
[Salter, 2010] Salter, J. (2010). The art of the ditch. New York Review of Books,
57(1). http://www.nybooks.com/articles/archives/2010/jan/14/the-art-of-the-ditch/
[Samuel, 1953] Samuel, A. L. (1953). Computing bit by bit, or digital computers made
easy. Proceedings of the IRE, 41(10):12231230.
[Samuelson, 1989] Samuelson, P. (1989). Why the look and feel of software user
interfaces should not be protected by copyright law. Communications of the ACM,
http://www.foo.be/andria/docs/p563-samuelson.pdf.
[Samuelson et al., 1994] Samuelson, P., Davis, R., Kapor, M. D., and Reichman,
J. (1994). A manifesto concerning the legal protection of computer programs.
Columbia Law Review, 94(8):23082431. Accessed 19 September 2014 from:
http://scholarship.law.duke.edu/cgi/viewcontent.cgi?article=1783&context=faculty scholarship.
From a special issue on the legal protection of computer programs; other articles
elaborate on, or reply to, Samuelson et al. 1 is a good overview; 2 (esp. 2.2) is
also good, as are 5 and the Conclusion section. A summary version appears as
[Davis et al., 1996].
[Samuleson, 1991] Samuleson, P. (1991). Digital media and the law. Communications
of the ACM, 34(10):2328.
[Sanford, 1967] Sanford, N. (1967). Where Colleges Fail. Jossey-Bass, San Francisco.
[Saussure, 1959] Saussure, F. d. (1959). Course in General Linguistics. Philosophical

Library, New York. Charles Bally, Albert Sechehaye, & Albert Reidlinger (eds.).
[Schagrin et al., 1985] Schagrin, M. L., Rapaport, W. J., and Dipert, R. R. (1985).
Logic: A Computer Approach. McGraw-Hill, New York.
822 BIBLIOGRAPHY
[Schank, 1983] Schank, R. C. (1983). The current state of ai: One mans opinion. AI
Magazine, 4(1):38.
[Scherlis and Scott, 1983] Scherlis, W. L. and Scott, D. S. (1983). First steps towards
inferential programming. In Mason, R., editor, Information Processing 83, pages
199212. Elsevier North-Holland. Technical-report version accessed 27 July 2014
from:
http://repository.cmu.edu/cgi/viewcontent.cgi?article=3542&context=compsci.
[Scheutz, 2001] Scheutz, M. (2001). Computational versus causal complexity. Minds
and Machines, 11:543566. Preprint accessed 23 October 2014 from:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.21.8293&represent=rep1&type=pdf.
[Scheutz, 2012] Scheutz, M. (2012). What it is not to implement a computation: A
critical analysis of chalmers
notion of implemention. Journal of Cognitive Science (South Korea), 13(1):75
106. http://j-cs.org/gnuboard/bbs/board.php?bo table= vol013i1&wr id=4
[Schwitzgebel, 2012] Schwitzgebel, E. (9 January 2012). For all x, theres philosophy
of x. http://schwitzsplinters.blogspot.com/2012/01/for-all-x-theres-philosophy-of-
x.html (accessed 12 February 2013).
[Schyns et al., 2008] Schyns, P. G., Goddelin, F., and Smith, M. L. (2008).
Information processing algorithms in the brain. Trends in Cognitive Sciences,
13(1):2026. Accessed 31 March 2014 from
http://www.mapageweb.umontreal.ca/gosselif/schynsetal TICS.pdf.
[Seabrook, 2007] Seabrook, J. (2007). Fragmentary knowledge. The New Yorker,
pages 94102.
[Searle, 1980] Searle, J. R. (1980). Minds, brains, and programs. Behavioral and
Brain Sciences, 3:417457.
[Searle, 1982] Searle, J. R. (1982). The myth of the computer. New York Review of
Books, pages 36. Cf. correspondence, same journal, 24 June 1982, pp. 5657.
[Searle, 1983] Searle, J. R. (1983). Intentionality: An Essay in the Philosophy of Mind.
Cambridge University Press, Cambridge, UK.
[Searle, 1990] Searle, J. R. (1990). Is the brain a digital computer? Proceedings and
Addresses of the American Philosophical Association, 64(3):2137.
[Searle, 1995] Searle, J. R. (1995). The Construction of Social Reality. Free Press,
New York.
[Sellars, 1962] Sellars, W. (1962). Philosophy and the scientific image of man. In
Colodny, R., editor, Frontiers of Science and Philosophy, pages 3578. University of
Pittsburgh Press, Pittsburgh. reprinted in Sellars, Wilfrid (1963), Science, Perception
and Reality (London: Routledge & Kegan Paul); online at:
http://www.ditext.com/sellars/psim.html (accessed 12 February 2013).
BIBLIOGRAPHY 823
[Shadmehr and Wise, 2005] Shadmehr, R. and Wise, S. (2005). The Computational
Neurobiology of Reaching and Pointing: A Foundation for Motor Learning. MIT
[Shagrir, 1999] Shagrir, O. (1999). What is computer science about? The Monist,
82(1):131149.
[Shagrir, 2006a] Shagrir, O. (2006a). Godel on turing on com-
putability. In Olszewski, A., Wo enski, J., and Janusz, R., edi-
tors, Churchs Thesis after 70 Years, pages 393419. Ontos-Verlag.
http://edelstein.huji.ac.il/staff/shagrir/papers/Goedel on Turing on Computability.pdf
(accessed 11 December 2013); page references are to the online version.
[Shagrir, 2006b] Shagrir, O. (2006b). Why we view the brain as a computer. Synthese,
153:393416. Preprint at
http://edelstein.huji.ac.il/staff/shagrir/papers/Why we view the brain as a computer.pdf
[Shagrir, 2010] Shagrir, O. (2010). Brains as analog-model computers. Studies in
History and Philosophy of Science, 41(3):271279.
[Shagrir, 2012] Shagrir, O. (2012). Can a brain
possess two minds? Journal of Cognitive Science (South Korea), 13(2):145165.
[Shannon, 1937] Shannon, C. E. (1937). A symbolic analysis of relay and switching
circuits. Technical report, MIT Department of Electrical Engineering, Cambridge,
MA. MS thesis, 1940; http://hdl.handle.net/1721.1/11173.
[Shannon, 1948] Shannon, C. E. (1948). A mathematical theory of communication.
The Bell System Technical
Journal, 27:379423, 623656. http://worrydream.com/refs/Shannon%20-
%20A%20Mathematical%20Theory%20of%20Communication.pdf
[Shannon, 1950] Shannon, C. E. (1950). Programming a computer for playing chess.
Philosophical Magazine, Ser. 7, 41(314). online at:
http://vision.unipv.it/IA1/ProgrammingaComputerforPlayingChess.pdf (accessed
5 March 2013).
[Shapiro and Benenson, 2006] Shapiro, E. and Benenson, Y. (2006). Bringing dna
computers to life. Scientific American, 294(5):4451. Accessed 24 March 2014
from
http://www.wisdom.weizmann.ac.il/udi/papers/ShapiroBenensonMay06.pdf.
[Shapiro, 1985] Shapiro, F. R. (1985). Debugging etymologies. New York Times. letter
to the editor; online at:
http://www.nytimes.com/1985/03/24/books/l-debugging-etymologies-114833.html
824 BIBLIOGRAPHY
[Shapiro, 2000a] Shapiro, F. R. (2000a). Origin of the term software: Evidence form
the jstor electronic journal archive. IEEE Annals of the History of Computing,
22(2):6971.
[Shapiro, 1993] Shapiro, S. (1993). Understanding churchs thesis, again. Acta

Analytica, 11:5977.
[Shapiro, 1996] Shapiro, S. (1996). Space, number and structure: A tale of two
debates. Philosophia Mathematica, 4(2):148173.
[Shapiro, 2000b] Shapiro, S. (2000b). Remarks on the development of computability.

History and Philosophy of Logic, 4(1):203220.
[Shapiro, 2009] Shapiro, S. (2009). We hold these truths to be self-evident: But what
do we mean
by that? Review of Symbolic Logic, 2(1):175207. Accessed 19 February 2014
from: http://www.pgrim.org/philosophersannual/29articles/shapirowehold.pdf.
[Shapiro, 1989] Shapiro, S. C. (1989). The cassie projects: An approach to

natural language competence. In Martins, J. and Morgado, E., editors, EPIA
89: 4th Portugese Conference on Artificial Intelligence Proceedings, pages
362380. Springer-Verlag Lecture Notes in Artificial Intelligence 390, Berlin.
http://www.cse.buffalo.edu/sneps/epia89.pdf (accessed 12 October 2015).
[Shapiro, 1992] Shapiro, S. C. (1992). Artificial intelligence. In Shapiro, S. C., editor,

Encyclopedia of Artificial Intelligence, 2nd edition, pages 5457. John Wiley &
Sons, New York.
[Shapiro, 1998] Shapiro,

S. C. (1998). Embodied cassie. In Cognitive Robotics: Papers from the 1998
AAAI Fall Symposium, pages 136143. AAAI Press, Menlo Park, CA. Technical
Report FS-98-02, http://www.cse.buffalo.edu/shapiro/Papers/embodiedcassie.pdf
[Shapiro, 2001] Shapiro, S. C. (2001). Computer science: The study of procedures.

Technical report, Department of Computer Science and Engineering, University at
Buffalo, Buffalo, NY. Accessed 30 September 2011 from:
http://www.cse.buffalo.edu/shapiro/Papers/whatiscs.pdf.
[Shapiro and Kwasny, 1975] Shapiro, S. C. and Kwasny, S. C. (1975). Interactive

consulting via natural language. Communications of the ACM, 18(8):459462.
http://www.cse.buffalo.edu/shapiro/Papers/shakwa75.pdf (accessed 28 August
2015).
[Shapiro and Rapaport, 1987] Shapiro, S. C. and Rapaport, W. J. (1987). SNePS

considered as a fully intensional propositional semantic network. In Cercone, N.
and McCalla, G., editors, The Knowledge Frontier: Essays in the Representation of
Knowledge, pages 262315. Springer-Verlag, New York.
BIBLIOGRAPHY 825
[Shapiro and Rapaport, 1991] Shapiro, S. C. and Rapaport, W. J. (1991). Models and
minds: Knowledge representation for natural-language competence. In Cummins,
R. and Pollock, J., editors, Philosophy and AI: Essays at the Interface, pages 215
259. MIT Press, Cambridge, MA.
[Shapiro and Rapaport, 1995] Shapiro, S. C. and Rapaport, W. J. (1995). An
introduction to a computational reader of narratives. In Duchan, J. F.,
Bruder, G. A., and Hewitt, L. E., editors, Deixis in Narrative: A Cognitive
Science Perspective, pages 79105. Lawrence Erlbaum Associates, Hillsdale, NJ.
http://www.cse.buffalo.edu/rapaport/Papers/shapiro.rapaport.95.pdf (accessed
12 October 2015).
[Shapiro et al., 1992] Shapiro, S. C., Srihari, S., and Jayaraman, B. (December 1992).
email discussion. http://www.cse.buffalo.edu/rapaport/scs.txt.
[Shelley, 1818] Shelley, M. W. (1818). Frankenstein; or, the Modern Prometheus.
online. http://www.literature.org/authors/shelley-mary/frankenstein/.
[Shepherdson and Sturgis, 1963] Shepherdson,
J. and Sturgis, H. (1963). Computability of recursive functions. Journal of the
ACM, 10(2):217255.
[Sheraton, 1981] Sheraton, M. (1981). The elusive art of writing precise recipes. New
York Times. http://www.nytimes.com/1981/05/02/style/de-gustibus-the-elusive-art-
of-writing-precise-recipes.html (accessed 13 November 2013).
[Sieg, 1994] Sieg, W. (1994). Mechanical procedures and mathematical experience.
In George, A., editor, Mathematics and Mind, pages 71117. Oxford University
Press, New York.
[Sieg, 2000] Sieg, W. (2000). Calculations by man and machine: Conceptual analysis.
Technical Report CMU-PHIL-104, Carnegie-Mellon University Department of
Philosophy, Pittsburgh, PA. http://repository.cmu.edu/philosophy/178.
[Sieg, 2006] Sieg, W. (2006). Godel on computability. Philosophia Mathematica,
14:189207. page references to preprint version, accessed 6 November 2012 from:
http://www.hss.cmu.edu/philosophy/sieg/A%20G del%20on%20Computability.pdf.
[Simon, 1962] Simon, H. A. (1962). The architecture of complexity. Proceedings of
the American Philosophical Society, 106(6):467482. reprinted as Ch. 8 of Simon
1996.
[Simon, 1996a] Simon, H. A. (1996a). Computational theories of cognition. In
ODonohue, W. and Kitchener, R. F., editors, The Philosophy of Psychology, pages
160172. SAGE Publications, London.
[Simon, 1996b] Simon, H. A. (1996b). The Sciences of the Artificial, Third Edition.
MIT Press, Cambridge, MA.
826 BIBLIOGRAPHY
[Simon and Newell, 1958] Simon, H. A. and Newell, A. (1958). Heuristic problem
solving: The next advance in operations research. Operations Research, 6(1):110.
[Simon and Newell, 1962] Simon, H. A. and Newell, A. (1962). Simulation of human
thinking. In Greenberger, M., editor, Computers and the World of the Future, pages
94114. MIT Press, Cambridge, MA.
[Skidelsky, 2014] Skidelsky, R. (2014). The programmed prospect before us. New
York Review of Books, 61(6):3537. Accessed 28 April 2014 from:
http://www.skidelskyr.com/site/article/the-programmed-prospect-before-us/.
[Skow, 2007] Skow, B. (2007). Are shapes intrinsic? Philosophical Studies, 133:111
130.
[Slagle, 1971] Slagle, J. R. (1971). Artificial Intelligence: The Heuristic Programming
Approach. McGraw-Hill, New York.
[Sloman, 1998] Sloman, A. (1998). Supervenience
and implementation: Virtual and physical machines.
http://www.cs.bham.ac.uk/research/projects/cogaff/Sloman.supervenience.and.implementation.pdf.
Accessed 21 October 2014.
[Sloman, 2002] Sloman, A. (2002). The irrelevance of Turing machines to AI. In
Scheutz, M., editor,
Computationalism: New Directions, pages 87127. MIT Press, Cambridge, MA.
http://www.cs.bham.ac.uk/research/projects/cogaff/sloman.turing.irrelevant.pdf
(accessed 21 February 2014). Page references are to the online preprint.
[Sloman, 1989] Sloman, A. (6 August 1989). Contribution to newsgroup
discussion of is there a definition of ai? Article 4702 of comp.ai.
http://www.cse.buffalo.edu/rapaport/defs.txt.
[Sloman, 2010] Sloman, A. (7 January 2010). Why symbol-grounding is both
impossible and unnecessary, and why symbol-tethering based on theory-tethering is
more powerful anyway. http://www.cs.bham.ac.uk/research/cogaff/talks/#models.
[Sloman and Croucher, 1981] Sloman, A. and Croucher, M. (1981). Why robots will
have emotions. In Proceedings of IJCAI 1981, page PGS. PUB, ADDR.
[Smith, 1985] Smith, B. C. (1985). Limits of correctness in computers. ACM SIGCAS
Computers and Society, 1415(14):1826. Also published as Technical Report
CSLI-85-36 (Stanford, CA: Center for the Study of Language & Information);
reprinted in Charles Dunlop & Rob Kling (eds.), Computerization and Controversy
(San Diego: Academic Press, 1991): 632646; reprinted in Timothy R. Colburn,
James H. Fetzer, & Terry L. Rankin (eds.), Program Verification: Fundamental
Issues in Computer Science (Dordrecht, Holland: Kluwer Academic Publishers,
1993): 275293.
[Smith, 1987] Smith, B. C. (1987). The correspondence continuum. Technical Report
CSLI-87-71, Center for the Study of Language & Information, Stanford, CA.
BIBLIOGRAPHY 827
[Smith, 1996] Smith, B. C. (1996). On the Origin of Objects. MIT Press, Cambridge,
MA.
[Smith, 2002] Smith, B. C. (2002). The foundations of computing. In Scheutz, M.,
editor, Computationalism: New Directions, pages 2358. MIT Press, Cambridge,
MA.
[Smith, 2010] Smith, P. (4 December
2010). Answer to what good is it to study philosophy?. AskPhilosophers.com.
http://www.askphilosophers.org/question/3710 (accessed 12 February 2013).
[Soare, 2009] Soare, R. I. (2009). Turing oracle machines, online computing, and
three displacements in
computability theory. Annals of Pure and Applied Logic, 160:368399. Preprint at
http://www.people.cs.uchicago.edu/soare/History/turing.pdf;
published version at
http://ac.els-cdn.com/S0168007209000128/1-s2.0-S0168007209000128-main.pdf? tid=8258a7e2-01ef-
11e4-9636-00000aab0f6b&acdnat=1404309072 f745d1632bb6fdd95f711397fda63ee2 (both
accessed 2 July 2014). A slightly different version appears as [Soare, 2013].
[Soare, 2012] Soare, R. I.
(2012). Formalism and intuition in computability. Philosophical Transactions of
the Royal Society A, 370:32773304. doi:10.1098/rsta.2011.0335.
[Soare, 2013] Soare, R. I. (2013). Interactive computing and relativized computability.
In Copeland, B. J., Posy, C. J., and Shagrir, O., editors, Computability: Turing,
Godel, Church, and Beyond, pages 203260. MIT Press, Cambridge, MA. A slightly
different version appeared as [Soare, 2009].
[Sprevak, 2010] Sprevak, M. (2010). Computation, individuation, and the received
view on representation. Studies in History and Philosophy of Science, 41(3):260
270.
[Spruit and Tamburrini, 1991] Spruit, L. and Tamburrini, G. (1991). Reasoning and
computation in leibniz. History and Philosophy of Logic, 12:114.
[Squires, 1970] Squires, R. (1970). On ones mind. Philosophical Quarterly,
20(81):347356.
[Staples, 2014] Staples, M. (2014). Critical rationalism and engineering: Ontology.
Synthese, 191(10):22552279. Preprint accessed 10 February 2015 from:
http://www.nicta.com.au/pub?doc=7397.
[Staples, 2015] Staples, M. (2015). Critical
rationalism and engineering: Methodology. Synthese, 192(1):337362. Preprint
at http://www.nicta.com.au/pub?doc=7747 (accessed 11 February 2015).
[Steed, 2013] Steed, S. (2013). Harnessing human intellect for computing. Computing
Research News, 25(2). Accessed 21 July 2014 from:
http://www.humancomputation.com/2013/CRN-Feb2013.pdf.
828 BIBLIOGRAPHY
[Sternberg, 1990] Sternberg, R. J. (1990). Metaphors of Mind: Conceptions of the

Nature of Intelligence. Cambridge University Press, New York.
[Stevens, 1996] Stevens, Phillip, J. (1996). Magic. In Levinson, D. and Ember, M.,
editors, Encyclopedia of Cultural Anthropology, pages 721726. Henry Holt, New
York.
[Stevens, 2001] Stevens, Phillip, J. (2001). Magical thinking in com-

plementary and alternative medicine. Skeptical Inquirer, 25(6):3237.
http://www.csicop.org/si/show/magical thinking in complementary and alternative medicine/,
[Stewart, 1994] Stewart, I. (1994). A subway named turing. Scientific American, pages
104, 106107.
[Stewart, 2000] Stewart, I. (2000). Impossibility theorems. Scientific American, pages

9899.
[Stewart, 2001] Stewart, I. (2001). Easter is a quasicrystal. Scientific American, pages

80, 8283. Accessed 11 December 2013 from:
http://www.whydomath.org/Reading Room Material/ian stewart/2000 03.html.
[Strasser and Antonelli, 2015] Strasser, C. and Antonelli, G. A. (2015). Non-

monotonic logic. In Zalta, E. N., editor, Stanford Encyclopedia of Philosophy
(Fall 2015 edition). Stanford University. http://plato.stanford.edu/entries/logic-
nonmonotonic/ (accessed 11 October 2015).
[Strawson, 2012] Strawson, G. (2012). Real naturalism. Proeedings and Addresses of

the American Philosophical Association, 86(2):125154.
[Strevens, 2013] Strevens, M. (2013). Looking into the black box. New York
Times Opinionator. http://opinionator.blogs.nytimes.com/2013/11/24/looking-into-
the-black-box (accessed 3 December 2013).
[Suber, 1988] Suber, P. (1988). What is software? Journal of Speculative

Philosophy, 2(2):89119. http://www.earlham.edu/peters/writing/software.htm
[Suber, 1997a] Suber, P. (1997a). Formal systems and machines: An isomorphism.

http://www.earlham.edu/peters/courses/logsys/machines.htm.
[Suber, 1997b] Suber, P. (1997b). Turing

machines ii. http://www.earlham.edu/peters/courses/logsys/turing2.htm (accessed
21 January 2014).
[Swade, 1993] Swade, D. D. (1993). Redeeming charles babbages mechanical

computer. Scientific American, pages 8691.
BIBLIOGRAPHY 829
[Tanenbaum, 2006] Tanenbaum, A. S. (2006). Structured Computer Organization,

Fifth Edition. Pearson Prentice Hall, Upper Saddle River, NJ. Accessed 24 July
2014 from:
http://e-book.az/download?id=821.
[Tedre, 2015] Tedre, M. (2015). The Science of Computing: Shaping a Discipline.

CRC Press/Taylor & Francis, Boca Raton, FL.
[Tedre and Sutinen, 2008] Tedre, M. and Sutinen, E. (2008). Three traditions of
computing: What educators should know. Computer Science Education, 18(3):153
170.
[Tenenbaum and Augenstein, 1981] Tenenbaum, A. M. and Augenstein, M. J. (1981).

Data Structures using Pascal. Prentice-Hall, Englewood Cliffs, NJ.
[Teuscher and Sipper, 2002] Teuscher, C. and Sipper, M. (2002). Hypercomputation:

Hype or computation? Communications of the ACM, 45(8):2330. Accessed
28 June 2014 from:
http://www.cs.bgu.ac.il/sipper/papabs/hypercomp.pdf.
[Thagard, 1978] Thagard, P. (1978). Why astrology is a pseudoscience. PSA:

Proceedings of the Biennial Meeting of the Philosophy of Science Association,
1:223234.
[Thagard, 1984] Thagard, P. (1984). Computer programs as psychological theories.

In Neumaier, O., editor, Mind, Language and Society, pages 7784. Conceptus-
Studien, Vienna.
[Thagard, 2006] Thagard, P. (2006). Hot Thought: Mechanisms and Applications of

Emotional Cognition. MIT Press, Cambridge, MA.
[Thagard, 2012] Thagard, P. (4 December 2012). Eleven dogmas of analytic

philosophy. Psychology Today. http://www.psychologytoday.com/blog/hot-
thought/201212/eleven-dogmas-analytic-philosophy (accessed 13 February 2013).
[The Economist, 2013] The Economist (2013). Unreliable research: Trouble at the
lab. The Economist. online at: http://www.economist.com/news/briefing/21588057-
scientists-think-science-self-correcting-alarming-degree-it-not-trouble (accessed
4 November 2013).
[Thomason, 2003] Thomason, R. H. (2003). Dynamic contextual intensional logic:

Logical foundations and an application. In Blackburn, P., editor, CONTEXT
2003: Lecture Notes in Artificial Intelligence 2680, pages 328341. Springer-
Verlag, Berlin. http://link.springer.com/chapter/10.1007/3-540-44958-2 26#page-1
[Tingley, 2013] Tingley, K. (2013). The body electric. The New Yorker, pages 7880,
82, 8686.
830 BIBLIOGRAPHY
[Toussaint, 1993] Toussaint, G. (1993). A new look at euclids second proposition.

The Mathematical Intelligencer,
15(3):1223. http://cgm.cs.mcgill.ca/godfried/publications/euclid.pdf (accessed
6 December 2013).
[Tucker et al., 2003] Tucker et al., A. (2003). A model curriculum for k12 computer
science: Final report of the acm k12 task force curriculum committee, second
edition. Computer Science Teachers Association and Association for Computing
Machinery. online at:
http://www.acm.org/education/education/curric vols/k12final1022.pdf (accessed
1 July 2013).
[Tukey, 1958] Tukey, J. W. (1958). The teaching of concrete mathematics. American

Mathematical Monthly, 65(1):19.
[Turing, 1936] Turing, A. M. (1936). On computable numbers, with an application to

the entscheidungsproblem. Proceedings of the London Mathematical Society, Ser. 2,
42:230265.
[Turing, 1937] Turing, A. M. (1937). Computability and -definability. Journal of

Symbolic Logic, 2(4):153163.
[Turing, 1938] Turing, A. M. (1938). On computable numbers, with an application

to the entscheidungsproblem: A
correction. Proceedings of the London Mathematical Society, Ser. 2, 43(1):544
546. http://cs.stanford.edu/people/trevisan/cs172/turing36.pdf (accessed 21 January
2014).
[Turing, 1948] Turing, A. M. (1948). Intelligent machinery. In Copeland, B.,

editor, The Essential Turing, pages 410432. Oxford University Press, Oxford.
Editorial commentary on pp. 395409. Typescript accessed 26 February 2014 from:
http://www.alanturing.net/intelligent machinery/.
[Turing, 1950] Turing, A. M. (1950). Computing machinery and intelligence. Mind,

59(236):433460.
[Turing, 1951] Turing, A. M. (1951). Intelligent machinery, a heretical theory.

Philosophia Mathematica, 4(3):256260. Alternate version accessed 21 May 2014
from:
http://commonsenseatheism.com/wp-content/uploads/2011/02/Turing-
Intelligent-Machinery-a-heretical-theory.pdf; typescripts and further bibliographic
information accessed 21 May 2014 from:
http://www.turing.org.uk/sources/biblio1.html.
[Tversky and Kahneman, 1974] Tversky, A. and Kahneman, D. (1974). Judgment

under uncertainty: Heuristics and biases. Science, 185(4157):11241131.
http://psiexp.ss.uci.edu/research/teaching/Tversky Kahneman 1974.pdf (accessed
20 April 2015).
BIBLIOGRAPHY 831
[Uebel, 2012] Uebel, T. (2012). Vienna circle. In Zalta,

E. N., editor, Stanford Encyclopedia of Philosophy (Summer 2012 Edition). online.
http://plato.stanford.edu/archives/sum2012/entries/vienna-circle/.
[Uglow, 2010] Uglow, J. (2010). The other side of science. New York Review of Books,
pages 3031, 34.
[Unger, 1979a] Unger, P. (1979a). I do not exist. In Macdonald, G. F., editor,

Perception and Identity. Cornell University Press, Ithaca, NY.
[Unger, 1979b] Unger, P. (1979b). Why there are no people. In ED, ed-
itor, Studies in Metaphysics (Midwest Studies in Philosophy, Vol. 4),
pages 177222. University of Minnesota Press, Minneapolis. (online at:
http://www.thatmarcusfamily.org/philosophy/Course Websites/Readings/Unger%20-
%20No%20People.pdf (accessed 6 February 2013)).
[Vahid, 2003] Vahid, F. (2003). The softening of hardware. [IEEE] Computer,

36(4):2734. Accessed 29 July 2014 from:
http://www.cs.ucr.edu/vahid/pubs/comp03 softerhw.pdf.
[van Leeuwen and Wiedermann, 2013] van Leeuwen, J. and Wiedermann, J. (2013).
The computational power of turings non-terminating circular a-machines. In
pages 8085. Elsevier, Amsterdam.
[Vardi, 2011] Vardi, M. Y. (2011). Computing for humans. Communications of the

ACM, 54(12):5.
[Vardi, 2012] Vardi, M. Y. (2012). What is an algorithm? Communications of the

ACM, 55(3):5. Accessed 26 February 2014 from:
http://cacm.acm.org/magazines/2012/3/146261-what-is-an-algorithm/fulltext.
[Veblen, 1908] Veblen, T. (1908). The evolution of the scientific point of view. The
University of California Chronicle: An Official Record, 10(4):395416. (online at:
http://archive.org/details/universitycalif08goog (accessed 6 February 2013).
[von Neumann, 1945] von

Neumann, J. (1945). First draft report on the edvac. IEEE Annals of the History of
Computing, 15(4):2775. edited by Michael D. Godfrey; accessed 30 October 2013
from http://virtualtravelog.net.s115267.gridserver.com/wp/wp-content/media/2003-
08-Th eFirstDraft.pdf; page references to online version.
[von Neumann, 1966] von Neumann, J. (1966). Theory of Self-Reproducing

Automata. University of Illinois Press, Urbana, IL. Arthur W. Burks (ed.).
[Wadler, 1997] Wadler, P. (1997). How to declare an imperative. ACM Computing

Surveys, 29(3):240263. Preprint accessed 26 May 2014 from:
https://wiki.ittc.ku.edu/lambda/images/3/3b/Wadler -
How to Declare an Imperative.pdf.
832 BIBLIOGRAPHY
[Wainer, 2007] Wainer, H. (2007). The most dangerous equation. American

Scientist, 95(3):249ff. http://www.americanscientist.org/issues/feature/2007/3/the-
most-dangerous-equation/ (accessed 20 April 2015).
[Wainer, 2012] Wainer, H. (2012). The survival of the fittists. American Scientist,
100:358361. online at:
http://www.americanscientist.org/issues/pub/the-survival-of-the-fittists/1 (accessed
23 July 2013).
[Wallich, 1997] Wallich, P. (1997). Cracking the u.s. code. Scientific American,
page 42.
[Walsh, 2014] Walsh, T. (2014). Candy crushs puzzling mathematics. American

Scientist, 102(6):430433. Accessed 17 October 2014 from:
http://www.americanscientist.org/issues/feature/2014/6/candy-crushs-puzzling-
mathematics/.
[Wang et al., 2013] Wang, Z., Busemeyer, J. R., Atmanspacher, H., and Pothos, E. M.,
editors (2013). Topics in Cognitive Science 5(4) (October): The Potential of Using
Quantum Theory to Build Models of Cognition. Cognitive Science Society, CITY.
[Wangsness and Franklin, 1966] Wangsness, T. and Franklin, J. (1966). algorithm

and formula. Communications of the ACM, 9(4):243.
[Wartofsky, 1966] Wartofsky, M. W. (1966). The model muddle: Proposals for an

immodest realism. In Wartofsky, M. W., editor, Models: Representation and the
Scientific Theory of Understanding, pages 111. D. Reidel, 1979, Dordrecht, The
Netherlands.
[Wartofsky, 1979] Wartofsky, M. W. (1979). Introduction. In Wartofsky, M. W., editor,

Models: Representation and the Scientific Theory of Understanding, pages xiii
xxvi. D. Reidel, Dordrecht, The Netherlands.
[Weatherson, 2012] Weatherson, B. (18 July

2012). What could leave philosohy? http://tar.weatherson.org/2012/07/18/what-
could-leave-philosophy/ (accessed 12 February 2013).
[Wegner, 1976] Wegner, P. (1976). Research paradigms in computer science. In ICSE

76 Proceedings of the 2nd International Conference on Software Engineering,
pages 322330. IEEE Computer Society Press, Los Alamitos, CA.
[Wegner, 1995] Wegner, P. (1995). Interaction as a basis for empirical computer

science. ACM Computing Surveys, 27(1):4548.
[Wegner, 1997] Wegner, P. (1997). Why interaction is more powerful than algorithms.
[Weinberg, 2002] Weinberg, S. (2002). Is the universe a computer? New York Review
of Books, 49(16).
BIBLIOGRAPHY 833
[Weizenbaum, 1966] Weizenbaum, J. (1966). Elizaa computer program for the

study of
natural language communication between man and machine. Communications of the
ACM, 9(1):3645. http://web.stanford.edu/class/linguist238/p36-weizenabaum.pdf
(accessed 28 August 2015); reprinted in Communications of the ACM 26(1)
(January 1983: 2328; be sure to read the correspondence between Weizenbaum
and Arbib on p. 28 of the reprint.
[Weizenbaum, 1967] Weizenbaum, J. (1967). Contextual understanding by

computers. Communications of the ACM, 10(8):474480.
[Weizenbaum, 1976] Weizenbaum, J. (1976). Computer Power and Human Reason.

W.H. Freeman, New York.
[Wheeler, 2013] Wheeler, G. (2013). Models, models, and models. Metaphilosophy,

44(3):293300. Preprint accessed 6 May 2014 from:
http://philsci-archive.pitt.edu/9500/1/Models.pdf.
[Wiebe and Rapaport, 1986] Wiebe, J. M. and Rapaport, W. J. (1986). Representing

de re and de dicto belief reports in discourse and narrative. Proceedings of the IEEE,
74:14051413.
[Wiener, 1964] Wiener, N. (1964). God and Golem, Inc.: A Comment

on Certain Points Where Cybernetics Impinges on Religion. MIT Press,
Cambridge, MA. http://www.scribd.com/doc/2962205/God-and-Golem-Inc-Wiener
or http://simson.net/ref/1963/God And Golem Inc.pdf.
[Wiesner, 1958] Wiesner, J. (1958). Communication sciences in a university

environment. IBM Journal of Research and Development, 2(4):268275.
[Wigner, 1960] Wigner, E. (1960). The unreasonable effectiveness of mathematics in

the natural sciences. Communications in Pure and Applied Mathematics, 13(1).
https://www.dartmouth.edu/matc/MathDrama/reading/Wigner.html,
http://www.maths.ed.ac.uk/aar/papers/wigner.pdf (both accessed 7 April 2015).
[Wilford, 2006] Wilford, J. N. (2006). Early astronomical computer found to be

technically complex. New York Times.
[Wilford, 2008] Wilford, J. N. (2008). A device that was high-tech in 100 b.c.
(discovering how greeks computed in 100 b.c.). New York Times, page A12.
[Williams, 1998] Williams, B. (1998). The end of explanation? The New York Review
of Books, 45(18):4044.
[Williamson, 2011] Williamson, T. (2011). What is naturalism? New York Times

Opinionator: The Stone. online at:
http://opinionator.blogs.nytimes.com/2011/09/04/what-is-naturalism/
834 BIBLIOGRAPHY
[Wilson and Papadopoulos, 2004] Wilson, D. G. and Papadopoulos, J. (2004).

Bicycling Science, Third Edition. MIT Press, Cambridge, MA.
[Wilson and Frenkel, 2013] Wilson, E. and Frenkel, E. (2013). Two views:
How much math do scientists need? Notices of the AMS, 60(7):837838.
http://dx.doi.org/10.1090/noti1032 (accessed 21 January 2014).
[Wing, 2006] Wing, J. M. (2006). Computational thinking. Communications of the
ACM, 49(3):3335.
[Winograd, 1983] Winograd, T. (1983). Language as a Cognitive Process; Vol. 1:
Syntax. Addison-Wesley, Reading, MA.
[Winsberg, 2001] Winsberg, E. (2001). Simulations, models, and theories: Complex
physical systems and their representations. Philosophy of Science, 68(3):S442
S454. Supplement: Proceedings of the 2000 Biennial Meeting of the Philosophy of
Science Association; Part I: Contributed Papers.
[Wirth, 1971] Wirth, N. (1971). Program development by stepwise refinement.
Communications of the ACM, 14(4):221227. Reprints online at:
http://oberoncore.ru/ media/library/wirth program development by stepwise refinement2.pdf,
http://sunnyday.mit.edu/16.355/wirth-refinement.html,
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.115.9376&rep=rep1&type=pdf
(all accessed 5 March 2015).
[Wittgenstein, 1921] Wittgenstein, L. (1921). Tractatus Logico-Philosophicus.
Humanities Press, New York. Second Edition, with corrections, 1972.
[Wittgenstein, 1958] Wittgenstein, L. (1958). Philosophical Investigations, Third
Edition. Macmillan, New York.
[Wittgenstein, 1980] Wittgenstein, L. (1980). Remarks on the Philosophy of
Psychology, Vol. I. University of Chicago Press, Chicago.
[Wolfram, 2002] Wolfram, S. (2002). A New Kind of Science. Wolfram Media.
[Worrall, 1989] Worrall, J. (1989). Structural realism: The best of both worlds?
Dialectica, 43(12):99124.
[Wulf, 1995] Wulf, W. (1995). Are we scientist or engineers? ACM Computing
Surveys, 27(1):5557.
[Young, 2008] Young, J. O. (2008). The coherence theory of truth. In
Zalta, E. N., editor, Stanford Encyclopedia of Philosophy (Fall 2008 Edition).
http://plato.stanford.edu/archives/fall2008/entries/truth-coherence/.
[Zeigler, 1976] Zeigler, B. P. (1976). Theory of Modeling and Simulation. Wiley
Interscience, New York.
[Zemanek, 1971] Zemanek, H. (1971). Was ist informatik? (what is informatics?).
Elektronische Rechenanlagen (Electronic Computing Systems), 13(4):157171.
BIBLIOGRAPHY 835
[Zhai and Kristensson, 2012] Zhai, S. and Kristensson, P. O. (2012). The word-
gesture keyboard: Reimagining keyboard interaction. Communications of the ACM,
55(9):91101.
[Zobrist, 2000] Zobrist, A. L. (2000). Computer games: Traditional. In Ralston, A.,
Reilly, E. D., and Hemmendinger, D., editors, Encyclopedia of Computer Science,
4th edition, pages 364368. Groves Dictionaries, New York.
[Zupko, 2011] Zupko, J. (2011). John Buridan, 7. In Zalta, E. N.,
editor, Stanford Encyclopedia of Philosophy (Fall 2011 Edition). online.
http://plato.stanford.edu/archives/fall2011/entries/buridan/.

Philosophy of Computer Science

Uploaded by

Copyright:

Available Formats

Philosophy of Computer Science

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Philosophy of Computer Science

Uploaded by

Copyright:

Available Formats

Philosophy of Computer Science

Department of Computer Science and Engineering,

November 23, 2015

1 What Is Philosophy of Computer Science? 21

3 What Is Computer Science? 63

3.3.1 Academic Motivations . . . . . . . . . . . . . . . . . . . . . 67

4 What Is Science? 149

5 What Is Engineering? 177

7 What Is an Algorithm? 223

CLARK & COWELL; SO COMPARE MY PRESENTATION WITH THEIRS!!

7.10 NOTES FOR NEXT DRAFT . . . . . . . . . . . . . . . . . . . . . . 257

8 Turings Analysis of Computation 289

10 What Is a Procedure? 411

11 What Is Hypercomputation? 449

11.6 NOTES FOR NEXT DRAFT . . . . . . . . . . . . . . . . . . . . . . 461

12 Software vs. Hardware 487

13 Copyright vs. Patent 509

14 What Is Implementation? 537

15 Are Programs Theories? 563

16 Can Computer Programs Be Verified? 587

17 How Do Programs Relate to the World? 613

18 Computer Ethics I: Decisions 629

19.5.4 A Better Way . . . . . . . . . . . . . . . . . . . . . . . . . . 665

20 Computer Ethics II: AI 677

21.8 NOTE FOR NEXT DRAFT . . . . . . . . . . . . . . . . . . . . . . 700

22 Syntax and Semantics 701

A Argument Analysis and Evaluation 705

B Position-Paper Assignments 719

C Term Paper Topics 763

D Final Exam 765

Version of 23 November 2015

This document is a continually-being-revised draft of a textbook on the philosophy

The course is described in:

A video of my Herbert Simon Keynote Address at NACAP-2006 describing the

http://www.hass.rpi.edu/streaming/conferences/cap2006/nacp 8 11 2006 9 1010.asx

And thanks for looking! (Changes are listed on next page.)

Changes since previous version (dated 20 November 2015):

Major changes to: Ch. 3, and Ch. 19,

1. Possibly divide into parts, each with chapters, e.g.:

2. Fix URLs, using BASHurl; see LaTeX Companion

3. Either include somewhere, or recommend to instructor to do, a slow reading of

put exercises at ends of approp chaps as per syllabus?

7. update section on reasoning to include sample valid, invalid args

8. add chap on syntactic semantics

9. write 2nd draft based on files

10. write a new closing chapter/article giving my phics.

13. Include all relevant emails from /web/faculty/rapaport/584/S07/EMAIL and

Version of 14 April 2015

Many people know about modern electronics in general and computers in

Figure 1.1: 1984

1.1 Recommended Readings:

1.2 What This Book Is About

(a) And, in particular: What is the philosophy of X? (where X = things like:

Then we will ask:

To answer this, we will need to consider a series of questions, each of which

Algorithms are said to be implemented in computer programs,

(a) What is AI?

(a) Should we trust decisions made by computers?

1.3 What This Book Is Not About

1.4 . . . Further Sources of Information