Black Boxes

Black Boxes
Black Boxes
How Science Turns Ignorance into Knowledge
M A R C O J. NAT HA N
1
3
Oxford University Press is a department of the University of Oxford. It furthers
the University’s objective of excellence in research, scholarship, and education
by publishing worldwide. Oxford is a registered trade mark of Oxford University
Press in the UK and certain other countries.
Published in the United States of America by Oxford University Press

198 Madison Avenue, New York, NY 10016, United States of America.
© Oxford University Press 2021
All rights reserved. No part of this publication may be reproduced, stored in

a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by license, or under terms agreed with the appropriate reproduction
rights organization. Inquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above.
You must not circulate this work in any other form

and you must impose this same condition on any acquirer.
Library of Congress Cataloging-in-Publication Data

Names: Nathan, Marco J., author.
Title: Black boxes : how science turns ignorance into knowledge / Marco J. Nathan.
Description: New York, NY : Oxford University Press, [2021] |
Includes bibliographical references and index.
Identifiers: LCCN 2021001998 (print) | LCCN 2021001999 (ebook) |
ISBN 9780190095482 (hardback) | ISBN 9780190095505 (epub)
Subjects: LCSH: Science—Philosophy. | Science—Methodology. |
Ignorance (Theory of knowledge)
Classification: LCC Q 175 . N38 2021 (print) | LCC Q 175 (ebook) | DDC 501—dc23
LC record available at https://lccn.loc.gov/2021001998
LC ebook record available at https://lccn.loc.gov/2021001999
DOI: 10.1093/oso/9780190095482.001.0001
1 3 5 7 9 8 6 4 2
Printed by Integrated Books International, United States of America
Dedicated, with love, to Jacob Aaron Lee Nathan.
Benvenuto al mondo.
Like all general statements, things are not as simple as I have written
them, but I am seeking to state a principle and refrain from listing
exceptions.
—Ernest Hemingway, Death in the Afternoon
Contents
Preface ix
1. Bricks and Boxes 1

1.1. The Wall 1
1.2. A Theory of Everything? 6
1.3. Pandora’s Gift to Science: The Black Box 12
1.4. Structure and Synopsis 16
1.5. Aim and Scope 21
2. Between Scylla and Charybdis 23
2.1. Introduction 23
2.2. The Rock Shoal and the Whirlpool 24
2.3. The Rise and Fall of Classical Reductionism 26
2.4. Antireductionism Strikes Back 31
2.5. The Reductionists Return 37
2.6. Scylla or Charybdis? 41
2.7. Why Not Have the Cake and Eat It Too? 47
3. Lessons from the History of Science 49
3.2. Darwin’s Quest for Inheritance and Variation 50
3.3. Mendel’s Missing Mechanisms: A Tale of Two Syntheses 55
3.4. Absent-Minded: Psychological Behaviorism 61
3.5. The Nature and Significance of Economics 67
3.6. The Many Faces of Black Boxes 77
4. Placeholders 82
4.2. Two Theses about Fitness 84
4.3. Are Dispositions Explanatory? 89
4.4. Placeholders in Scientific Theory and Practice 94
4.5. Two Types of Placeholders 105
5. Black-Boxing 101 109
5.2. Step One: The Framing Stage 111
5.3. Step Two: The Difference-Making Stage 118
viii Contents
5.4. Step Three: The Representation Stage 126

5.5. What Is a Black Box? A Three-Part Recipe 132
6. History of Science, Black-Boxing Style 136
6.2. Darwin’s Black Boxes 138
6.3. Mendel’s Black Boxes 144
6.4. Skinner’s Black Boxes 149
6.5. Friedman’s Black Boxes 155
6.6. Black Boxes in Science 160
7. Diet Mechanistic Philosophy 162
7.2. The New Mechanistic Philosophy 164
7.3. Three Steps: A Neo-Mechanistic Perspective 166
7.4. Mechanisms, Models, and Metaphysics 171
7.5. Mechanisms and Black Boxes 177
7.6. What’s in a Mechanism? 186
7.7. The Diet Recipe 189
8. Emergence Reframed 191
8.2. Emergence: Metaphysical or Epistemic? 193
8.3. Emergence in Action: Systems Neuroscience 197
8.4. Emergents and Black Boxes 202
8.5. Refinements and Implications 205
8.6. Adequacy Conditions and Advantages 209
8.7. Concluding Remarks 213
9. The Fuel of Scientific Progress 215
9.2. The Roots of Incommensurability 218
9.3. Referential Models and Incommensurability 225
9.4. Black Boxes and Reference Potential, Part I 232
9.5. Black Boxes and Reference Potential, Part II 238
9.6. Incommensurability, Progress, and Black Boxes 246
10. Sailing through the Strait 250
10.1. Back to Bricks and Boxes 250
10.2. The Quagmire 253
10.3. Rocking the Rock, Weathering the Whirlpool 256
10.4. A History of Black Boxes 263
10.5. The Devil in the Details? 275
References 281
Index 291
Preface
Every intellectual project has its “eureka!” moment. For this book, it happened
a few summers ago during a lonely walk on the beach of Marina di Campo,
on the beautiful island of Elba, off the Tuscan coast. I suddenly started seeing
a guiding thread, a leitmotiv, that connected much of my reflections on
the nature of science since I started working on these issues back in grad-
uate school. Simply put, it dawned upon me how I viewed most scientific
constructs as placeholders. Explanations, causal ascriptions, dispositions,
counterfactuals, emergents, and much else. They could all be viewed as
boxes, more or less opaque, standing in for more detailed descriptions. That
got me thinking about how to provide a more unified account that puts all
the tiles of the mosaic together. At the same time, I had realized that the very
concept of a black box, so frequently cited both in specialized and popular
literature, has been unduly neglected in philosophy and in the sciences alike.
This book is the result of my attempts to bring both insights together, in a
more or less systematic fashion.
The intellectual journey sparked by my preliminary reckoning on a sandy
beach has taken several years to complete. Along the way, I have been hon-
ored by the help and support of many friends and colleagues. Philip Kitcher
and Achille Varzi encouraged me to pursue this project from the get-go. Many
others provided constructive comments on various versions of the manu-
script. I am especially grateful to John Bickle, Giovanni Boniolo, Andrea
Borghini, Stefano Calboli, Guillermo Del Pinal, George DeMartino, Enzo
Fano, Tracy Mott, Emanuele Ratti, Sasha Reschechtko, Michael Strevens, and
Anubav Vasudevan for their insightful comments. A special thank you goes
to Mika Smith, Roscoe Hill, Mallory Hrehor, Naomi Reshotko, and, espe-
cially, Bill Anderson, all of whom struggled with me through several drafts
and minor tweaks, in my endless—futile, but no less noble—quest for clarity
and perspicuity.
Over the years, the University of Denver and, especially, the Department
of Philosophy have constantly provided a friendly, supporting, and stimu-
lating environment. Various drafts of the manuscript were presented as part
of my advanced seminar Ignorance and Knowledge in Contemporary Scientific
x Preface
Practice. I am indebted to all the students who shared thoughts, comments,

and frustrations with me—on this note, a shoutout to Blake Harris, Olivia
Noakes, and Jack Thomas. Two reading groups, one at the University of Milan
and one at the University of Denver, have been quite helpful. Bits and pieces
of this project have been presented, in various forms, at several institutions,
too many to adequately acknowledge. Audiences across the world have pro-
vided invaluable feedback.
I am very grateful to the production staff at Oxford University Press—
especially Peter Ohlin and his team for believing in the project from the very
beginning and their constant support, as well as Dorothy Bauhoff ’s careful
proofreading. Two anonymous reviewers provided precious feedback. Also,
a heartfelt “Grazie!” to Stefano Mannone of MadMinds for crafting the
images and bearing with all my nitpicky-ness.
Finally, none of this would have been possible without the unfaltering
help, patience, and support of my extended family. Lots of love to my wife
Heather, my parents Alida and Jacob, my sister Shirli, my brother David,
my two “adopted” brothers Federico and Matteo, and my nieces and
nephews: Martina, Virginia, Alexander, and Nicoló. Thanks for brightening
my days. While this book was slowly seeing the light, my own life has been
rocked and blessed by the birth of my son, Jacob. This work is dedicated to
him: welcome to the world!
1
Bricks and Boxes
Knowledge is a big subject.

Ignorance is bigger.
And it is more interesting.
—Stuart Firestein, Ignorance, p. 10
§1.1. The Wall
At the outset of a booklet, aptly entitled The Art of the Soluble, the eminent
biologist Sir Peter Medawar characterizes scientific inquiry in these terms:
Good scientists study the most important problems they think they can
solve. It is, after all, their professional business to solve problems, not
merely to grapple with them. The spectacle of a scientist locked in combat
with the forces of ignorance is not an inspiring one if, in the outcome, the
scientist is routed. (Medawar 1969, p. 11)
Many readers will find this depiction captivating, intuitive, perhaps even
self-evident. What is there to dispute? Is modern science not a spectacularly
successful attempt at solving problems and securing knowledge?
Yes, it is. Still, one could ask, what makes the spectacle of a scientist
locked in combat with the forces of ignorance so uninspiring? Why is it that
we seldom celebrate ignorance in science, no matter how enthralling, and
glorify success instead, regardless of how it is achieved? To be fair, we may
excuse ignorance and failure, when they have a plausible explanation. But ig-
norance is rarely—arguably never—a goal in and of itself. Has a Nobel Prize
ever been awarded for something that was not accomplished?
The key to answering these questions, and for understanding Medawar’s
aphorism, I maintain, is to be sought in the context of a long-standing
image of science that has, more or less explicitly, dominated the scene well
into the twentieth century. The goal of scientific inquiry, from this hallowed
Black Boxes. Marco J. Nathan, Oxford University Press. © Oxford University Press 2021.
DOI: 10.1093/oso/9780190095482.003.0001
2 Black Boxes
perspective, is to provide an accurate and complete description of the uni-

verse, or some specific portion of it. Doing science, the story goes, is anal-
ogous to erecting a wall. Or, borrowing a slightly artsier metaphor, it is like
tiling a mosaic. Walls are typically made out of stone or clay. Mosaics are
crafted by skillfully assembling tiles. The bricks of science, its fundamental
building blocks, are facts. Science is taken to advance through a slow, pains-
taking, constant discovery of new truths about the world surrounding us.
Intuitive as this may sound, we must be careful not to read too much
into the simile. First, there are clear discrepancies of sheer scale. The task
confronting the most creative builder or ambitious artist pales in compar-
ison to the gargantuan endeavor that the scientific community, taken in its
entirety, is striving to accomplish. Second, despite an impressive historical
record, the current outlook is, at best, very partial. What we have discovered
about the universe is dwarfed by what we do not yet know. Third, many of our
findings are bound to be inexact, at some level of description, and, when we
turn our attention to the most speculative theories, they may even be widely
off the mark. In short, the likening of buildings, works of art, and science
should be taken with more than just a grain of salt.
Still, the old image is suggestive, optimistically indicating that overall
trends and forecasts are positive. Despite a few inevitable hiccups, our grand
mosaic of the universe is slowly but surely getting bigger by the day. The goal
of the scientist is to identify a problem, solve it, and discover some new facts.
This, I maintain, is the backdrop to Medawar’s quote. Paraphrasing Pink
Floyd, all in all it’s just another brick in the wall.
Here is where things start to get more interesting. Over the last few
decades, a growing number of scientists, philosophers, historians, and
sociologists have argued that the depiction of science as progressive accu-
mulation of truth is, at best, a simplification. Consequently, Medawar’s sug-
gestive image of the scientist battling against the evil forces of ignorance has
gradually but inexorably fallen out of style. Much has been said about the
shortcomings of this old perspective, and some of these arguments will be
rehearsed in the ensuing chapters. The point, for the time being, is that very
few contemporary scholars, in either the sciences or the humanities, would
take it at face value any longer. The wall has slowly crumbled.
What is missing from the traditional picture of science as accumulating
truths and facts? Simply put, it is lopsided. Knowledge, broadly construed, is
a constituent—and an important one at that—of the scientific enterprise. Yet,
it is only one side of the coin. The other side involves a mixture of what we do
Bricks and Boxes 3
not know, what we cannot know, and what we got wrong. In a multifaceted
word, what is lacking from the old conception of science is the productive
role of ignorance. But what does it mean for ignorance to play a “productive”
role? Could ignorance, at least under some circumstances, be positive, per-
haps even preferable to the corresponding state of knowledge?
The inevitable presence of ignorance in scientific practice is neither novel
nor especially controversial. Generations of philosophers have extensively
discussed the nature and limits of human knowledge and their implications.
Nevertheless, ignorance was traditionally perceived as a hurdle to be over-
come. Over the last few decades it has turned into a springboard, and a more
constructive side of ignorance began to emerge. Allow me to elaborate.
In a recent booklet, Stuart Firestein (2012, p. 28), a prominent neurobiolo-
gist, remarks that “Science [ . . . ] produces ignorance, possibly at a faster rate
than it produces knowledge.” At first blush, this may sound like a pessimistic
reading of Medawar, where the spectacle of a scientist routed in combat by
forces of ignorance becomes uninspiring. Yet, as Firestein goes on to clarify,
this is not what he has in mind: “We now have an important insight. It is that
the problem of the unknowable, even the really unknowable, may not be a
serious obstacle. The unknowable may itself become a fact. It can serve as a
portal to deeper understanding. Most important, it certainly has not inter-
fered with the production of ignorance and therefore of the scientific pro-
gram. Rather, the very notions of incompleteness or uncertainty should be
taken as the herald of science” (2012, p. 44).
Analogous insights abound in psychology, where the study of cognitive
limitations has grown into a thriving research program.1 Philosophy, too, has
followed suit. Wimsatt (2007, p. 23) fuels his endeavor to “re-engineer phi-
losophy for limited beings” with the observation that “we can’t idealize devi-
ations and errors out of existence in our normative theories because they are
central to our methodology. We are error prone and error tolerant—errors are
unavoidable in the fabric of our lives, so we are well-adapted to living with
and learning from them. We learn more when things break down than when
they work right. Cognitively speaking, we metabolize mistakes!” In short, ig-
norance pervades our lives.
Time to connect a few dots. We began with Medawar’s insight that sci-
ence is in the puzzle-solving business. When one grapples with a problem
and ends up routed, it is a sign that something has gone south. All of this fits
1 See, for instance, Gigerenzer (2007) and Kahneman (2011).

4 Black Boxes
in well with the image of science as a cumulative, brick-by-brick endeavor,

which has dominated the landscape until and for the better part of the twen-
tieth century. Still, scholars are now well aware that ignorance is not always
or necessarily a red flag. The right kind of mistakes can be a portal to suc-
cess, to a deeper understanding of reality. As Wimsatt puts it, while some
errors are important, others are not. Thus, “what we really need to avoid is
not errors, but significant ones from which we can’t recover. Even significant
errors are okay as long as they’re easy to find” (2007, p. 24).
All of this well-known. But what does it mean to claim that ignorance
may become a portal to deeper understanding? Early on in the discussion,
Firestein acknowledges that his use of the word “ignorance” is intentionally
provocative. He subsequently clarifies his main point by drawing a distinc-
tion between two very different kinds of ignorance. As he elegantly puts it,
“One kind of ignorance is willful stupidity; worse than simple stupidity, it is
a callow indifference to facts of logic. It shows itself as a stubborn devotion
to uninformed opinions, ignoring (same root) contrary ideas, opinions, or
data. The ignorant are unaware, unenlightened, uninformed, and surpris-
ingly often occupy elected offices. We can all agree that none of this is good”
(2012, p. 6). Nevertheless, he continues, “there is another, less pejorative
sense of ignorance that describes a particular condition of knowledge: the
absence of fact, understanding, insight, or clarity about something. It is not
an individual lack of information but a communal gap in knowledge. It is a
case where data don’t exist, or more commonly, where the existing data don’t
make sense, don’t add up to a coherent explanation, cannot be used to make
a prediction or statement about some thing or event. This is knowledgeable
ignorance, perceptive ignorance, insightful ignorance. It leads us to frame
better questions, the first step to getting better answers. It is the most impor-
tant resource we scientists have, and using it correctly is the most important
thing a scientist does” (2012, pp. 6–7).
Needless to say, it is the latter notion of ignorance—knowledgeable, per-
ceptive, insightful ignorance—that I intend to further explore throughout
this book. Recognizing that failure can be important for the advancement
of science, and that not all ignorance is made equal, is a small step in a long,
tortuous stride. Once the spotlight is pointed in this direction, many provoc-
ative questions arise. What is productive ignorance? What is its role in sci-
entific research? Is it merely an indicator of where further work needs to be
done, or are there really scientific questions where ignorance may actually be
preferable to knowledge? Can ignorance teach us something, as opposed to
merely showing us what is missing? What makes some errors fatal and others
Bricks and Boxes 5
negligible or even useful? To put all of this in general terms, how does science
turn ignorance and failure into knowledge and success?
Much will be said, in the chapters to follow, about the nature of productive
ignorance, what distinguishes it from “a stubborn devotion to uninformed
opinions,” and how it is incorporated into scientific practice. Before doing so,
in the remainder of this section, I want to draw attention to a related issue,
concerning pedagogy: how science is taught to bright young minds. If, as
Firestein notes, ignorance is so paramount in science, why is its role not ex-
plicitly incorporated into the standard curriculum?
As noted, most scholars no longer take seriously the cumulative ideal of
science as the slow, painstaking accumulation of truths. Now, surely, there
is heated debate and disagreement on how inaccurate this picture really is.
More sympathetic readings consider it a benign caricature, promoting a
simple but effective depiction of a complex, multifaceted enterprise. Crankier
commentators dismiss it as an inadequate distortion that has overstayed its
welcome. Still, few, if any, take it at face value, for good reason.
This being said, this superseded image is still very much alive in some
circles. It is especially popular among the general public. Journalists,
politicians, entrepreneurs, and many other non-specialists explicitly en-
dorse and actively promote the vision of science as a gradual accumula-
tion of facts and truths. While some tenants of the ivory tower of academia
might take this as an opportunity to mock and chastise the incompetence
of the masses, we should staunchly resist this temptation. First, note, we
are talking about highly educated and influential portions of society. Lack
of schooling can hardly be the main culprit. Second, and even more im-
portant, it is not hard to figure out where the source of the misconception
might lie.
Textbooks, at all levels and intended for all audiences, present science as a
bunch of laws, theories, facts, and notions to be internalized uncritically. This
pernicious stereotype trickles down from schools and colleges to television
shows, newspapers, and magazines, eventually reaching the general public.
Now, surely, some promising young scholars will go on to attend graduate
schools and pursue research-oriented careers: academic, clinical, govern-
mental, and the like. They will soon learn, often the hard way, that actual
scientific practice is very different—and way more interesting!—than what is
crystallized in journals, books, and articles. Yet, this is a tiny fraction of the
overall population. Most of us are only exposed to science in grade school or
college, where the old view still dominates. By the time one leaves the class-
room to walk into an office, the damage is typically done.
6 Black Boxes
Where am I going with this? The bottom line is simple. The textbooks by
which students learn science, which perpetuate the cumulative approach, are
written by those same specialists who, in their research, eschew that brick-
by-brick analogy. Why are experts mischaracterizing their own disciplines,
promoting an old image that they themselves no longer endorse?
This book addresses this old question, which has troubled philosophers of
science, at least since the debates of the 1970s in the wake of Kuhn’s Structure.
My answer can be broken down to a pars destruens and a pars construens.
Beginning with the negative claim, textbooks do not adopt the cumulative
model with the intention to deceive. The reason is that no viable alternative
is available. The current scene is dominated by two competing models of
science, neither of which supplants the “brick-by-brick” ideal. There must
be more effective ways of popularizing research, exposing new generations
to the fascinating world of science. The positive portion of my argument
involves a constructive proposal. The next two sections present these two
theses in embryonic form. Details will emerge in the ensuing chapters.
§1.2. A Theory of Everything?
The previous section concluded by posing a puzzle. Why do textbooks pre-

sent science as an accumulation of truths and facts, given that their authors
know perfectly well how inaccurate this is? The answer, I maintain, is that no
better alternative is available. We currently have two competing models of
science: reductionism and antireductionism. Neither provides an accurate
depiction of the productive interaction between knowledge and ignorance,
supplanting the old image of the wall. This section introduces the status quo,
the two philosophical models presently dominating the scene.
Before getting started, two quick clarifications. First, these prelimi-
nary paragraphs merely set the stage for further discussion to be devel-
oped throughout the book. Skeptical readers unsatisfied with my cursory
remarks are invited to hold on tight. Second, in claiming that “philosophical”
models guide the presentation in textbooks and other popular venues, I am
not suggesting that scientific pedagogy is directly responding to the work
of philosophers—wishful thinking, one could say. Rather, my claim is that
these models, developed systematically within the philosophy of science, re-
flect the implicit attitude underlying much empirical practice. With these
qualifications in mind, we can finally get down to business.
Bricks and Boxes 7
Classic and contemporary philosophy of science has been molded by a

long-standing clash between two polarizing metaphysical stances. Allow me
to illustrate them by introducing a colorful image that will recur throughout
our discussion. Assume that the natural world can be partitioned into dis-
tinct, non-overlapping levels, arranged in terms of constitution, with smaller
constituents at the bottom and larger-scale entities arranged toward the top.
From this standpoint, these levels will correspond to traditional subdivisions
between fields or inquiries. Intuitively, physics, broadly construed, will be
at the very bottom. At a slightly coarser scale, we find entities described by
biology, followed by neuropsychology. Toward the top we will find the mac-
roscopic structures postulated and studied by economics and other social sci-
ences. It is customary to depict this layering of reality as a wedding cake. For
this reason, I shall refer to it as the “wedding cake model” (Figure 1.1). While
such partition will be further developed and refined, it is important to clarify
from the get-go that this representation is admittedly oversimplified and in-
complete. It is incomplete because there are numerous fields and subfields of
science that are not included, such as sociology and anthropology. It is over-
simplified because any student of science worth their salt is perfectly aware
that this partition does not capture complex, nuanced relations between dis-
ciplines. To wit, I am here clashing together psychology and neuroscience
ECON O MIC S
NEUROPSYCH
BIOLOGY
PHYSICS
Figure 1.1. The “wedding cake” layering of reality.

8 Black Boxes
under the moniker “neuropsychology,” knowing perfectly well that their

ontologies and methodologies diverge in important respects. Similarly, parts
of biology overlap with psychology, neuroscience, and the social sciences,
spawning hybrid, interdisciplinary approaches such as neuroethics, evolu-
tionary psychology, and biolinguistics. Finally, one could contest my place-
ment of physics at the very bottom, since a prominent part of this science,
astrophysics, studies some of the largest systems in the universe.
With these limitations well in mind, I present this idealized model of
reality to illustrate our two metaphysical stances. On the one hand, reduc-
tionism contends that all special sciences boil down to fundamental physics,
in the sense that “higher levels” effectively reduce to the bottom level.2 On
the other hand, antireductionism maintains that “higher” layers are ontolog-
ically or epistemically autonomous from “lower” ones.3 How should these
claims be understood? What is the crux of the disagreement? A more precise
articulation of the debate will be set aside until Chapter 2. For the time being,
my aim is a general reconstruction of the main point of contention between
reductionists and their foes.
Allow me to kick things off by introducing a simple thought experiment.
Imagine that, at some time in the future, physics—or more accurately, the
descendants of our current physical theories—will advance to the point of
providing an exhaustive description of the bedrock structure of the entire
cosmos, together with a complete specification of all the fundamental laws
of nature.
At first, this picture will be easier to visualize if we depict the universe as an
enormous container (“absolute space”) filled with an astronomical number
of indivisible particles. Further, suppose that these atomic elements interact
with each other in a fully deterministic fashion along a temporal dimension
(“absolute time”) in a reversible, that is, time-symmetric manner.
This scenario roughly corresponds to the depiction of the cosmos pro-
vided by Newton. Note that it is completely deterministic. This means that
any complete future state of the universe may, in principle, be predicted
2 The expression “special sciences” refers to all branches of science, with the exception of funda-
mental (particle) physics. This moniker is somewhat misleading, as there is nothing inherently “spe-
cial” about these disciplines, aside from having been relatively underexplored in philosophy. Yet, this
label has become commonplace and I shall stick to it.
3 To be clear, any characterization of levels as “coarse-grained” vs. “fine-grained,” “higher-level” vs.
“lower-level,” or “micro” vs. “macro,” should be understood as relativized to a specific choice of expla-
nandum. From this standpoint, the same explanatory level Ln can be “micro” or “lower-level” relative
to a coarser description Ln+1, and “macro” or “higher” relative to a finer-grained depiction Ln−1.
Bricks and Boxes 9
precisely from any comprehensive description of a past state, in conjunction

with the relevant laws of nature. Analogously, past states can be retrodicted
from future ones.
One final flight of fancy. Suppose that, down the line, researchers will also
find a way to achieve the astounding computing power required for these
derivations. What this means is that, given an exhaustive specification of the
location and momentum of every bedrock particle in the universe, any past
or future state can be determined with absolute certainty.
This thought experiment was popularized by the physicist and mathema-
tician Pierre Simon de Laplace. For this reason, I shall follow an established
tradition and refer to a creature with such stupendous predictive and retro-
dictive capacities as a “Laplacian Demon.”4
As some readers will surely realize, making Laplace’s insight cohere with
contemporary physics requires some fancy footwork. First and foremost, we
need a characterization of the universe that avoids any reference to the super-
seded concepts of absolute space and absolute time, and which reframes
these notions in relativistic terms. Second, a Laplacian Demon worth their
salt needs to be at least compatible with the fundamental laws of nature being
indeterministic. This is a possibility left open by current quantum mechanics
but, it should be noted, it is fundamentally at odds with one of Laplace’s
core tenets, namely, determinism. On this interpretation, the Demon will,
at best, be able to assign to every past and future state a precise probability
that falls short of unity, that is, certainty. Finally, a more realistic characteri-
zation of Laplace’s thought experiment requires addressing thermodynamic
and quantum irreversibility as well as in-principle unpredictable behaviors
described by chaos theory.
Be that as it may, suppose that future physics follows this path, devel-
oping into a powerful framework capable of describing, down to its most
basic constituents, every corner of the physical universe that exists, has
existed, and will ever exist. These considerations raise a host of philosophical
questions.5 Under these hypothetical circumstances, what would happen to
science as we know it today? Could physics explain every event? Or, more
modestly, could it explain any event that currently falls under the domain
4 For a discussion of the origins and foundations of Laplacian determinism, see van Strien (2014).
5 “Philosophical” discussions of Laplacian Demons and the future of physics have been under-
taken by prominent philosophers and scientists such as Dennett (1987); Putnam (1990); Weinberg
(1992); Mayr (2004); Chalmers (2012); Nagel (2012); and Burge (2013).
10 Black Boxes
of some branch of science? Would we still need the special sciences? Would
physics replace them and truly become a scientific theory of everything?
There are two broad families of responses to these questions. The first,
which may be dubbed reductionism, answers our quandaries in the posi-
tive. Reductionism comes in various degrees of strength. Any reductionist
worth their salt is perfectly aware that current physics is still eons away from
achieving the status of a bona fide accurate and exhaustive description of the
universe. And even if our best available candidates for fundamental laws of
nature happened to be exact, the computing power required to rephrase rel-
atively simple higher-level events in physical terms remains out of reach, at
least for now. This is to say that physics, as we presently have it, is not yet an
overarching theory of everything. Still, the more radical reductionists claim,
physics will eventually explain all events in the universe, and contemporary
theories have already put us on the right track. More modest reductionists
make further concessions. Perhaps physics will never actually develop to the
point of becoming the utmost theory of reality. Even if it did, gaining the
computing power to completely dispose of all non-physical explanations
may remain a chimera. Hence, the special sciences will always be required
in practice. Nevertheless, in principle, physics could do without them. In this
sense, the special sciences are nothing but convenient, but potentially dis-
posable, scaffoldings.
The second family of responses, antireductionism, provides diametrically
opposite answers to our questions concerning the future of physics. Like its
reductionist counterpart, antireductionism comes in degrees. More radical
versions contend that, because of the fundamental disunity or heterogeneity
of the universe, a physical theory of everything is a metaphysical impossi-
bility. Even the most grandiose, all-encompassing, and successful physical
theories could not explain all that goes on in the material universe because
many—or, some would say, most—events covered by the special sciences fall
outside of the domain of physics. Less uncompromising antireductionists
may make some concessions toward modest forms of reductionism. Perhaps
physics could, in principle, explain every scientific event. Still, the special
sciences are nevertheless not disposable. This is because they provide objec-
tively “better” explanations of higher-level happenings. In short, the antire-
ductionist says, the success of physics is no threat for higher-level theories.
Special sciences are not going anywhere, now or ever.
In sum, the debate between reductionism and antireductionism in general
philosophy of science boils down to the prospects of developing a physical
Bricks and Boxes 11
theory of everything, a grand characterization of the world that, in principle,

could describe, predict, and explain any and every scientific event. These
metaphysical considerations carry methodological implications. Are dis-
ciplines other than fundamental physics mere scaffoldings awaiting replace-
ment? Or are special sciences a necessary, ineliminable component of any
comprehensive description and explanation of the universe?
These thought-provoking questions have inspired much discussion. The
divide between reductionism and antireductionism has dominated classic
and contemporary philosophy of science. The widespread propensity to
privilege fundamental physics over other sciences digs its roots deep into
the history of Naturwissenschaft.6 Logical positivism, the founding move-
ment of the philosophy of science as it is currently understood, assumed that
the physical and mathematical sciences set both the agenda and the tone
of the conversation. This attitude that physics is the paradigm of true sci-
ence and all other fields should sooner or later follow suit, still prominent
in some circles, has attracted criticism among scientists, historians, and
philosophers, spawning healthy debate. However, after decades of discus-
sion, these positions have hardened, stifling productive conversation. The
topic of reduction has attracted a disproportional amount of attention in phi-
losophy, at the expense of other equally pressing and important issues. Let
me motivate this claim, which lies at the core of my argument.7
First and foremost, debating whether all scientific inquiries can be
addressed in terms of physics has a major drawback. It encourages laying all
our eggs in one basket—the nature, limits, and boundaries of ideal, future
science—setting aside questions pertaining to current inquiries. But how do
we adjudicate, in the present, the prospects of research yet to come? In this
respect, general philosophy of science has much to learn from contemporary
philosophies of particular disciplines, which understood decades ago the im-
portance of focusing on the here-and-now of empirical practice.
Second, and more relevant to our concerns, reductionism and antireduc-
tionism, as traditionally conceived, leave us wanting. One problem is that
there is little consensus on how to characterize “reduction” and “autonomy,”
often leaving disputants talking past each other. In addition, neither stance is
6 For an excellent discussion of the rise of scientific philosophy, see Friedman (2001).
7 Similar conclusions have been reached, via a different route, by Wimsatt (2007), and developed
by the “new wave of mechanistic philosophy,” presented and examined in Chapter 7. Relatedly, Gillett
(2016) has noted some discrepancy between the models of reduction and emergence developed in
philosophy versus the ones adopted in scientific practice.
12 Black Boxes
able to successfully capture the interplay between knowledge and ignorance

in contemporary science. Reductionist metaphysics goes together well with
the “brick-by-brick” epistemology, which views the advancement of science
as breaking down complex, higher-level systems into smaller blocks. This
leaves little room for productive ignorance, that is, the idea that ignorance
itself may, at times, be preferable to the corresponding state of knowledge.
Antireductionism, in contrast, legitimizes the role of ignorance by stressing
the autonomy of the macrocosm. Still, it fails to convincingly explain why
higher-level explanations can be objectively superior to lower-level ones.
These two perspectives need to be combined, not pitched against each other.
Throughout the book, I shall argue that the reductionism vs. antireduc-
tionism opposition, as traditionally conceived, confronts us with a false di-
lemma. Paraphrasing Lakatos, despite its long-standing history and renewed
popularity, this debate has now turned into a regressive research program.
Both parties often talk past each other, rehashing old “this was never my point
to begin with” claims. In the meantime, substantive historical, scientific, and
philosophical questions accumulate, awaiting examination. After decades
of discussion, it is finally time to move on. What we need is a framework
that can bring together both aspects of science—autonomy and reduction—
without pitching them against each other. What is needed, in short, is a dif-
ferent, alternative model of science. What could this be?
§1.3. Pandora’s Gift to Science: The Black Box
Section 1.2 introduced my pars destruens: the debate between reductionism

and antireductionism fails to provide an adequate image of the dynamic
interplay between scientific knowledge and productive ignorance. If the
current path is not the way to go, in what direction should we point the
discussion?
Answering this question is the chief goal of the constructive portion of
my argument. The chapters that follow sketch an account of scientific ex-
planation that is neither reductionist nor antireductionist. Or, perhaps, it is
both, in the sense that it combines core aspects from each perspective. The
proposed shift brings attention to a prominent construct—the black box—
which underlies a well-oiled technique for incorporating a productive role
of ignorance and failure into the acquisition of empirical knowledge. While
Bricks and Boxes 13
a comprehensive analysis will be undertaken throughout the monograph, let

me briefly introduce you to this important albeit neglected concept.
Most readers will likely have some pre-theoretical, intuitive sense of what a
black box is and may even have candidate examples in mind. If your acquaint-
ance with electronics is anything like mine, your smartphone constitutes a
perfectly good illustration, as it surely does for me. I am able to use my phone
reasonably well. I can easily make calls, check my email, and look up driving
directions. I am aware that, by opening a specific app, the screen will display
my boarding pass, allowing me to board my flight. I obviously know that, if
I do not charge the battery every other day, the darn thing will eventually run
out of power. For all intents and purposes, I am a typical user. The point is
that, like most customers, I have no clear conception of what happens inside
the phone itself, as witnessed by the observation that, if something were to
break or otherwise stop working, I would have to take it to a specialized store
for repair. In brief, I have a fairly good grasp of the functional organization of
my phone—systems of inputs and outputs that govern standard usage—and
even a vague sense of the algorithms at play. But the inner workings that un-
derlie this functionality and make it possible are a mystery to me. I am per-
fectly aware that they must be there. But, being no technician, I have no clue
as to what exactly they are and how they work. In this sense, a smartphone is
a black box to me, and I am confident that many others are in the same boat.
Do not let the mundane nature of the example deceive you. There is a long-
standing historical tradition referring to “known unknowns,” going back, at
least, to the medieval debate on insolubilia and, more recently, Emil du Bois-
Reymond’s ignoramus et ignorabimus. In contemporary settings, references
to black boxes can be found throughout the specialized literature across a
variety of fields. In contemporary philosophy, the black box par excellence
is the human mind itself. As Daniel Dennett (1991, p. 171) notes, in a pas-
sage of his book Consciousness Explained extracted from a section entitled
“Inside the Black Box of Consciousness”: “It is much easier to imagine the
behavior (and behavioral implications) of a device you synthesize ‘from the
inside out’ one might say, than to try to analyze the external behavior of a
‘black box’ and figure out what must be going on inside.” The point is rein-
forced, from a slightly different perspective, in Pinker’s How the Mind Works
(1997, p. 4): “In a well-designed system, the components are black boxes that
perform their function as if by magic. That is no less true of the mind. The
faculty with which we ponder the world has no ability to peer inside itself or
our other faculties to see what makes them tick.” But consciousness is just
14 Black Boxes
the tip of the iceberg. References to black boxes can be found in the work of
many prominent philosophers, such as Hanson (1963), Quine (1970), and
Rorty (1979), just to pick a few notable examples.
And, of course, black boxes are hardly limited to the philosophy of mind,
or even the field of philosophy tout court. As a contemporary biologist puts it,
“the current state of scientific practice [ . . . ] more and more involves relying
upon ‘black box’ methods in order to provide numerically based solutions
to complex inference problems that cannot be solved analytically” (Orzack
2008, p. 102). And here is an evolutionary psychologist: “The optimality
modeler’s gambit is that evolved rules of thumb can mimic optimal behavior
well enough not to disrupt the fit by much, so that they can be left as a black
box” (Gigerenzer 2008, p. 55). These are just a few among many representa-
tive samples, which can be found across the board.
The use (and abuse) of black boxes is criticized as often as it is praised. Some
neuroscientists scornfully dub the authors of functional models containing
boxes, question marks, or other filler terms, “boxologists.” In epidemiology—
the branch of medicine dealing with the incidence, distribution, and possible
control of diseases and other health factors—there is a recent effort to over-
come the “black box methodology,” that is, “the methodologic approach that
ignores biology and thus treats all levels of the structure below that of the in-
dividual as one large opaque box not to be opened” (Weed 1998, p. 13). Many
reductionists view black boxes as a necessary evil: something that does occur
in science, but that is an embarrassment, not something to celebrate.
In short, without—yet—getting bogged down in details, references to
black boxes, for better or for worse, are ubiquitous. Analogous remarks can
be found across every field, from engineering to immunology, from neuro-
science to machine learning, from analytic philosophy to ecology. What are
we to make of these boxes that everyone seems to be talking about?
Familiar as it may ring, talk of boxes here is evidently a figure of speech.
You may actually find a black box on an aircraft or a modern train. But you
will not find any such thing in a philosophy professor’s dusty office any more
than you will find it in a library or research lab. What exactly is a black box?
Simply put, it is a theoretical construct. It is a complex system whose struc-
ture is left mysterious to its users, or otherwise set aside. More precisely, the
process of black-boxing a specific phenomenon involves isolating some of its
core features, in such a way that they can be assumed without further micro-
explanation or detailed description of its structure.
Bricks and Boxes 15
These issues will be developed throughout the monograph. Meanwhile,

let me stress that not all black boxes are the same. Some work perfectly well.
Others muddy the waters by hiding or obliterating crucial detail. Some boxes
are opaque for everyone, as no one is currently able to open them. Others, like
phones, depend on the subject in question. Some black boxes are constructed
out of necessity: ignorance about the underlying mechanisms or computa-
tional intractability. Others are the product of error, laziness, or oversight.
Yet again, some are constructed to unify fields. Still others draw disciplinary
boundaries. In a nutshell, black-boxing is a more complex, nuanced, and
powerful technique than is typically assumed.
Given these epistemic differences in aims and strategies, does it even make
sense to talk about the practice of black-boxing, in the singular? This book
is founded on the conviction that, yes, it does make sense. I shall argue that
there is a methodological kernel that lies at the heart of all black boxes. This
core phenomenon involves the identification of some entity, process, or
mechanism that, for various reasons, can be idealized, abstracted away, and
recast from a higher-level perspective without undermining its effectiveness,
explanatory power, or autonomy. But how does this work? Answering these
deceptively simple questions will require some effort.
In sum, here is our agenda. What is a black box? How does it work? How
do we construct one? How do we determine what to include and what to
leave out? What role do boxes play in contemporary scientific practice? If you
have the patience to explore some fascinating episodes in the history of sci-
ence, together we will address all of these issues. I shall argue that a detailed
analysis of this widespread practice is descriptively accurate, in the sense
that it captures some prominent albeit neglected aspects of scientific prac-
tice. It is also normatively adequate, in that it triggers a plausible analysis of
the relation between explanations pitched in different fields, at different epi-
stemic levels. Moreover, my proposal promises to bring together the insights
on both reductionism and antireductionism, while avoiding the pitfalls of
both extremes. Simply put, there are two aspects of ignorance in science. On
the one hand, reductionism captures how ignorance points to what needs
to be discovered. Antireductionism, on the other hand, reveals how science
can proceed in the face of ignorance. Black boxes capture both dimensions,
showing how autonomy and reduction are complementary, not incompat-
ible, and offering a fresh perspective on stagnating debates. For all these
reasons, black boxes are the perfect candidate for replacing stale models at
the heart of philosophical discussions of scientific methodology.
16 Black Boxes
This is our ambitious goal. Before getting down to business, we still have a
couple of chores. First, I need to map the terrain ahead of us and clarify my
perspective. This will be the task of section 1.4, which offers a synopsis of the
chapters to come. Finally, section 1.5 concludes this preliminary overview
with a couple of heads up and caveats about the aims and scope of the project.
§1.4. Structure and Synopsis
This book is divided into ten chapters, including the introduction you are
currently reading. Chapter 2, “Between Scylla and Charybdis,” provides an
overview of the development of the reductionism vs. antireductionism de-
bate, which has set the stage for philosophical analyses of science since the
early decades of the twentieth century. Our point of departure is the rise and
fall of the classical model of reduction, epitomized by the work of Ernest
Nagel. Next is the subsequent forging of the “antireductionist consensus” and
the “reductionist anti-consensus.” Once the relevant background is set, the
chapter concludes by arguing that modest reductionism and sophisticated
antireductionism substantially overlap, making the dispute more termino-
logical than is often appreciated. Even more problematically, friends and foes
of reductionism tend to share an overly restrictive characterization of the in-
terface between levels of explanation. Thus, it is time for philosophy to move
away from these intertwining strands, which fail to capture the productive
interplay between knowledge and ignorance in science, and to develop new
categories for charting the nature and advancement of the scientific enter-
prise. Reductionism and antireductionism will return in the final chapter.
Before doing so, we shall explore a new path by focusing on an explanatory
strategy that, despite being well known and widely employed, currently lacks
a systematic analysis. This strategy is black-boxing.
Chapter 3, “Lessons from the History of Science,” starts cooking our alter-
native to “Scylla” and “Charybdis” by providing four historical illustrations
of black boxes. The first two case studies originate from two intellectual
giants in the field of biology. Darwin acknowledged the existence and signif-
icance of the mechanisms of inheritance. But he had no adequate proposal to
offer. How could his explanations work so well, given that a crucial piece of
the puzzle was missing? A similar shadow is cast on the work of Mendel and
his early-twentieth-century followers, the so-called classical geneticists, who
posited genes having little to no evidence of the nature, structure, or even
Bricks and Boxes 17
the physical reality of these theoretical constructs. How can the thriving field
of genetics be founded on such a fragile underpinning, a crackling layer of
thin ice? The answer to both conundrums lies in the construction of black
boxes, which effectively set to the side the unknown or mistaken details of
these explanations without impacting their accuracy and robustness. Then
came the Modern Synthesis, first, and the Developmental Synthesis later,
which began to fill in the blanks, opening Darwin’s and Mendel’s black boxes,
only to replace them with new black boxes. This phenomenon is by no means
unique to biology. Another illustration is found in the elimination of mental
states from the stimulus-response models advanced by psychological behav-
iorism. A final example comes from neoclassical economics, whose “as if ”
approach presupposes that the brain can be treated as a black box, essen-
tially setting neuropsychological realism aside. The history of science, I shall
argue, is essentially a history of black boxes.
In addition to illustrating the prominence of black boxes across the sci-
ences, these episodes also show that, contrary to a common if tacit belief,
black-boxing is hardly a monolithic, one-size-fits-all strategy that allows
scientific research to proceed in the face of our ignorance. These theoret-
ical constructs can play various subtly different roles. Yet, despite substantial
methodological differences, there is a common thread. All four case histories
point to the same core phenomenon: the identification of mechanisms that,
for various reasons, are omitted from the relevant explanations. This is done
via the construction of a black box. But what is a black box and how does
black-boxing work? How are these entities constructed? What distinguishes
a “good” box from a “bad” one? To answer these questions, we need to pro-
vide a more systematic analysis of this explanatory strategy.
Chapter 4, “Placeholders,” poses the foundations of this project. It should
be evident, even just from this cursory introduction, that black boxes func-
tion as placeholders. But what is a placeholder? What role does it play in
science? I set out to answer these questions by introducing two widespread
theses concerning the concept of biological fitness. First, fitness is commonly
defined as a dispositional property. It is the propensity of an organism or trait
to survive and reproduce in a particular environment. Second, since fitness
supervenes—that is, depends, in a sense to be clarified—on its underlying
physical properties, it is a placeholder for a deeper account that dispenses
with the concept of fitness altogether. Plausible as they both are, these two
theses are in tension. Qua placeholder, fitness is explanatory. Qua disposi-
tion, it explicates but cannot causally explain the associated behavior. In the
18 Black Boxes
second part of the chapter, I suggest a way out of this impasse. My solution,
simply put, involves drawing a distinction between two kinds of placeholders.
On the one hand, a placeholder may stand in for the range of events to be
accounted for. In this case, the placeholder functions as a frame. It spells
out an explanandum: a behavior, or range of behaviors, in need of explana-
tion. On the other hand, a placeholder may stand in for the mechanisms,
broadly construed, which bring about the patterns of behavior specified by
the frame, regardless of how well their nature and structure are understood.
When this occurs, the placeholder becomes an explanans and I refer to it as a
difference-maker.
Both kinds of placeholders—frames and difference-makers—play a piv-
otal role in the construction of a black box. Chapter 5, “Black-Boxing 101,”
breaks down this process into three constitutive steps. First, in the framing
stage, the explanandum is sharpened by placing the object of explanation
in the appropriate context. This is typically accomplished by constructing a
frame, a placeholder that stands in for patterns of behavior in need of expla-
nation. Second, the difference-making stage provides a causal explanation of
the framed explanandum. This involves identifying the relevant difference-
makers, placeholders that stand in for the mechanisms producing these
patterns. The final representation stage determines how these difference-
makers should be portrayed, that is, which mechanistic components and
activities should be explicitly represented, and which can be idealized or ab-
stracted away. The outcome of this process is a model of the explanandum, a
depiction of the relevant portion of the world. This analysis provides and jus-
tifies the general definition we were looking for. A black box is a placeholder—
frame or difference-maker—in a causal explanation represented in a model.
By now, we will be ready to put this to work.
Is this three-step recipe adequate and accurate? Does the proposed defi-
nition capture the essence of black-boxing? What are its advantages? What
are its limitations? These questions are taken up in the following chapters.
Chapter 6, “History of Science, Black-Boxing Style,” revisits our case studies
from the perspective of the present analysis of black boxes. By breaking down
these episodes into our three main steps, we are able to see how it was pos-
sible for Darwin to provide a simple and elegant explanation of such a com-
plex, overarching explanandum: distributions of organisms and traits across
the globe. It also explains why Mendel is rightfully considered the founding
father of genetics, despite having virtually no understanding of what genes
are, how they work, and even if they existed from a physiological perspective.
Bricks and Boxes 19
Furthermore, if Darwin and Mendel are praised for skillfully setting the
mechanisms of inheritance and variation aside and keeping them out of
their explanations, why is Skinner criticized for providing essentially the
same treatment of mental states? What distinguishes Darwin’s and Mendel’s
pioneering insights from Skinner’s influential, albeit outmoded, approach
to psychology? Finally, our analysis of black boxes sheds light on the con-
temporary dispute over the goals and methodology of economics, dividing
advocates of traditional neoclassical approaches from more or less revolu-
tionary forms of contemporary “psycho-neural” economics.
After providing a systematic definition of black boxes and testing its ad-
equacy against our case histories, it is time to address and enjoy the phil-
osophical payoff of all this hard work. This begins in Chapter 7, “Diet
Mechanistic Philosophy,” which compares and contrasts the black-boxing
approach with a movement that has gained much traction in the last couple
of decades within the philosophy of science: the “new wave of mechanistic
philosophy.” The new mechanistic philosophy was also born as a reaction to
the traditional reductionism vs. antireductionism divide. Unsurprisingly, it
pioneers many of the theses discussed here. This raises a concern. Is my treat-
ment of black boxes as novel and original as I claim? Or it is just a rehashing
of ideas that have been on the table since the turn of the new millennium? As
we shall see, the black-boxing recipe fits in quite well with the depiction of
science being in the business of discovering and modeling mechanisms. All
three steps underlying the construction of a black box have been stressed,
in some form or degree, in the extant literature. Nevertheless, the construc-
tion of black boxes, as I present it here, dampens many of the ontological
implications that characterize the contemporary landscape. This allows us
to respond to some objections raised against traditional mechanism. For this
reason, I provocatively refer to black-boxing as a “diet” mechanistic philos-
ophy, with all the epistemic flavor of your ole fashioned mechanistic view, but
hardly any metaphysical calories. Now, we can really begin to explore philos-
ophy of science, black-boxing style.
Reductionism contends that science invariably advances by descending to
lower levels. Antireductionism flatly rejects this tenet. Some explanations,
it claims, cannot be enhanced by breaking them down further. But why
should this be so? What makes explanations “autonomous”? A popular
way of cashing out the antireductionist thesis involves the concept of emer-
gence. The core intuition underlying emergence is simple. As systems be-
come increasingly complex, they begin to display properties which, in some
20 Black Boxes
sense, transcend the properties of their parts. The main task of a philosoph-
ical analysis of emergence is to cash out this “in some sense” qualifier. In
what ways, if any, do emergents transcend aggregative properties of their
constituents? How should one understand the alleged unpredictability, non-
explainability, or irreducibility of the resulting behavior? Answering these
questions might seem simple at first glance. But it has challenged scientists
and philosophers alike for a hot minute. Chapter 8, “Emergence Reframed,”
presents, motivates, and defends a strategy for characterizing emergence
and its role in scientific research, grounded in our analysis of black boxes.
Emergents, I maintain, can be characterized as black boxes: placeholders
in causal explanations represented in models. My proposal has the wel-
come implications of bringing together various usages of emergence across
domains and to reconcile emergence with reduction. Yet, this does come at
a cost. It requires abandoning a rigid perspective according to which emer-
gence is an intrinsic or absolute feature of systems, in favor or a more contex-
tual approach that relativizes the emergent status of a property or behavior to
a specific explanatory frame of reference.
Chapter 9, “The Fuel of Scientific Progress,” addresses a classic topic
that, over the last couple of decades, has been unduly neglected: the ques-
tion of the advancement of science. Setting up the discussion will require
us to retrace our steps back to the roots of modern philosophy of science.
Logical positivism provided an intuitive and prima facie compelling ac-
count of scientific knowledge. Science advances through a slow, constant,
painstaking accumulation of facts or, more modestly, increasingly precise
approximations thereof, in a “brick-by-brick” fashion (§1.1). These good old
days are gone. In the wake of Kuhn’s groundbreaking work, positivist philos-
ophy of science was replaced by a more realistic and historically informed
depiction of scientific theory and practice. However, over half a century has
now passed since the publication of Structure. Despite valiant attempts, we
still lack a fully developed, viable replacement for the cumulative model
presupposed by positivism. At the dawn of the new millennium, mainstream
philosophy eventually abandoned the project of developing a grand, over-
arching account of science. The quest for generality was traded in for a more
detailed analysis of particular disciplines and practices. I shall not attempt
here a systematic development of a post-Kuhnian alternative to logical posi-
tivism. More modestly, my goal is to show how the black-boxing strategy can
offer a revamped formulation of scientific progress, an important topic that
Bricks and Boxes 21
lies at the core of any general characterization of science, and to bring it back
on the philosophical main stage, where it legitimately belongs.
Chapter 10, “Sailing through the Strait,” takes us right back to where we
started. Chapter 2 characterizes contemporary philosophy of science as
metaphorically navigating between Scylla and Charybdis, that is, between
reductionism and antireductionism. There, I ask two related families of
questions. First, is it possible to steer clear of both hazards? Is there an al-
ternative model of the nature and advancement of science that avoids the
pitfalls of both stances and, in doing so, provides a fresh way of presenting
science to an educated readership in a more realistic fashion? Second, how
does science bring together the productive role of ignorance and the pro-
gressive growth of knowledge? The final chapter cashes out these two prom-
issory notes. These two sets of problems have a common answer: black
boxes. Specifically, the first four sections argue that the black-boxing strategy
outlined throughout the book captures the advantages of both reductionism
and antireductionism, while eschewing more troublesome implications. The
final section addresses the interplay of ignorance and knowledge.
§1.5. Aim and Scope
I conclude this introductory overview by clarifying the aim and scope of this
work. At the most general level, I have three main targets in mind.
My first goal is a philosophical analysis of an important theoretical con-
struct and how it affects scientific practice. Talk about black boxes is ubiq-
uitous. This metaphor is widely employed by scientists, philosophers,
historians, sociologists, politicians, and many others. Yet, no one ever tells
us exactly what to make of this figure of speech. I offer to pick up this tab.
I should make it clear that my aim is hardly to dismiss and replace all the
excellent work in general history and philosophy of science that has been de-
veloped over the last few decades. The collective goal of the ensuing chapters
is twofold. On the one hand, they develop and refine the process of black-
boxing by appealing to some traditional debates in general philosophy of sci-
ence. On the other hand, I also want to suggest that black boxes provide a
clear and precise framework to systematize an array of traditional concepts,
whose nature has proven to be notoriously elusive, especially after the uni-
fying force of logical positivism has waned. More generally, my objective
is to advocate and justify a shift in perspective. If we move away from the
22 Black Boxes
traditional reductionism vs. antireductionism divide and recast these impor-

tant questions in terms of black boxes, we shed new light on these projects.
Admittedly, this is an ambitious endeavor. Thus, I must convince you that
this bold move is worth the cost, by showing the potential payoff in the form
of many new exciting research projects.
My second goal is, broadly speaking, historical. I maintain that black
boxes play a prominent role in the development and advancement of science.
For my arguments to be persuasive, the case studies need to be both accurate
and illuminating. Admittedly, this book is not, primarily, a scholarly contri-
bution to the history of science—an endeavor that lies beyond my profes-
sional competence. My project falls more within the confines of a rational
reconstruction, as traditionally conceived in the philosophy of science.8 At
the same time, this work is also intended to have some historical significance,
as black boxes are important to understand not merely what these authors
should have done, but also what they actually said and did.9
My third goal pertains more directly to the philosophy of science. As we
shall see in the ensuing chapters, black-boxing provides the intellectual tools
to understand and advance various foundational concepts. My objective is
to begin revisiting questions about causation, explanation, emergence, prog-
ress, and the old reductionism vs. antireductionism debate in a new guise.
This requires a blend of conceptual rigor and historical accuracy.
Finally, a word about my intended readership. This book should be of in-
terest to both philosophers and theoretically inclined scientists, as well as
students of each discipline. Beginning and advanced scholars should have
no problem navigating the arguments and concepts presented here. The ex-
position is entirely self-contained, with the exception of some basic formal
logic. My hope, in keeping the discussion widely accessible, is to encourage
researchers from various disciplines to venture into the depths of this ex-
citing topic.
8 The notion of rational reconstruction is rooted in the groundbreaking work of Carnap (1956b);
Kuhn (1962); and, later, Lakatos (1976).
9 I try to adhere to two adequacy conditions borrowed from Kitcher (1993, pp. 12–13). First, if
something is attributed to a figure, that attribution is correct. Second, nothing is omitted which, if
introduced into the account, would undermine the point made.
2
Between Scylla and Charybdis
Come fa l’onda là sovra Cariddi,

che si frange con quella in cui s’intoppa,
così convien che qui la gente riddi.
—Dante, Inferno, VII, 16–18*
§2.1. Introduction
Our discussion began, in Chapter 1, with a well-known image of science.

From this perspective, the scientific enterprise is properly conceived as a
progressive “brick-by-brick” accumulation of knowledge and facts. Despite
its hallowed history and familiar ring, this view is now rejected by most
specialists. Yet it remains popular among the educated general public, who
take it straight out of our textbooks. These considerations raise a question.
Why are we actively promoting such a distorted characterization of science?
This book constitutes a long answer to this query, together with a sketch of
an alternative solution. The constructive proposal will be spelled out in the
ensuing chapters. Before doing so, we need to address the status quo. My pars
destruens can be presented as follows. We currently have two general models
that implicitly guide much scientific practice, and which have been devel-
oped, more explicitly, within theoretical and philosophical discussions of
scientific methodology. Neither stance is fully satisfactory. Overcoming the
old cumulative image while, at the same time, explaining the productive role
of ignorance requires us to move past extant conceptualizations. Motivating
this claim is our present task. To get the proverbial ball rolling, I introduce a
metaphor that will prove useful along the way.
* “Even as waves that break above Charybdis, //each shattering the other when they meet, //so
must the spirits here dance the round dance.” Translation by A. Mandelbaum.
DOI: 10.1093/oso/9780190095482.003.0002
24 Black Boxes
§2.2. The Rock Shoal and the Whirlpool
In Greek mythology, sailors who dared to navigate across the Strait of

Messina were confronted by two deadly perils, positioned so close to each
other to pose an inescapable trap. The first threat was Scylla, a rock shoal
located along the Italian shore, colorfully portrayed by Homer as a six-
headed monster. The second was Charybdis, a huge whirlpool off the island
of Sicily. While the beautiful Mediterranean coastline has long lost this poetic
aura of mystery, the image has stuck. Being “between Scylla and Charybdis”
figuratively refers to a forced choice between one of two evils.
This metaphor portrays the condition of much contemporary philos-
ophy of science. Philosophers are currently “sailing” through two conceptual
hazards: the Scylla of reductionism and the Charybdis of antireductionism.
Let me begin by presenting my thesis in embryonic form. These claims are
unraveled, in greater depth, in this chapter and throughout the book.
During the heyday of logical positivism, the received view was that all
of science would eventually boil down to fundamental physics.1 To be
sure, logical empiricists were perfectly aware of how far science was from
accomplishing even a distal approximation of such a complete derivation.
Furthermore, positivists did not overlook how translating all scientific inqui-
ries into questions for physics would pose practical hurdles for cognitively
limited creatures like us. Predicting fluctuations in the stock market is chal-
lenging enough when financial transactions are understood in economic
terms. Approaching them as interactions between subatomic particles would
not make it any easier. Nevertheless, the in-principle feasibility of these
reductions, together with their unificatory and guiding roles, were central
tenets of positivist philosophy of science. This regulative ideal was conceived
along the lines of what came to be known as the “classical model,” which
views reduction as a logical relation between theories.
Influential as it once was, the classical model has fallen on hard times.
As we shall see, the positivist ideal was gradually eroded by powerful
objections, fueled by the multiple-realizability of higher-level kinds, the lack
of appropriate law-like generalizations, and the shortcomings of “syntactic”
conceptions of theories. Consequently, throughout the 1970s and the 1980s,
1 Some authors draw a fine-grained distinction between “logical positivism” and “logical empir-
icism,” separating the early stages of the movement, revolving around the Vienna Circle, from later
phases, following the post–World War II diaspora. Unless specified otherwise, for the sake of sim-
plicity, I shall use the two expressions interchangeably.
Between Scylla and Charybdis 25
an “antireductionist consensus” was forged. On this view, progressive reduc-

tion is doomed to fail, both normatively and descriptively, as an account of
science.
Unfortunately, it turns out that antireductionism itself is not devoid of
controversy. When advanced as a negative thesis—the in-principle impossi-
bility of reducing all of science to physics—it leaves us without a viable meth-
odological alternative. When cashed out as a constructive proposal, most
antireductionist projects are confronted with an array of nagging concerns.
Is antireductionism consistent with a modest materialist ontology, which
accepts the supervenience of all scientific entities on physical properties? Can
antireductionism, which purports to vindicate the “autonomy” of higher sci-
ences, explain the thriving of interdisciplinary projects, which successfully
borrow concepts and merge theories from various disciplines?
Because of these and related worries, the last few decades have witnessed
the rise of a “reductionist anti-consensus” across the philosophy of science.
Recent approaches steer clear of the logical derivations which character-
ized earlier forms of reductionism, replacing them with revamped epistemic
frameworks. The general contention fueling the new wave of reductionism
is that scientific explanations are always empowered by shifting down to
lower levels, that is, by providing descriptions of the explanandum at smaller
scales. While contemporary reductionist models avoid many of the problems
that plagued their positivist predecessors, they have attracted a brand new
host of objections. Antireductionists have new targets in sight.
Over half a century after the demise of logical positivism, the battle over
the nature and status of reductionism rages on. While no winners or losers are
in sight, both armies look fairly worn out. Each side strenuously continues its
fight, constantly refining its arguments and offering sharper rebuttals of op-
posing frameworks. But is this the best strategy for heading forward? Might
it not be time to call for a truce and reshuffle the deck?
This chapter argues that the reductionism vs. antireductionism rhetoric
poses a false dilemma. After surveying, in greater detail, the unfolding of the
debate, I suggest that sophisticated versions of both stances overlap, making
the opposition more verbal and less substantive than many disputants like
to admit. The root of the problem is that reductionists and antireductionists
alike share a restrictive interface between levels of explanations, viewing re-
duction and autonomy as antithetical, as opposed to conceiving them as two
sides of the same coin. Thus, both stances provide an impoverished concep-
tion of the nature and advancement of science.
26 Black Boxes
If this old-fashioned debate is not a promising way forward, in what

directions should we point our research? The goal of this book is to trace a
different route, thereby suggesting a recharacterization of the current phi-
losophy of science agenda. My proposal focuses on the role of black boxes
in scientific explanation. These epistemic constructs identify entities, pro-
cesses, and mechanisms that, for a variety of reasons, may be idealized or
abstracted away from our models without any significant loss of explanatory
power. These black boxes shed light on the productive side of ignorance.
The following chapters provide a systematic analysis of black-boxing and
illustrate the general theme by presenting some prominent episodes in the
history of science. We shall return to the present topic—the standoff between
reductionism and antireductionism—in Chapter 10. Specifically, I shall
then make the case that black-boxing captures insights common to both
modest reductionism and sophisticated antireductionism, while eschewing
problems and accommodating ignorance. Yet, before embarking on these
ambitious quests, allow me first to convince you that it is time to sail away
from both Scylla and Charybdis, in search of calmer waters.
Before moving on, a fair warning is due. The present chapter may feel a
little technical, especially for readers unacquainted with the relevant litera-
ture. However, the details are crucial to appreciate the character and limita-
tions of the current debate. Scholars who are already familiar with the issues
under discussion should consider moving directly to section 2.6. Empirically
oriented readers with little patience for philosophical analysis are advised to
skip this chapter entirely and begin with Chapter 3.
§2.3. The Rise and Fall of Classical Reductionism
Our excursus into the modern history of reductionism sets out by going back
to the dawn of the contemporary debate. The story starts with the classical
model of derivational reduction, a legacy of logical positivism.2
In what came to be a locus classicus, Ernest Nagel (1961) characterized
reduction as the logical deduction of the laws of a reduced theory S from the
laws of a reducing theory P, a condition known as “derivability.”3 For such
2 The exposition in this section draws heavily from Nathan and Del Pinal (2016).
3 “S” and “P” stand in, respectively, for one of the special sciences and fundamental physics. Yet, the
same schema can be applied, mutatis mutandis, to any pair of sciences hierarchically arranged: neu-
roscience and psychology, biology and chemistry, economics and sociology, etc. For the sake of sim-
plicity, let us assume that the language of these theories does not substantially overlap, that is, most
derivation to get off the ground, a second condition of “connectability” must

also be met. It has to be possible for the predicates figuring in the laws of
S to be expressed in terms of predicates belonging to the language of P. To
wit, suppose that among the laws of S there is the generalization LS: S1x →
S2x, which connects property S1x with property S2x in a nomological fashion.
To reduce this law to theory P, on Nagel’s model, one needs to show that LS
can be derived from a law LP: P1x → P2x, expressed entirely in the language
of theory P. What is needed is a series of bridge laws, principles that govern
the systematic translation of the relevant S-predicates into P-predicates. Such
principles will have the following form:
(R1) S1x ↔ P1x

(R2) S2x ↔ P2x
How should the biconditional be interpreted, for bridge principles like R1

and R2 to fulfill their role in the context of Nagel-style reduction?
First, “↔” must be transitive. This ensures that, if S1 is reduced to Q1, and
Q1 is reduced to P1, then S1 is thereby reduced to P1. This proviso allows the
reduction of any branch of science to physics to occur in a stepwise fashion.
Thus, one need not derive, say, economics from biology in a single swoop.
Such operation may be enacted gradually by, first, reducing economics to
psychology and, subsequently, reducing psychology to biology.
Second, and more important, one should not read “S ↔ P” as “S causes P.” The
reason is that causal relations are typically asymmetric. Causes bring about
their effects, but effects usually do not bring about their causes. Bridge prin-
ciples, in contrast, must be symmetric. If an S1-event is a P1-event, then a P1-
event is also an S1-event. Given these two requirements, bridge laws are most
naturally interpreted as expressing contingent event identities. Thus under-
stood, R1 states that S1 is type-identical to P1.4 To illustrate, a lightning-event
is the same type of event as a discharge-of-electricity-in-the-atmosphere
predicates of S are not predicates of P and, vice versa, the majority of P-predicates do not belong to
the vocabulary of S.
4 As Fodor (1974) notes, the bridge laws required by Nagelian reduction express a stronger posi-
tion than token physicalism, the plausible view that all events which fall under the laws of a special
science supervene on physical events. Statements like R1 and R2 presuppose type physicalism, a more
controversial tenet according to which kinds figuring in the laws of special sciences must be type-
identical to more fundamental kinds.
28 Black Boxes
event. Along the same lines, tasting sea salt and tasting sodium chloride
(NaCl) are identical types of events.
Intuitive and influential as it once was, Nagel’s framework has now fallen
on hard times, undermined by powerful methodological arguments. Before
presenting some objections, it is instructive to focus on a couple of desiderata
that the classical model accomplishes and does so quite well.
First, reductionism, in general, offers a clear and precise concept of the
unity of science. Two theories are said to be unified when one is reduced to
the other or both are subsumed under a broader, more general theory. To be
sure, many contemporary scholars reject the identification of unity with der-
ivational reduction. Some even question whether science should be viewed
as unified at all.5 Yet, this is typically a side effect of the shortcomings of re-
ductionism, specifically, of Nagel’s model. To the extent that reduction offers
a viable general model of science, it also captures its unity. Thus, unsurpris-
ingly, during the heyday of positivism, the unity of science was treated as a
matter of logical relations between the terms and laws of various fields, to be
achieved through a series of inter-theoretic reductions.6
A second accomplishment of classical reductionism is the provision of
a clear-cut account of how lower-level discoveries can, in principle, inform
higher-level theories. Let me illustrate the point with simple examples. Suppose
that we are testing a psychological hypothesis LPsy: Psy1x → Psy2x, which posits
a law-like connection between two psychological predicates: Psy1 and Psy2. If
we had a pair of reductive bridge laws that map Psy1 and Psy2 onto neural kinds
Neu1 and Neu2, then we could confirm and explain the nomological status of LPsy
directly by uncovering the neural-level connection LNeu: Neu1x → Neu2x. This is
because, as noted, the bridge laws presupposed by derivational reduction ex-
press contingent type-identities. If Psy1 and Psy2 are type-identical to Neu1 and
Neu2, and there is a law-like connection between Psy1 and Psy2, there will also be
a corresponding nomological connection between Neu1 and Neu2.
Readers unfamiliar with these debates might find the following analogy
more intuitive. Assume that water and sea salt are type-identical to H2O
and NaCl, respectively. If one provided a successful explanation of why
NaCl dissolves in H2O, under specific circumstances, then one has thereby
explained why sea salt dissolves in water, under those same conditions.
5 Prominent examples are Dupré (1993) and Cartwright (1999).

6 This tenet is made explicit in Carnap (1938) and Oppenheim and Putnam (1958).
In short, the reductive approach suggests a general model for how micro-
theories can be used to advance their macro-counterparts. The goal is to look
for lower-level implementations of higher-level processes, which—on the
presupposition of reductionism—can then be used directly to test and ex-
plain macroscopic laws and generalizations. Reduction is unification, and
unification, the story goes, is the goal and benchmark of scientific progress.
The good news is over. Now, on to some objections. The best-known
problem with derivational reduction stems from the observation that natural
kinds seldom correspond neatly across levels in the way presupposed and
required by reductive bridge laws. One could arguably find a handful of suc-
cessful Nagelian reductions in the history of science. For instance, one could
make a strong case that sea salt and NaCl are type-identical, that the action-
potential of neurons is derivable from electric impulses, or that heat has been
effectively reduced to the mean kinetic energy of constituent molecules.
Assume, for the sake of the argument, that these reductions do, in fact, fit in
well with the classical model presented earlier.7 Still, contingent event iden-
tities are way too scarce to make classical reductionism a plausible, accurate,
and general inter-theoretic model of scientific practice. In most cases, there
are no physical, chemical, or macromolecular kinds that correspond—in
the sense of being type-identical—to biological, psychological, or economic
kinds, in the manner required by Nagel’s framework. This, in a nutshell, is the
multiple-realizability argument, first spelled out by Putnam and Fodor. The
basic idea is that most higher-level kinds are multiply-realizable and func-
tionally describable. Consequently, we rarely have bridge principles like R1
and R2 (“S1x ↔ P1x”; “S2x ↔ P2x”), which posit one-to-one mappings of kinds
across levels. Rather, what we typically find in science are linking laws such
as R3, which capture how higher-level kinds can be potentially realized by a
variety of lower-level states:
(R3) S1x ↔ P1x ∨ . . . ∨ Pnx
For most higher-level predicates S, the number of corresponding lower-level

predicates P is likely to be quite large. To make things worse, all the Pi will
typically have little to nothing in common, from a lower-level perspective.
7 This is hardly uncontroversial. For some difficulties and qualifications affecting the “classical” re-
duction of thermodynamics to statistical mechanics, see Sklar (1993).
30 Black Boxes
A second difficulty with classical reductionism concerns the underlying

conception of theories. Recall that, on Nagel’s model, the relata in a reductive
relation are theories. On the standard post-positivist, “syntactic” interpreta-
tion, theories are conceived as collections of statements, including both em-
pirical laws and testable conclusions derivable from them. Note how well this
approach, adopted by many introductory scientific textbooks, fits in with the
“brick-by-brick” conception presented in Chapter 1. However, as we shall see
in section 2.4, the complex structure of many areas of the special sciences
cannot be captured in terms of laws and law-like generalizations, on pain of
generating a dramatically impoverished reconstruction.
There are two standard responses to this worry. One is to revise the defi-
nition of “theory.” For example, one could replace the syntactic conception
with a semantic approach that identifies scientific theories with collections
of models.8 Alternatively, one can deny that genetics, biochemistry, and ev-
olution count as theories at all. “Theory” can then be substituted with an
altogether different term, such as Kuhn’s “paradigms,” Toulmin’s “discip-
lines,” Laudan’s “research programs,” or Shapere’s “fields.” Whether scientific
subfields are better defined as “theories” or as something else is a question
that transcends the scope of our discussion. The relevant point is that neither
strategy will salvage classical reduction. No matter what terms we use to de-
fine them, the special sciences cannot be adequately captured and described
as the interpreted axiomatic systems presupposed by Nagel’s model. This
hammers another nail in the coffin of derivational reduction.
Trouble is not quite over. Setting aside the problem of multiple realizability
and the limitations of interpreted axiomatic systems, derivational reduction
has also been dubbed inadequate as a characterization of current trends in
scientific practice. As noted, classical reductionism identifies scientific prog-
ress with successful reduction. Hence, it carries the presupposition that, with
the advancement of the scientific enterprise, we should witness a recurring
reduction of higher-level fields to lower-level ones. Nevertheless, this pre-
diction has not been borne out. As stressed by many, the proliferation of new
fields is as frequent, or even more common, than their merger and reduc-
tion. Moreover, with the possible exception of a few hackneyed and contro-
versial examples, clear-cut instances of reduction are hard to come by. The
situation is especially troublesome in the context of biology, where even the
best candidates for Nagelian derivation, namely, Mendelian and molecular
8 This view is spelled out in Suppes (1960); van Fraassen (1980); and Lloyd (1988).
genetics, notoriously fail (§2.4). Finally, classical reductionism is not only

seemingly descriptively inaccurate. Even its normative force as a regulative
ideal, as a goal for science, has been questioned. Derivational reduction, it is
claimed, leaves unanswered—indeed, unasked—important questions about
interactions between branches of science, intermediate or linking theories,
and the way the unity of science actually comes about.9
In sum, classical theoretical reductionism has fallen on all fronts. The mul-
tiple realizability of kinds poses an apparently insurmountable challenge.
The traditional syntactic conception of theories is drastically limited in
scope. And the framework lacks both normative and descriptive force.10 To
be sure, as we shall see in section 2.5, reductionism remains a popular stance,
with feisty defenders across the philosophy of science. Yet, contemporary
reductionism has morphed. Current debates focus on epistemic questions
concerning whether higher-level events are always best explained at more
fundamental levels, not on the logical derivation of laws across science.11
Theory reduction, in its traditional guise, has been beaten to death.
§2.4. Antireductionism Strikes Back
If antireductionism were tantamount to the rejection of Nagel’s model

of derivational reduction, then it would be virtually uncontroversial.
However, antireductionists do not merely advance a negative tenet. Their
pars destruens is supplemented with a pars construens. The general posi-
tive thesis can be characterized as the explanatory autonomy of the higher
levels. This section surveys some influential arguments which attempt to
9 For an excellent discussion of these points, see Maull (1977).

10 This list of objections is by no means exhaustive. Other influential difficulties with Nagel’s orig-
inal model include an incompatibility with various forms of meaning holism and the observation
that, when a theory succeeds another in an area of science, the theories typically make contradictory
claims (Hütteman and Love 2016).
11 There have been notable attempts to amend Nagel’s framework. Schaffner (1967, 1993) argues
that the target of the deduction from the lower-level theory is not the original higher-level theory,
but a suitably corrected version. More recently, Dizadji-Bahmani et al. (2010) have generalized
Schaffner’s approach by further restricting the lower-level theory with boundary conditions and
auxiliary assumptions. While these refinements constitute significant improvements, they admit-
tedly fail to provide a general model of reduction that holds across the sciences. Following a different
strategy, other authors have attempted to rehabilitate Nagel by arguing that traditional efforts to apply
derivational reduction misinterpret his “condition of connectability” (Fazekas 2009; Klein 2009). The
prospects of salvaging Nagel’s model by reinterpreting it constitute an interesting project, albeit one
that I shall not address here. My point is simply that classical derivational reduction, in its original,
most popular, and overarching form, is doomed.
32 Black Boxes
establish this epistemic claim.12 Next, section 2.5 will consider some re-
ductionist comebacks.
One of the most influential arguments in support of explanatory autonomy
was originally developed by Putnam in his article “Philosophy and Our
Mental Life.” Putnam’s intended goal was to show that traditional philosoph-
ical discussions of the mind-body problem rest on a misleading assumption.
He begins by noting how all parties involved presuppose, more or less im-
plicitly, the following premise. If human beings are purely material entities,
then there must be a physical explanation of our behavior. Materialists em-
ploy this conditional claim in a modus ponens inference. Together with the
additional premise that we are material entities, it vindicates the in-principle
possibility of physically explaining our behavior:
(a) If humans are purely material entities, then there is a physical

explanation of human behavior.
(b) Humans are purely material entities.
(c) ∴ There is a physical explanation of human behavior.
Dualists accept premise (a) but turn the modus ponens into a modus tol-
lens by revising (b) and (c). Given that there can be no physical explana-
tion of human behavior, they argue, we cannot be purely material beings:
(a) If humans are purely material entities, then there is a physical

explanation of human behavior.
(c*) There is no physical explanation of human behavior.
(b*) ∴ Humans are not purely material entities.
12 It is important to distinguish the epistemic form of antireductionism addressed here from var-
ious metaphysical variants. For instance, Galison, Hacking, Cartwright, Dupré, and other members
of the “Stanford School” have defended an ontological thesis, the heterogeneity of the natural world,
from which follows a methodological tenet fundamentally at odds with the positivist outlook: the
disunity of science. Reductionism, which Dupré (1993, p. 88) defines as “the view that the ultimate
scientific understanding of phenomena is to be gained exclusively from looking at the constituents of
those phenomena and their properties” becomes a derivative target, in virtue of its connection with
the unity of science. A different form of metaphysical antireductionism, associated with the possi-
bility of downward causation and strong emergence, has become popular in science in the context of
systems biology. We shall discuss some of these metaphysical stances—in particular, emergence—in
Chapter 8. For the moment, I shall only be concerned with antireductionism understood primarily as
an epistemic tenet of scientific explanation.
Both arguments, Putnam contends, miss the mark. Physicalists and dualists
make the same mistake by accepting the conditional premise (a). To establish
his point, he introduces an example that has since become famous.
Consider a simple physical system constituted by a rigid board with two
holes—a circle one inch in diameter and a square one inch high—and a rigid
cubical peg just under one inch high (Figure 2.1). Our task is to explain the
intuitive observation that the peg can pass though the square hole, but it will
not go through the round hole. Why is this the case?
Putnam sketches two kinds of explanations. The first begins by noting that
both the board and the peg are rigid lattices of atoms. Now, if we compute the
astronomical number of all possible trajectories of the peg, we will discover
that no trajectory passes through the round hole. There is, however, at least
one (and likely quite a few) trajectories that will pass through the square hole.
A second kind of explanation begins in exactly the same way, by noting that
the board and the peg are both rigid systems. But, rather than computing
trajectories, it points out that the square hole is slightly bigger than the cross-
section of the peg, whereas the round hole is smaller. Let us call the first kind
of explanation “lower-level,” “micro,” or “physical” and, correspondingly,
label the second one “higher-level,” “macro,” or “geometrical.” The predict-
able follow-up question is: are both explanations adequate? If not, why not?
And, if so, which one is better?
Putnam maintains that the geometrical explanation is objectively superior,
from a methodological perspective, to its physical counterpart. (Actually, he
goes as far as saying that the physical description is not an explanation at all.
Figure 2.1. Putnam’s square-peg-round-hole scenario.

34 Black Boxes
Yet, we can set this more controversial thesis to the side.) The reason is that
the physical description evidently applies only to the specific case at hand,
since no other pegs and boards will have exactly the same atomic structure.
In contrast, the geometrical story can be straightforwardly generalized to
similar systems. Whereas the macro-explanation brings out the relevant ge-
ometrical relations, the micro-explanation conceals these laws. As Putnam
puts it: “in terms of real life disciplines, real life ways of slicing up scientific
problems, the higher-level explanation is far more general, which is why it is
explanatory” (1975a, p. 297).
The moral drawn by Putnam from his example is the explanatory au-
tonomy of the mental from the physical. Higher-level explanations, regard-
less of whether they involve pegs or mental states, should not be explained at
lower levels, in terms of biochemical or physical properties. Doing so does
not produce a better explanation of the explanandum under scrutiny.
Putnam’s conclusion has fired up a long-standing debate that rages on
to the present day. We can all agree that, for creatures like us, the geomet-
rical explanation is simple and good enough for most intents and purposes.
Furthermore, the physical explanation is likely intractable and overkill. That
much virtually no one disputes. But is it really true that macro-explanations
are objectively superior to their lower-level counterparts, in the sense that, if
we were to advance them, they would explain more and better? As we shall
see in section 2.5, reductionists beg to disagree. Before getting there, in the
remainder of this section, let us focus on how Putnam’s insight has contrib-
uted to forging an antireductionist consensus.
The square-peg–round-hole analogy involves a simple toy system. Yet,
its general conclusion, the explanatory autonomy of the macro-level, can be
readily applied to scientific cases. A famous extension of Putnam’s example
in the philosophy of biology is Kitcher’s article: “1953 and All That: A Tale
of Two Sciences.” Developing insights originally pioneered by David Hull,
Kitcher considers the prospects of reducing classical Mendelian genetics to
molecular biology. His conclusion is negative. He examines three arguments
for the in-principle impossibility of the reduction in question. First, genetics
does not contain the kind of general laws presupposed by Nagelian reduction.
Second, the vocabulary of Mendelian genetics cannot be translated into the
language of lower-level sciences, nor can these terms be connected in other
suitable ways for reasons pertaining to multiple-realizability. Third, even if
such reduction were possible, it would not be explanatory. Sure, looking at
the underlying mechanisms allows one to extend higher-level depictions in
various ways. Nevertheless, appeals to molecular biology do not deepen cy-

tological explanations of genetic phenomena.
While all three points are worthy of discussion, I focus mainly on the
third one, for two basic reasons. First, Kitcher’s final objection is directly
connected to Putnam’s square-peg–round-hole example. Second, and more
important, the first two claims are, nowadays, seldom contested. Even the
staunchest contemporary reductionist would concede the hopelessness of
reducing Mendelian to molecular genetics along the lines of Nagel’s model.
This is because, as Kitcher aptly notes, we lack the laws and bridge principles
required for applying the framework in question. In contrast, as we shall see
in the following, current reductionism takes issue with the third argument.
To establish his conclusion, Kitcher considers a modern paraphrase of
Mendel’s second law, the “law of independent assortment,” stating that
genes on non-homologous chromosomes assort independently. How does
one explain this generalization? Biology textbooks typically do so by pro-
viding a description of the cytological process of meiosis at the level of cell
bodies and their interactions. The standard story goes something like this.
Chromosomes line up with their homologues, making it possible for them to
exchange genetic material and produce pairs of recombinant chromosomes.
In the course of this process, one member of each recombinant pair goes to
each gamete in a way that makes the assignment of a member of another pair
to that same gamete statistically independent. It follows that genes on homol-
ogous chromosomes will likely be transmitted together, whereas genes on
non-homologous chromosomes tend to assort independently.
Just like Putnam, Kitcher considers the prospects of explaining Mendel’s
law of independent assortment at a lower level. What makes the example
especially interesting is that such a description is available not just in prin-
ciple, but in practice. We already know many of the molecular mechanisms
underlying the process of meiosis. The crucial question is: does this know-
ledge enhance the original cytological explanation? Kitcher recognizes that
some additional details will deepen our understanding. For instance, mo-
lecular descriptions of the formation of the spindle and the migration of the
chromosomes at the poles of the spindle, just before meiotic division, reveal
how the chromosomes are not selectively oriented toward these poles. Still,
this is roughly where Kitcher draws the line. According to an informal prin-
ciple of diminishing returns, we soon reach a threshold where providing fur-
ther molecular details contributes nothing to the explanation.
36 Black Boxes
This point deserves to be clarified. In general, richer biochemical

descriptions obviously do contribute to our biological knowledge. Borrowing
Kitcher’s terminology, molecular biology constitutes an “explanatory exten-
sion” of classical genetics. Yet, most molecular depictions of genetic processes
do not enhance the explanations offered by higher-level biological theories.
The reasons for this failure of transitivity in explanation is that, as noted by
Putnam and Fodor, the ascent or descent of levels often yields explanations
that do not generalize well. Disjunctive bunches of properties at lower levels
typically do not translate into higher-level natural kinds. Unnecessarily de-
tailed descriptions may end up muddying the waters.13
The connection between Kitcher’s argument and Putnam and Fodor’s
original insights should now be evident. Atomic properties, it is argued, add
nothing to the geometrical square-peg–round-hole explanation. If anything,
they hinder the explanatory power of higher-level descriptions. For analo-
gous reasons, biochemical details have no place in higher-level biological
theories. At the same time, Kitcher’s examples went much further than its
predecessors, effectively revealing how the problems for reductionism did
not just apply in the abstract, as with Putnam’s toy example, or to relatively
young and underdeveloped fields, as with Fodor’s discussion of the psycho-
neural interface. The strength of Kitcher’s “1953” argument was tackling
directly one of the best-case scenarios for reductionism, in one of the most
established and faster-growing fields of science. If reductionism could not
take flight in genetics, where we already have a clear and detailed picture of
the underlying molecular mechanisms, what prospects could there be for
disciplines like psychology, sociology, or economics?
In sum, problems with the classical model of reduction spurred an antire-
ductionist consensus according to which the special sciences cannot—and
should not—be progressively reduced until we hit the bedrock of funda-
mental physics. We lack the bridge laws for achieving many such reductions
in practice. And, even they were feasible in principle, these derivations would
not increase the explanatory power of macro-theories. This pars destruens
was supplemented with a pars construens, which can be characterized, in
13 As Strevens (2016, pp. 156–57) aptly summarizes it in a recent discussion, “Kitcher appears to
have the following view: moving down the levels of potential explanation from ecology to physi-
ology to cytology to chemistry to fundamental physics, there is some point beyond which fur-
ther unpacking of mechanisms becomes entirely irrelevant. [ . . . ] In ‘1953’ Kitcher accounts for
failures of transitivity by proposing that, when transitivity falls through it is because categories es-
sential to explaining high-level phenomena cannot be ascribed explanatory relevance by lower-level
explanations due to their not constituting natural kinds from the lower-level point of view.”
general, as the explanatory autonomy of the higher levels. The main thesis,
pioneered by Putnam and Fodor with an eye to the philosophy of mind,
made some form of non-reductive physicalism the default position. These
original insights were developed and applied to various other branches of
science, such as biology, psychology, and the social sciences.14
Still, some would not budge. For one, despite the clear shortcomings
outlined earlier, many continued to find the overall reductionist stance con-
vincing. Furthermore, as we shall shortly see, antireductionism is confronted
by controversial implications, just like its reductive counterpart. For these
and related reasons, the last few decades have witnessed the forging of a “re-
ductionist anti-consensus” across philosophical discussions of the sciences.
This new wave of reductionism proposes a refined framework. Moving away
from the original logico-positivist model, revamped epistemic reductionism
contends that explanations are always improved by shifting down to lower
levels. It is time to focus on this theoretical comeback.
§2.5. The Reductionists Return
As discussed in the previous section, antireductionism emerged from

problems confronting the classic model of reduction. However, antireduc-
tionism is not devoid of difficulties of its own. First, it is often presented as
a negative stance, leaving us without a fully developed methodological al-
ternative. Even its main positive core tenet—namely, autonomy—is typically
introduced as the in principle impossibility of reduction.15 Despite strenuous
attempts at clarification, concepts such as “explanatory autonomy” and “ex-
planatory extension” remain just as elusive as their reductive counterparts.
Second, the most popular and influential strand of antireductionism—
non-reductive physicalism—appears dangerously unstable. The instability,
the objection runs, arises because non-reductive physicalism is either not
really physicalism, or it is not truly non-reductive.16 Allow me to elaborate.
As a form of physicalism, non-reductive physicalism is committed to the
supervenience of all scientific events on physical properties. This is to say
that a change in any event described by the special sciences requires some
14 See Hull (1974); Garfinkel (1981); Kitcher (1984, 1999); Fodor (1999); Kincaid (1990).
15 For a recent discussion of these issues, see Nagel (2012).
16 This objection is presented and discussed in detail by Bickle (1998, 2003).

38 Black Boxes
corresponding change in the underlying micro-structure. Yet, qua anti-

reductionism, non-reductive physicalism eschews the type-identification
of properties across levels. These individually plausible assumptions seem
jointly in tension. Non-reductive physicalism is a form of property dualism,
as it posits non-physical properties. Property dualism comes with a dualist
ontology which grounds the claim that even mature physics could not ex-
haustively explain higher-level events. Thus, critics conclude, non-reductive
physicalism becomes a form of dualism, which makes it incompatible with
bona fide physicalism.
A third challenge arises from the need to reconcile the success of inter-
disciplinary endeavors with the autonomy of the special sciences.17 In one
of his early contributions, Fodor noted that “the development of science
has witnessed the proliferation of specialized disciplines at least as often as
it has witnessed their reduction to physics, so the widespread enthusiasm
for reduction can hardly be a mere induction over its past successes” (1974,
p. 97). Over four decades later, Fodor’s assessment remains accurate. Indeed,
it has been reinforced. Rather than being progressively reduced to physics,
the special sciences have spawned, developing into a number of burgeoning
subfields. The rise of interdisciplinarity is a double-edged sword for antire-
ductionism. For, if the special sciences are “autonomous,” as Fodor and many
fellow antireductionists maintain, then what explains the constant prolif-
eration of fields such as neurolinguistics, moral psychology, and neuroeco-
nomics, which bring together and integrate independent disciplines?
While all these points deserve discussion, this section focuses on a dif-
ferent objection to antireductionism, more central to our present concerns.
The problem challenges the conceptual core of antireductionism, namely, the
argument for autonomy. Should we simply accept Putnam and Kitcher’s con-
tention that micro-details do not invariably deepen macro-explanations?
Let us begin by clarifying the main issue at stake. As just noted, all proper-
ties investigated in the special sciences supervene on physics. Few scholars,
including the most fervent supporters of autonomy, would deny this basic
materialist tenet.18 Thus, the crux of the debate over reductionism is inde-
pendent of physicalism or materialism. These are typically taken for granted,
at least in contemporary philosophy of science. The crucial question is the
17 For a more detailed discussion of this point, see Nathan and Del Pinal (2016).
18 Fodor (1974) calls this thesis “token physicalism,” or “the generality of physics,” and accepts it
as unproblematic. Hütteman and Love (2016) dub it “metaphysical reductionism,” to distinguish it
from the “epistemological” reductionism presently under discussion.
one first raised in section 1.2 of Chapter 1: in principle, can physics explain
everything? Is it true that every fact explained by a special science can be
explained just as well—indeed, better, in greater depth—by physics?
Antireductionism offers a negative answer. As Putnam’s square-peg–
round-hole example purports to show, size, shape, rigidity, and other macro-
states supervene on the micro-structure: position, velocity, and other atomic
properties. And, yet, the macro-level is superior, from an epistemic stand-
point. The physical depiction is less perspicuous than the geometrical one
because it cites inessential facts and obscures relevant features. Other influ-
ential antireductionism arguments, such as Kitcher’s, build on this funda-
mental insight: the explanatory autonomy of higher-level descriptions.
Not everyone, however, was convinced. One response insists that de-
tailed physical stories do, in fact, provide enhanced illuminations of
macro-systems. We all acknowledge that biochemistry explains why taking
morphine alleviates pain. Why then would atomic structure not explain
the basic geometrical structure of the square-peg–round-hole system? The
problem with this line of reasoning is that it quickly turns into a slippery
slope. Once we apply it across the board, it turns out that the special sciences
seemingly have no explanatory problems of their own. In practice, our lim-
ited cognitive capacities may prevent human beings from being able to com-
pute the micro-explanation or from employing it. Yet, in principle, physics
always provides a deeper account of higher-level events. From this stand-
point, the autonomy of the special sciences is a mere pragmatic byproduct of
our ignorance or lack of computing power. Many reductionists suggest that
the theoretical disposability of higher levels does not undermine the intrinsic
value of biology, neuropsychology, or economics, vis-à-vis the omnipotence
of physics.19 Antireductionists typically beg to disagree.
A stronger objection turns Putnam’s argument on its head, dismissing
allegedly “autonomous” geometrical explanations as either incomplete or
straight up false.20 The effectiveness of geometrical depictions, the story goes,
presupposes a host of physical information regarding, say, the rigidity of
materials and their behavior in the conditions explicitly presented—or tac-
itly assumed—in the description of the system. As these depictions are pro-
gressively completed, filling in gaps, the crucial relevance of the fundamental
physical states and corresponding laws will become more evident.
19 For a response along these lines, see Sober (1999, 2000).
20 This rejoinder to Putnam’s argument is found in Rosenberg (2006, p. 35, fn. 3).
40 Black Boxes
Related worries have been raised against Kitcher’s arguments developed

in his article titled “1953 and All That.” As discussed in section 2.4, Kitcher
questioned the possibility of reducing classical Mendelian genetics to mo-
lecular biology. Succinctly put, does molecular theory really not explain the
classical theory of genetics? Are the shallower explanations of Mendelian ge-
netics objectively preferable and more powerful than the detailed accounts
provided by current biochemistry? Molecular biology has greatly improved
our understanding of gene replication, expression, mutation, and recombi-
nation. Does this not make the antireductionist perspective outdated? Why
such pessimism? Why not expect molecular biology to continue on its path
toward explaining genetics in accordance with the spirit, if not the form, of
post-positivist reduction?21
These debates were hardly confined to biology. A similar interplay or
arguments and counterarguments can be seen, around the same time, in phil-
osophical discussions of mind, psychology, and neuroscience. In the first half
of the twentieth century, Descartes’s mind-body problem was reformulated
in terms of reduction. Can mental states be reduced to brain states? Initial
attempts by U. T. Place and J. J. C. Smart to spell out a version of psycho-
neural reduction in accordance to the post-positivist model were thwarted
by the same problems discussed earlier in section 2.3: multiple-realizability,
lack of laws, and the like. This led to the formulation of various antireduc-
tionist theories of mind, such as functionalism and the token-identity theory.
But not all scholars viewed the endorsement of autonomy as the appro-
priate response. Identifying the problem of early type-identity approaches
with their crudeness, this neo-reductive wave in the philosophy of mind
produced several alternative approaches, such as eliminative materialism,
functionalization, and various revised versions of the identity theory.22
21 This is Ken Waters’s (1990) response to Kitcher’s “explanatory incompleteness objection.” Waters
also responds to another class of antireductionist arguments, which attempt to establish an unbridge-
able conceptual gap between these areas of biology because of subtle differences in the meaning of
parallel terms (Hull 1974; Rosenberg 1985). The main idea underlying this “unconnectability objec-
tion” is that, while both classical geneticists and molecular geneticists talk about “genes,” the term is
used very differently across these theories. Mendelian genes are identified through their phenotypes.
And the relation between molecular genes and phenotypes is exceedingly complex, frustrating any
systematic attempt to connect the two concepts along Nagel’s lines. Waters responds by arguing that
this unconnectability objection presupposes an oversimplified conception of Mendelian genes. Once
the original theory is correctly understood, he claims, “The Mendelian gene can be specified in mo-
lecular biology as a relatively short segment of DNA that functions as a biochemical unit. [ . . . ] I con-
clude that the antireductionist thesis that there is some unbridgeable conceptual gap lurking between
[Classical Mendelian Genetics] and its molecular interpretation is wrong” (1990, p. 130).
22 Eliminative materialism was jointly developed by Patricia and Paul Churchland (1979, 1986).
On this view, mental states are theoretical entities posited by “folk psychology.” Commonsensical as
it seems, folk psychology, they argue, is a flawed theory of mind and is thus not a credible candidate
Time to take stock. This chapter began with a discussion of classical reduc-
tionism and its well-known shortcomings. Next, we presented the gradual
emergence of an antireductionist consensus. While antireductionism re-
mains an influential player in the game, it is no longer hegemonic, as its
foundations have been challenged by a resurgence of the reductionist per-
spective. This revitalized reductionism acknowledges the limits of the clas-
sical model. It contends, however, that this was never the real issue at stake.
Reductionism, its new-wave defenders claim, should not be abandoned. It
should be reconfigured, from a logical relation between theories to an ep-
istemic tenet regarding scientific explanation. This shift is evident in the
writing of many influential contemporary authors.23 But the main questions
remain largely unsettled. Do micro-explanations always deepen the explan-
atory power of macro-depictions? Or are the higher levels epistemically au-
tonomous from the lower levels, as suggested by Putnam and his followers?
Can this long-standing debate be settled, once and for all?
§2.6. Scylla or Charybdis?
So, who wins? The reductionist or the antireductionist? Are we going to get
crushed by the rock or drowned by the whirlpool? The answer hinges on
whether we can—and should—describe every scientific event at more fun-
damental levels, and whether these micro-depictions invariably enhance
for integration. Rather, it should be eliminated and replaced by a developed neuroscience, which
will be more predictive, more explanatory, and better connected with extant scientific research.
Eliminative materialism is, at the most general level of description, a detailed and provocative at-
tempt to apply a revised account of classical reduction to the traditional mind-body problem. To be
sure, eliminativists prefer to talk about “elimination,” rather than “reduction.” Yet, the former can be
treated as a limiting case of the latter. Other notable reductionist strategies include Lewis’s (1972) and
Kim’s (2005) sophisticated versions of identity theory and Bickle’s (1998, 2003) “ruthlessly reductive”
account.
23 For example, Rosenberg (2006, p. 12) characterizes biological reductionism as the tenet that
“there is a full and complete explanation of every biological fact, state, event, process, trend, or gener-
alization, and [ . . . ] this explanation will cite only the interaction of macromolecules to provide this
explanation. This is the reductionism that a variety of biologists and their sympathizers who identify
themselves as antireductionists need to refute.” Or, again, “the impossibility of postpositivist reduc-
tion reveals the irrelevance of this account of reduction to contemporary biology, not the impossi-
bility of its reduction to physical science. Thus, the debate between reductionists and antireductionists
must be completely reconfigured” (Rosenberg 2006, p. 22). Similarly, in a recent article addressing
Kitcher’s arguments, Strevens (2016) suggests that the in-practice autonomy of the special sciences
can be made compatible with reductionism. This requires understanding the high-level sciences’ sys-
tematic explanatory disregard of lower-level details of implementation as practically, as opposed to
intellectually, motivated. For similar stances, see Schaffner (1967, 1993, 2006) and Hooker (1981).
42 Black Boxes
explanatory power. Once again, clear-cut answers are still wanting, and not
for lack of trying. How come? The reason, I maintain, has largely to do with
the two parties talking past each other. Allow me to elaborate.
A popular litmus test to locate one’s position on the reductionism-
antireductionism spectrum is to make an informed prediction concerning
the future state of science. This is because no one really wants to argue that
current physics is in a position to replace biology, psychology, economics, or
any other branch of science. We lack the deep understanding of subatomic
systems and the computing power required to approximate the perfect vi-
sion of a Laplacian Demon. Still, reductionism claims, in principle, it would
be possible to provide a host of physical depictions that could not merely
replace extant higher-level accounts but enhance their explanatory power.
Antireductionists beg to disagree. Even if we were in a position to translate
biological, psychological, or economic generalizations in molecular terms,
we should not do so. If anything, this would muddy the waters.
This is how the debate was first introduced in section 1.2 of Chapter 1.
Intuitive and thought provoking as they are, these questions have a short-
coming: they are untestable. The reason is simple. How do we assess, at our
present time, the fate of future science, that is, its development in the long
run? To adjudicate between these competing stances, we need to look at
whether current science better conforms to the standards of reductionism or
antireductionism. When we try to do so, the matter becomes less substantive
than most participants like to admit. To corroborate this claim, I focus, in
turn, on how the dispute has unraveled in two fields: biology and psychology.
Can all biological events be subsumed under a physical explanation?
Unfortunately, this question cannot yet be answered, and this is unlikely to
change anytime soon. Sure, we are starting to see some fruitful interdiscipli-
nary overlap between physics and biology. Still, the truth is that we are too far
from integrating these two disciplines to warrant an informed response, one
way or the other. For this reason, the fate of reductionism in biology hinges
on a more modest thesis, namely, molecular reductionism.24 According to
this tenet, there is an explanation of every biological fact that mentions only
biochemical properties of molecules and their interactions.
Whereas the general reductionist query—can physics explain every-
thing?—pertains to the realm of science fiction, molecular reductionism
may be assessable in practice, not merely in principle. As noted in section 2.5,
24 This thesis is spelled out in detail in Sarkar (1998) and Rosenberg (2006).
several functional biological concepts have already been characterized from

a molecular perspective: gene replication, expression, mutation, recombina-
tion, and so forth. Now, to be sure, even if the truth or falsity of molecular
reductionism could be successfully established beyond reasonable doubt,
we would still be far from a vindication or rejection of the tout court reduc-
tionism envisioned by logical positivism. Still, failure (success) in a specific
field would provide evidence for (against) the overarching thesis.
So, is molecular reductionism a plausible assumption? Is there an explana-
tion of every biological fact that appeals only to a purely structural language,
eschewing functional concepts and any other higher-level term?
The trouble with this question is that a clear-cut answer presupposes a rig-
orous definition of “molecular concept” and “purely molecular language,”
which we lack. These expressions turn out to be dangerously vague. Precise
characterizations have never been articulated, and it is unlikely that they will
ever be found. It is important to stress that the problem does not lie with
the explanations themselves. My claim is not that explanations in molecular
biology are imprecise or otherwise inaccurate. Quite the contrary. What is
unclear is whether the explanations in question should be classified as “struc-
tural” or “purely molecular” and, therefore, whether they adhere to the re-
ductionist paradigm or to its antireductionist counterpart.
Some illustrations should help drive the message home.25 Consider the
typical explanation of drepanocytosis, a disease commonly known as sickle-
cell anemia. Its potentially devastating effects are triggered by a point muta-
tion, a single-base substitution in the chain of nucleotides of an individual
gene, the hemoglobin gene. In short, we have a genetic mutation that produces
a physiological dysfunction, due to a structural modification of cells. This
looks like a textbook case of reduction. If this story does not conform to the
molecular reductionist paradigm, I am not sure what will.
Upon further scrutiny, things turn out to be much more complicated than
they initially appear. Is drepanocytosis a successful instance of molecular re-
duction? The key is whether only structural biochemical concepts figure in
the explanans. Is this the case? Standard textbook explanations of sickle-cell
anemia appeal to various dispositional properties, such as the flexibility of
erythrocytes and the elasticity of capillaries. Similarly, terms like “transcrip-
tion factor” involve an implicit appeal to function, that is, the regulatory role
of these proteins in the transcription of genes. Translating all dispositional
25 For a more detailed discussion of these examples, see Nathan (2012).

44 Black Boxes
and functional notions in structural terms would be anywhere from frus-

tratingly hard to impossible. Are all these properties part of a molecular lan-
guage? Or do such appeals to non-structural concepts reveal that even the
best-case scenarios for the reductionist are doomed to fail? What seemed a
cut-and-dried empirical question now looks like a matter of stipulation: what
counts as a “purely structural” or “molecular” language?
Things are about to get worse. Once we move away from relatively straight-
forward genetic explanations—arguably, the clearest examples of molecular-re-
ductionism in biology—the question of whether the concepts used are “strictly
biochemical” becomes, if anything, more quixotic. Think about reaction-dif-
fusion models, which depict chemical reactions between types of molecules
as they spread throughout a system. Are the mathematical models describing
the geometry of the diffusion-space part of the language of molecular biology?
What about textbook explanations of the segmentation of Drosophila, where
difference-makers simultaneously occur both at the molecular and geometrical
levels?
The examples themselves might seem complicated, but the underlying
point is simple. A cursory glance at current biological research reveals how
explanations help themselves to a variety of concepts: structural properties
of genes and gene products, functional-dispositional features of reactants,
spatial-geometrical parameters of systems, and diffusion processes. Are all of
these part of a “fully molecular language”? How do we even begin answering
this question objectively, that is, without appealing to stipulation?
What conclusions should be drawn? The question presently under con-
sideration is: does biological practice vindicate molecular reductionism?
A clear answer, I suggested, is thwarted by the lack of clear definitions of
“molecular language” and derivative terms. Reductionists typically maintain
that functional concepts, dispositional terms, and whatever else is required
to explain biological systems are part of molecular biology and its toolkit.
Antireductionists contend that this heterogeneous conceptual toolkit
transcends the pure biochemical realm, as the ubiquitous appeal to teleology
and function thwart any characterization of a “fully molecular” language.26 If
both parties can agree on the range of explananda and the required explan-
antia, then what exactly is under dispute? Is it the nature of biological expla-
nation or whether or not to label these explanations “molecular”?
26 These divergent characterizations of a “fully molecular” language can be found in the work
of Rosenberg (2006) and Schaffner (2006), on the reductionist side, and Culp and Kitcher (1989),
Kincaid (1990), and Franklin-Hall (2008), from the opposite perspective.
It is important to stress that the issue is independent of biology. Analogous

problems affect similar debates across the board. As noted in section 2.5, con-
temporary philosophy of mind has been dominated by the question: can
mental states, broadly construed, be reduced to brain states, broadly construed?
Behaviorists, type-identity theorists, and eliminativists answer in the positive.
Psychological states, they contend, are reducible to neurological ones. Token-
identity theorists, functionalists, and property dualists answer in the negative,
arguing that any such reduction is doomed to fail.
If reductionism is so central to the philosophy of mind, one could legitimately
ask, where do we currently stand on psycho-neural reduction after almost a
century of extensive debate? A glance at the relevant literature reveals that not
much success can be boasted. Now, we have surely made significant progress
in unveiling the psychological and neural mechanisms underlying both higher
and, especially, lower cognition. Have these results tipped the scale with respect
to the question of the reduction of mental states to brain states? If so, the news
has not yet been broken, as there seems to be no more consensus on this issue
than there was in the 1950s.
When confronted with this lack of resolution, many scholars inter-
ested in the nature of mental states—scientists and philosophers alike—
respond by pointing to the extreme intricacy of the subject matter. The
human brain is arguably the most complex system discovered so far in
the universe, with billions of cells and an astronomical number of pos-
sible connections among them. No wonder that solving these disputes is
proving so hard!
I wholeheartedly agree. I am nevertheless skeptical that the complexity
of the structures under investigation is responsible for the lack of tangible
progress in the philosophy of mind. The main culprit, I maintain, is the
alleged contrast between reductionism and antireductionism themselves.
Can theories, concepts, or laws in psychology be “reduced” to the level
of neuroscience? Once again, this question is in need of clarification. What
exactly are we asking? If reduction is understood along the lines of the clas-
sical model, then the answer is clearly negative. Let us all agree, once and
for all, that the principles of psychology cannot be translated into the princi-
ples of neuroscience via Nagelian bridge laws, in practice or in theory. Case
closed. If the question is interpreted as whether we can and should provide
a “functionalized” characterization of mental states then, all of a sudden, an
46 Black Boxes
affirmative answer looks more promising.27 Yet again, if the issue is whether
or not all psychological events could and should be explained more thor-
oughly by describing them in neuroscientific terms, the answer depends on
how we choose to characterize psychology and neuroscience. If the “language
of neuroscience” is restricted to talking about individual neurons and their
additive interactions, this is clearly insufficient to explain psychology, period.
Indeed, such a narrowly construed neuroscience would not even be powerful
enough to describe neural events, let alone mental ones. As we shall discuss,
in greater detail, in Chapter 8, neural systems, as currently understood, are
irreducible to individual causes and transcend the localization and decom-
position of their elements. In contrast, sufficiently rich characterizations of
the appropriate vocabulary—including functional, behavioral, and cognitive
concepts—will make much of psychology describable in neural terms, not
only in principle, but in practice as well. And such a reduction may well be
fruitful and insightful.
It is important to emphasize, once again, that the vagueness affecting
current debates on psycho-neural reductionism is not grounded in factual
ignorance. Sure, cognitive neuroscience is, relatively speaking, a young dis-
cipline and much still needs to be learned. Nevertheless, a look at more es-
tablished fields suggests that lack of knowledge cannot be the main issue at
stake. Consider, once again, the situation in biology. Biologists are often able
to successfully pinpoint the implementation of functional structures at the
molecular level. We already know quite a bit about, say, how phylogenetic
adaptations develop at the ontogenetic level. Many complex conditions can
be causally explained by identifying their genetic difference-makers and the
subsequent cascade of processes. Prima facie, this might suggest that the case
for or against molecular reductionism has been settled, or is getting close to
being settled, in the life sciences. Yet, as we saw earlier, this is not the case.
The debate over molecular reductionism is as open as ever. For this reason,
we should not expect advances and discoveries concerning where and
how cognitive functions are computed in the brain to solve the question of
psycho-neural reduction. Philosophers of neuropsychology should learn the
hard lesson from their colleagues in biology.
Before exploring alternatives, it is time to tie up some loose ends.
27 From a functional perspective, to “reduce” a higher-level theory to a lower-level one is to provide

an analysis of the macro-properties in terms of inputs, outputs, and the realization of their causal role
at the more fundamental level (Kim 2005).
§2.7. Why Not Have the Cake and Eat It Too?
To wrap things up, let us return to the question that has fueled our discussion.
Who wins, the reductionist or the antireductionist? Scylla or Charybdis? The
bulk of this chapter laid out several ways of presenting the two stances, and
various arguments for and against them. Section 2.6 argued that the question
of whether macro-explanations can be enhanced by rephrasing them at the
micro-level has no clear, definitive answer. The issue depends more on termi-
nology or linguistic preference than substantive disagreement. Paraphrasing
one of Wittgenstein’s aphorisms, reductionism and antireductionism are
matters of expression, not facts of the world.28 Are there more substantive
disagreements to be found? I believe there are.
Both reductionism and antireductionism center on explanation. This
strikes me as correct. Still, the interface between higher and lower levels,
implicitly adopted by both parties, is overly restrictive. Reductionists
suggest that micro- explanations invariably advance macro- depictions.
Antireductionists respond that some higher levels exhibit an epistemic au-
tonomy of sorts, in the sense that they are perfectly explanatory without the
addition of further details that would only muddy the waters. Autonomy and
reduction are typically assumed to be mutually exclusive. Indeed, autonomy
is routinely defined as the rejection of reduction and, vice versa, reduction
is characterized as the rejection of autonomy. I want to argue that these two
tenets may actually be reconciled. What does this entail?
Compared to geometrical accounts, physical descriptions provide more
comprehensive understanding of why square pegs do not go through round
holes whose diameter is approximately the length of their cross section. Still,
the geometrical explanation delivers the goods in a more succinct fashion,
omitting unnecessary detail. How is this possible? Do these observations
not flatly contradict each other? Contrary to conventional philosophical
wisdom, I shall answer these questions in the negative. By focusing not on
what our models state, but on what they leave out, it will become clear that
autonomy and reduction are really two sides of the same coin.
28 Here is Wittgenstein’s original quote: “The fallacy we want to avoid is this: when we reject some
form of symbolism, we are inclined to look at it as though we had rejected a proposition as false. It is
wrong to treat the rejection of a unit of measure as though it were rejection of the proposition. “The
chair is three feet rather than two”. This confusion pervades all of philosophy. It is the same confusion
that considers a philosophical problem as though such a problem concerned a fact of the world in-
stead of a matter of expression” (1979, p. 69).
48 Black Boxes
The reconciliation of reduction and autonomy will provide a more system-

atic explanation of the interplay between productive ignorance and know-
ledge. The reason is that neither reductionism nor antireductionism, by itself,
captures one side of the equation, but remains moot on the rest of the story.
Reductionism, it should now be clear, captures well the image of science as a
wall progressively constructed in a brick-by-brick fashion. Science advances
by breaking down complex macro-systems into smaller and smaller building
blocks, until we reach bedrock. This approach, unfortunately, seems to
leave little to no room for productive ignorance, as replacing a vague char-
acterization with more detailed descriptions invariably constitutes a step
forward. The notion of autonomy underlying antireductionism, in con-
trast, legitimizes the role of ignorance in science, by stressing how the ad-
dition of micro-information does not add anything to the explanation. If
anything, it muddies the waters. Yet, as discussed, the claim that higher-level
explanations can be objectively superior to lower-level ones remains unper-
suasive. In short, a plausible account of productive ignorance needs to com-
bine these two perspectives, not pitch one against the other. Seeing how this
can be done, however, will require some work.
In conclusion, it is time to sail away from the Scylla of Reductionism
and the Charybdis of Antireductionism. I propose to do so by focusing on
Firestein’s question introduced in Chapter 1. How does science turn igno-
rance into knowledge? Our first step will be to illustrate my general thesis
with some examples from the history of science. We will come back to the
rock shoal and the whirlpool in Chapter 10, with much more under our belts.
By then, paraphrasing the lyrics of Judas Priest, you’ll be heading out to the
highway, ready to weather every storm that’s coming atcha.
3
Lessons from the History of Science
I know, for I told me so,

And I’m sure each of you quite agrees:
The more it stays the same,
The less it changes!
—Spinal Tap, “The Majesty of Rock”
§3.1. Introduction
Chapter 1 started to draw attention to the place of ignorance in science. Can

ignorance play a genuinely productive role? How? Why is this positive con-
tribution systematically overlooked, to the point of mischaracterizing the
scientific enterprise in textbook presentations? Chapter 2 identified the main
culprit as the reductionism vs. antireductionism rhetoric, which underlies
much contemporary philosophy of science. First, the disagreement is largely
an illusion. Antithetical as they may seem, sophisticated versions of these
stances overlap, making the opposition less substantive and more semantic
than is typically assumed. In addition, neither strand adequately captures the
interplay between knowledge and ignorance. The cumulative reductionist
model simply leaves no space for ignorance. Antireductionism, in contrast,
accommodates ignorance, but fails to show how it may play a productive role
in the advancement of science.
My alternative proposal revolves around an epistemic strategy that is oft
mentioned, but seldom analyzed explicitly: a form of explanation called
black-boxing. A systematic analysis of black boxes will be postponed until
Chapters 4 and 5. Before getting there, I need to convince readers that these
constructs are as prominent as I claim them to be. To this effect, this chapter
presents four episodes in the history of science where black-boxing takes
center stage. These case studies show that our strategy is not a single, mono-
lithic practice merely intended to mask lack of knowledge. It is a flexible and
multifaceted process that accomplishes a broad range of tasks.
DOI: 10.1093/oso/9780190095482.003.0003
50 Black Boxes
§3.2. Darwin’s Quest for Inheritance and Variation
When one speaks of Darwinism today, what is usually meant is evolution by

natural selection.1 Natural selection arguably was Darwin’s most daring and
novel idea, underlying his ambitious account of apparent harmony and adap-
tation in the organic world. What made the proposal so revolutionary?
Darwin’s theory of evolution by natural selection provides a powerful
framework for explaining patterns of distribution of species and traits across
the globe. The main conceptual ingredients are introduced in the opening
four chapters of On the Origin of Species (1859). Here is a brief overview.2
The first principle of variation states the obvious: within every species,
there is always variation. The second principle of competition notes that be-
cause of the tendency of populations to grow exponentially, at almost any
stage in the history of a species, more organisms are born than can survive
and reproduce. This is essentially the “struggle for existence” theorized by
Malthus and other “classical economists” of the time. The third principle of
variation in fitness notes that some of the differences between organisms,
captured in the first principle of variation, will affect the organism’s capacity
to survive and reproduce. By combining these ideas with his fourth principle
of inheritance, according to which some of this variation is heritable, Darwin
was able to derive the idea of differential survival. From these four simple
premises follows a fifth statement, the celebrated principle of natural selec-
tion: heritable variations that improve the ability to survive and reproduce
become more prevalent over successive generations.3
Notably, none of these tenets—natural selection included—was original.
All five propositions would have been accepted by most mid-nineteenth-
century intellectuals. Darwin’s insight thus does not consist in his formula-
tion of evolution by natural selection. As Kitcher argues, the breakthrough
1 Following conventional wisdom, I refer to “evolution by natural selection” in the singular, treating
it as an individual, monolithic, and cohesive theory. Yet, strictly speaking, the “theory” of evolution
encompasses a package of independent theses. These include the constant presence of reproductive
surplus, the continuous production of individual differences, the heritability of traits, sexual selec-
tion, and a few other tenets (Mayr 1991).
2 The following reconstruction of Darwin’s tenets is inspired by Kitcher (1985, 1993).
3 Darwin has a simpler conception of natural selection, compared to contemporary evolutionists
(Mayr 1991). For Darwin, selection is essentially a one-step process, which involves a steady produc-
tion of individuals, generation after generation. Some of these are bound to be “superior,” in virtue
of having some reproductive advantage over competitors. Modern mainstream evolutionary theory,
in contrast, depicts natural selection as a two-tiered process. The first step consists in the generation
of genetically distinct individuals, the production of variation. The second step is the actual selection
process, which determines the survival and reproductive success of these organisms themselves.
Lessons from the History of Science 51
rather lies in Darwin’s pioneering understanding of the far- reaching

implications of evolution, which is capable of breaching species boundaries.
Essentially, Darwin appreciated how the simple phenomenon of differential
survival, over many generations, becomes the engine for the evolution of
species.
A comprehensive overview and discussion of evolution by natural se-
lection lies well beyond the scope of this work. The important historical
point, for present purposes, is straightforward. The availability of an unlim-
ited supply of variation, especially heritable variation in fitness, is a corner-
stone of Darwin’s own formulation of evolutionary theory. Natural selection
requires the existence of intra-species variation, that some of this variation
affects the capacity of organisms to survive and reproduce, and that some of
these differences be passed on from one generation to the next. Indeed, once
natural selection is assumed as the chief engine of evolutionary change, the
only missing piece in Darwin’s puzzle is a reliable set of processes to generate
and spread the required heritable variability within and across populations.
So, what is the source of all this heritable variation in fitness? This question
preoccupied Darwin for his entire career. Merely recognizing the abundance
of differences within the biological realm did not satisfy the great natu-
ralist, who refused to accept spontaneous variation or the manifestation of
the imperfection of the organic world as plausible accounts. Darwin rightly
demanded more. He realized how the fate of his theory depended on finding
a plausible causal explanation for the mechanisms of variation and inherit-
ance. While Darwin admittedly privileged the former over the latter,4 I mo-
mentarily set variation aside and focus mainly on inheritance.
Until late in the nineteenth century, it was widely believed that both envi-
ronmental forces and usage could affect the heritable qualities of traits. The
underlying intuition—frequently albeit misleadingly dubbed “inheritance of
acquired characters”5—is that the material basis of inheritance, what has now
4 Darwin considered inheritance and its laws as a less-immediate concern than variation and
its causes, which troubled him from the initial stages of his evolutionary thinking. Two aspects
of variation were especially problematic (Mayr 1982). First, while Darwin was clearly thinking
about the range of variation for domesticated species, he never draws a clear distinction between
interpopulational and intrapopulational variation, that is, between individual and geographical vari-
eties. Second, although Darwin acknowledges the existence of discontinuous variation, he stresses
the prevalence and biological significance of continuous variation. It was genetics which eventually
showed that there is no fundamental difference between continuous and discontinuous variation.
5 This label is imprecise because such a view typically included the modifiability of genetic material
by climatic and other environmental conditions (“Geoffrism”) or by nutrition directly, without the
intermediary role of phenotypic traits (Mayr 1982, p. 687).
52 Black Boxes
been identified as genetic material, broadly construed, is pliable or “soft.” Soft

inheritance was so commonly taken for granted that hardly anyone even
bothered to actually try to figure out its mechanistic underpinnings before
the 1850s. Justifying the existence of soft inheritance, however, became of
paramount importance with the emergence of the evolutionary framework,
as substantial questions now hinged on it. Is evolution fueled by the in-
heritance of acquired characters, as proclaimed by the French evolutionist
Lamarck and his followers? Or are there different processes at play?
As previously noted, Darwin was explicitly committed to most indi-
vidual variation being heritable. At the same time, contrary to popular be-
lief, throughout his life he accepted both soft and hard inheritance, changing
opinion only on the relative importance of the two. Even mature works
such as the Origin occasionally cite evidence in favor of soft inheritance.6
Incidentally, in this respect, Darwin’s own account of the material basis of ev-
olution is closer to Lamarck’s than many contemporary readers realize.
What was Darwin’s view of the basic laws and mechanisms of inheritance?
In the Origin, he candidly confesses his ignorance on the matter:
The laws governing inheritance are quite unknown; no one can say why
a peculiarity in different individuals of the same species, or in individuals
of different species, is sometimes inherited and sometimes not so; why the
child often reverts in certain characters to its grandfather or grandmother
or other more remote ancestor; why a peculiarity is often transmitted from
one sex to both sexes, or to one sex alone, more commonly but not exclu-
sively to the like sex. (1859, p. 13)
The tone, however, becomes less cautious in later works. In Chapter 27 of

Variations of Animals and Plants under Domestication, Darwin sketches his
own theory of inheritance. This is a “provisional hypothesis of pangenesis,”
geared toward explaining the plethora of variation within and across spe-
cies. Briefly, Darwin surmises that the transmission of heritable qualities,
as well as the unfolding of development, is caused by individually different
particles, small to the point of being invisible, which he calls “gemmules.”
In order to accommodate the possibility of organisms inheriting acquired
characters, either from the environment or through use and disuse, the view
6 As stressed by Mayr (1982), Darwin discusses three potential sources of soft variation: changes
in the environment that induce increased variability via the reproductive system, unmediated
influences of the environment, and the effect of use and disuse.
is supplemented with a “transmission hypothesis.” At any stage of the life

cycle, he suggests, cells may throw off gemmules and, when supplied with
nutrients in sufficient quantity, they multiply by self-division.
To be sure, Darwin was rather reticent about his own account of pangen-
esis.7 This caution turned out to be well-placed. His theory was first rendered
obsolete and subsequently disproven by the German evolutionist August
Weismann, who identified the nucleus as the vehicle of inheritance. Roughly
fifteen years after the formulation of Darwin’s “provisional hypothesis of
pangenesis,” Weismann successfully unraveled the complex structure of the
chromatin inside the nucleus, conclusively rejecting the very possibility of
soft inheritance. Unfortunately, Darwin and his British colleagues were un-
aware of the spectacular advancements in cytology in Mitteleuropa at this
time. Consequently, Darwin remained quite uncertain throughout his life
about the nature of continuous variation, which remained controversial until
genetics finally settled the issue after 1910.
In sum, abundance of variation, especially heritable variation in fitness,
was a cornerstone of Darwin’s theory of evolution by natural selection. The
nature and source of inheritance and variation concerned the great naturalist
throughout his long and productive career, leading him to provide his own
speculative hypotheses. One can hardly fault Darwin for what he admittedly
did not know or failed to grasp. Lacking any deep understanding of genetics,
whose rudiments had to await the rediscovery of Mendel’s work at the turn
of the twentieth century, Darwin’s explanations in Origin made no reference
to mechanisms of inheritance or variation. Furthermore, many of his subse-
quent proposals strikingly missed the mark. Pangenesis and, in particular,
the transmission hypothesis did not withstand serious scrutiny.
These well-known observations raise an interesting quandary. Given that,
by Darwin’s own light, inheritance plays a crucial role in ontogeny and, in-
directly, in phylogeny, should one not expect his evolutionary explanations
to be dreadfully inadequate? Somewhat surprisingly, the answer is negative.
Glossing over Origin, many contemporary readers are struck by the variety
and detail of examples collected therein, together with the depth of Darwin’s
insightful explanations. Topics range from traits changing function in the
course of evolution (what we now call “exaptations”), to comparative analyses
of covariation occurring in distant albeit structurally similar environments.
7 According to Mayr (1982, p. 694), he was especially wary about the transmission hypothesis,
which he referred to as a “mad dream” that nonetheless “contains a great truth.”
54 Black Boxes
One also finds rudimentary, and yet remarkably plausible, hypotheses re-
garding the formation of complex organs that require myriad generations to
evolve but could not obviously do so in a piecemeal fashion. What use are a
quarter of a wing or a not fully functional eye? Origin’s answer turned out to
be on the right track.
In short, Darwin was aware of the role of variation and inheritance in his
theory of evolution and realized how much information he lacked. He also
later integrated his views with speculative and, for the most part, incorrect
assumptions. How is it possible for his explanations to be so successful?
This puzzling aspect of Darwin’s work has not gone unnoticed among
scientists, philosophers, historians, and other scholars interested in biology.
A clear statement of the main issue, together with a sketch of a solution, can
be found in Mayr’s seminal excursus into the growth of biological thought:
The most fascinating aspect of Darwin’s confusions and misconceptions

concerning variation is that they did not prevent him from promoting a
perfectly valid, indeed a brilliant, theory of evolution. Only two aspects of
variation were important to Darwin: (1) that it was at all times abundantly
available, and (2) that it was reasonably hard. Instead of wasting his time
and energy on problems that were insoluble at his time, Darwin treated
variation in much of his work as a “black box.” It was forever present and
could be used in the theory of natural selection. But the investigation of the
contents of the box, that is, the causes of the variation, occupied Darwin
only occasionally and with little success [ . . . ] Fortunately for the solution
of the major problems with which Darwin was concerned [ . . . ] a study of
the contents of the box was unnecessary. It could be postponed until more
auspicious times. (Mayr 1982, p. 682)
Mayr can be paraphrased as claiming that the two projects of identifying the
mechanisms of transmission and variation and of spelling out the processes
of evolution are not inextricably tied to each other. One can acknowledge the
existence of hereditary variation and its fundamental biological role while
effectively setting its nature and structure aside. This seems plausible. Yet,
Mayr’s observation raises a further philosophical question. How is it possible
to segregate the mechanisms of inheritance—a notion which lies at the very
heart of the phenomenon we are explaining: evolution by natural selection—
without compromising the integrity of the explanations, making them par-
tial, superficial, or otherwise inadequate?
Mayr’s insight already contains the key to answering this question. Darwin
treats inheritance and variation as black boxes. The variability and herita-
bility of certain traits is undeniable. Darwin was clear and explicit about this.
Nonetheless, he deliberately set aside and postponed puzzles concerning
their nature and structure until more auspicious times. And, when he did try
to answer these issues, in later publications, his wrongheaded speculations
left his evolutionary analyses unscathed.
Taking stock, Darwin stressed the role of variation and heritability in his
theory. Still, he “black-boxed” the underlying mechanisms. He acknowl-
edged their existence and significance but took them for granted. Now, surely
Darwin would have loved to know more about these processes. Nevertheless,
the concepts of inheritance and transmission, as they figure in Origin, did
not require any micro-explanation to be effectively employed within evolu-
tionary theory. All this is well known. But how exactly does it work?
The aim of this book is to spell out the foundations and implications of
this black-boxing strategy. What is a black box? How do we create one?
Under what conditions is this epistemic construct legitimate, successful, or
justified? Are there better or worse ways of isolating a phenomenon? Before
delving into the nuts and bolts of our explanatory strategy, I would like to in-
troduce a few more examples. These fascinating case studies emphasize how
pervasive black-boxing is across disciplines. In addition, they reveal a mul-
tifarious array of reasons underlying the deliberate decision to leave certain
phenomena unspecified, while unveiling others. As we shall see, these motiv-
ations are vastly richer, more complex, and more exciting than Mayr’s mere
need to “postpone and await for more auspicious times.”
§3.3. Mendel’s Missing Mechanisms: A Tale

of Two Syntheses
On the Origin of Species marks a true watershed in the history of biology.

It provides a powerful recipe for explaining the evolution of organisms and
traits across the globe. Still, Darwin left many questions unanswered. It thus
became of paramount importance for researchers following in his wake to
obtain a detailed understanding of the mechanisms governing the varia-
tion and transmission of traits. Unveiling them was a slow and painstaking
achievement. A seminal contribution came from Mendel’s studies that, inci-
dentally, provide another paradigmatic illustration of black-boxing.
56 Black Boxes
In his pioneering experiments with pea plants, Mendel investigated

the process by which traits are passed on from one generation to the next.
His observations were crystallized into three famous principles, here
paraphrased in contemporary biological jargon. First, the law of segregation
states that during gamete formation, the alleles for each gene segregate one
another so that each gamete carries only one allele per gene. Second, the law
of independent assortment (previously encountered in §2.4 of Chapter 2)
asserts that genes located on non-homologous chromosomes segregate in-
dependently during the formation of gametes. Third, the law of dominance
distinguishes between “dominant” and “recessive” alleles by stressing that an
organism with at least one dominant allele will display its effect over the re-
cessive variant.
A comprehensive discussion of these celebrated results transcends the
scope of this work. Our key observation is that Mendel did not provide any
description of the mechanisms responsible for the transmission of “factors.”
In fairness, Mendel was not seeking a mechanistic characterization and, argu-
ably, he did not even have a sophisticated concept of mechanism in mind. In
accordance with his training in physics within the Austro-Hungarian school,
he was after a mathematical description of laws and their consequences,
which could capture and systematize his meticulous data. Furthermore,
Mendel predated many new cytological findings, many of which were ac-
complished in the 1870s and 1880s, years after the publication of his paper.
Still, this raises a question. Given his ignorance of basic cytology, how
could Mendel explain the transportation of characters in gametes? Simply
put, he did not explain it! As Mayr (1982, p. 716) eloquently puts it:
He [Mendel] postulated that the characters are represented by “gleichartige

[identical] oder differiende [differing] Elemente.” He does not specify what
these “Elemente” are—who could have done so in 1865?—but considers
this concept sufficiently important that he refers to these “Elemente” no less
than ten times [ . . . ] [in] the Versuche.”
The obvious follow-up is: why is Mendel unanimously considered the fa-
ther on modern genetics, given that he did not understand the fundamental
structure of genes any better than other fellow naturalists of his time?
In response, Mayr stresses two crucial differences between the Bohemian
scientist and illustrious colleagues of his, such as Darwin, Galton, and
Weismann. First, Mendel discovered a consistent 3:1 ratio in inheritance
patterns. This effectively refuted the widespread “multiple particles postu-

late,” according to which numerous identical determinants for a given unit
character are present in each cell, and many replicas of a single determinant
might be transmitted simultaneously to the germ cell. Second, Mendel real-
ized that these particles exist in sets—genes and their alleles, we would say
today—which allowed an explanation of segregation and recombination.
In short, Mendel did have a deeper understanding of variation and herit-
ability, compared to other biologists of his time, Darwin included. Yet, this
does not fully capture his achievements. Mendel’s knowledge involved more
accurate generalizations that constrain the behavior of underlying entities.
But, from a mechanistic standpoint, the Bohemian scientist black-boxed
variation and inheritance no less than his illustrious English colleague. And,
contrary to Darwin, this did not seem to bother Mendel in the least.
The key to Mendel’s call to fame lies in the story of how biology opened his
black box: the gene. This is a tale that involves more black boxes.
Mendel’s groundbreaking work, published in 1866, remained ignored by
many naturalists until its “rediscovery” in 1900. Even then, it took decades
for genetics to finally prosper as an independent discipline. In the meantime,
the life sciences were becoming increasingly fragmented. Researchers in dif-
ferent biological fields approached the study of evolution at dramatically dif-
ferent scales. Paleontologists focused on the macro-picture: the fossil record
and the phylogeny of higher taxa. Systematists were largely concerned with
the nature of species and the process of speciation. Early geneticists, in turn,
studied the variation of traits in a handful of model organisms—most notably
various species of the fruit fly Drosophila. All these disciplines were discon-
nected and, at times, even mutually hostile. Colleagues with offices across the
hall from each other bickered over who provided the deepest, most fruitful
insights into the process of evolution.
Harmony was eventually achieved through the integration of all these ev-
olutionary perspectives, pitched at different levels. This achievement can be
broken down to two broad phases. The first stage occurred in the 1920s and
1930s, when population geneticists such as Fisher, Haldane, and Wright rec-
onciled the conflict between “saltationism,” the discontinuity of characters
of Mendelian alleles, with the continuously varying characters postulated
by traditional Darwinism. The second phase, which began in the late 1930s
and continued well into the 1970s, witnessed the ecumenical efforts of
Dobzhansky, Stebbins, Simpson, Mayr, Huxley, and Rensch in quilting popu-
lation genetics together with other branches of evolutionary theory.
58 Black Boxes
This grand union—epitomized by Julian Huxley’s 1942 book Evolution: The

Modern Synthesis—rested on the recognition and acceptance of two related
conceptual pillars. The first was that gradual evolution could be explained
by minute genetic changes which produce variation. This variation subse-
quently becomes the raw material upon which natural selection acts. The
second tenet held that evolution at “higher” taxonomic levels and larger
magnitudes could be explained by these same gradual evolutionary pro-
cesses sustained over long periods of time. Thanks to the collective work of
generations of biologists, the insights of Darwin and Mendel were finally rec-
onciled, or so it seemed. The road ahead remained long and winding.
The forging of the Modern Synthesis of genetics and evolution is a story
that has been recounted many times.8 The relevant observation, from our
perspective, is that, for all its ecumenical efforts, it remained incomplete:
The Modern Synthesis established much of the foundation for how evo-
lutionary biology has been discussed and taught for the past sixty years.
However, despite the monikers of “Modern” and “Synthesis,” it was in-
complete. At the time of its formulation and until recently, we could say
that forms do change, and that natural selection is a force, but we could
say nothing about how forms change, about the visible drama of evolution
as depicted, for example, in the fossil record. The Synthesis treated embry-
ology as a “black box” that somehow transformed genetic information into
three-dimensional, functional animals. (Carroll 2005, p. 7)
Embryology was treated as a “black box.” To what is Carroll alluding?

Biologists had been aware of the intimate connection between develop-
ment and evolution for quite some time. To wit, Darwin, T. H. Huxley, and
their allies relied heavily on the—somewhat limited—embryological know-
ledge of their time as evidence for descent with modification, thereby phy-
logenetically connecting human beings to the rest of the animal kingdom.
Given that both Darwin and Huxley correctly diagnosed development as the
key to evolution, why was it left out of the Modern Synthesis?
The reason, Carroll maintains, is that for the century or so following the
publication of Darwin’s masterpiece, hardly any progress was achieved in
understanding the principles of ontogeny. How a single unicellular zygote
gradually unfolds to grow into a complex multicellular organism remained
8 A classic recollection can be found in Mayr and Provine (1980).

a mystery, eluding any plausible biological explanation. As a consequence of

this frustrating impasse, the study of embryology, heredity, and evolution—
once inextricably tied at the core of biological thought—fractured, early in the
twentieth century, into separate, independent fields. Each discipline strived
to establish its own questions, principles, and methodology. This stalemate
perdured for several decades. During this same time, embryologists became
chiefly preoccupied with phenomena that could be studied by experimen-
tally manipulating the eggs and embryos of a few species. As a result, the evo-
lutionary framework, for them, faded into the background.
This black-boxing of the mechanisms of development was far from incon-
sequential for biological research at large. Evolutionists were studying genetic
variation in populations, still largely ignorant of the relation between genes
and form. Lack of embryological knowledge turned many serious scientific
studies of the evolution of form into pointless speculative exercises. It also
promoted the pernicious stereotype of evolutionary biology as a discipline
better suited for dusty museums than research labs. How could we shed light
on the evolution of form without having any firm empirical understanding
of how said form is generated in the first place? Refining Mendel’s original
insights, population genetics had successfully established that evolution is
grounded in changes in genes. Yet, as Carroll puts it, this by itself amounted
to a “principle without an example” (2005, p. 8), as no gene affecting the form
and evolution of any animal had been pinpointed, no matter how vaguely. In
short, it gradually became clear that new insights in evolution would require
breakthroughs in embryology.
Voices calling for a reunion of embryology and evolution, filling in the
gaps left open by the Modern Synthesis, began to make themselves heard in
the 1970s. A notable contribution came from Gould’s 1977 book Ontogeny
and Phylogeny, which revived discussions of how modifications of develop-
ment may influence evolution. By this point, everyone was quite aware that
genes are among the main common denominators of development and evo-
lution. The problem was that little was known about which genes mattered in
the development of organisms. This long-standing quandary was eventually
solved by researchers working on Drosophila, the workhorse of genetics, who
cleverly devised experiments to determine the genes that control the growth
of fruit flies. And the discoveries were flabbergasting.
For over a century, biologists had assumed that various types of animals
were constructed in completely distinct ways. If genes held the key to phe-
notypic variation, it was commonly supposed, then the greater the disparity
60 Black Boxes
in form, the fewer developmental processes two organisms would have in

common at the genetic level. Things turned out very differently. Contrary
to any expectation, most of the genes identified as governing the structure
and organization of a fruit fly were found to have quasi-exact counterparts
in most organisms, including humans. This was rapidly followed by another
striking discovery. The development of various body parts—eyes, limbs,
hearts, and the like—vastly different in structure across species and long
assumed to have evolved in entirely distinct ways, was also governed by the
same genes in different animals. So much for intuition!
All of this opened up an entirely new way of looking at evolution. The
comparison of developmental genes across species spawned a new discipline
at the interface of embryology and evolution: evolutionary developmental
biology, or “evo-devo,” for short. The discovery of this genetic toolkit, highly
preserved across eras and phyla, provides further irrefutable evidence, in
case additional proof was even needed, for the descent and modification of
all organisms, humans included, from a common ancestor. Darwin’s hypo-
thesis was finally vindicated in full. This wealth of new data paints a vivid
picture of how animal forms originate and evolve. And, perhaps even more
importantly, it raises a host of new questions that will likely shape the future
of much biological research for the years and decades to come.
The emergence of a common genetic toolkit and other analogies among
species’ genomes presents an apparent paradox, somewhat mirroring
Darwin’s own concerns regarding variability. If genes are so widely shared,
how do differences arise? If species are so genetically similar, then why do
they look so different? This puzzle was resolved by corroborating two re-
lated hypotheses. First, the development of form depends on the switching
on and off of genes at different times and places in the course of develop-
ment. Second, only a small fraction of DNA codes for protein—about 1.5%,
according to recent estimates. Yet, ∼3% is regulatory. Its function is to deter-
mine where, when, and how much of a gene’s product is synthesized.
Time to start tying up some loose ends. Paraphrasing Carroll, we can con-
ceive the story of evolution to date as a drama in three acts: Act One, Darwin’s
theory of evolution by natural selection; Act Two, the forging of the Modern
Synthesis, which bridged Darwinism and Mendelism at various scales; Act
Three, the ongoing Developmental Synthesis, spurred by the birth of evo-
devo and other developments in molecular biology. All three acts star black
boxes as protagonists, as a leitmotiv. At any point in time, crucial facts are
unknown, misrepresented, or otherwise set to the side. There are questions
that cannot be answered and hypotheses that cannot be verified. This igno-
rance does not stifle biological research; it enhances it. Borrowing a Kuhnian
expression, these glitches are anomalies, not falsifications. As Mayr put it,
Darwin wisely did not waste time and energy on problems then insoluble.
He effectively postponed these issues until more auspicious times. Mendel
followed suit and so did many other successful scientists. Unraveling the his-
tory of the life sciences requires keeping track of black boxes being set up and
broken down. Our goal is figuring out how this works.
In conclusion, two points should be stressed. First, while the first part of
this chapter focused on biology, black-boxing is by no means confined here.
It permeates every area of science. Second, the tales of Darwin, Mendel, and
their successors illustrate how black boxes play various crucial and subtly dif-
ferent parts in the advancement of science. The next two sections establish
these claims by looking at two more cases.
§3.4. Absent-Minded: Psychological Behaviorism
Before the twentieth century, psychology was commonly viewed as the

science of mental life, the systematic study of consciousness. Of course,
Wilhelm Wundt, William James, Sigmund Freud, and many other prom-
inent nineteenth-century psychologists realized that behavior is a form of
evidence, and an important one for sure. However, none of these authors
assumed that their entire discipline was directly aimed at predicting human
conduct. To be sure, their approaches to the study of the mind presented sig-
nificant methodological differences. For instance, James accepted physio-
logical explanations as part of psychology and emphasized the descriptive
power of viewing the mind teleologically. In contrast, the only form of causal
explanation that Wundt allowed in his psychology was mentalistic causal ex-
planation. Nevertheless, both agreed that conscious states could be explained
by appealing to the laws governing mental life, and both presupposed that
psychological differences would be relevant to such explanations.9
Early in the twentieth century, psychologists began to question the as-
sumption that conscious states are their primary objects of study. It be-
came increasingly popular to maintain that psychology concerns itself with
9 The overview in this section draws on Flanagan (1991) and Hatfield (2002).
62 Black Boxes
explaining and predicting human conduct and, more generally, the behavior
of organisms.
Behavioral psychology comes in various forms and degrees of rigidity.
Early proponents, such as William McDougall and Walter Pillsbury, advo-
cated an account of behavior that unabashedly helped itself to the mental-
istic vocabulary of traditional psychology. Thus, introspection was among
the conceptual tools that could be employed in the process of explaining the
mind. This proto-behavioristic science essentially introduced behavior as the
chief object of study without eschewing mentalism tout court. More radical
approaches rejected the validity of introspection for psychological analysis
but retained the use of mentalistic terms in the description of human con-
duct. The most extreme proposals purported to expunge entirely all mental-
istic talk from the vocabulary of psychology.
The most influential radical variant of early behaviorism was offered by
John B. Watson. Watson was a hardcore materialist, strongly committed to
the tenet that ultimate explanations of behavior would be grounded in the
principles and language of physics and chemistry. Watson and his followers
were well aware that psychology had ways to go before developing into a full-
fledged behavioral science which eschewed all forms of intentional language.
For this reason, they allowed the use of provisional explanations which
charted stimulus-response patterns along the lines of Pavlov’s conditioning
theory and Thorndyke’s laws of effects. Yet, this is where the line was drawn.
Stimuli, responses, and other measurable bodily states were supposed to be
rigorously described by using only the—allegedly—objective vocabulary of
physical theories. In rejecting all inherently mentalistic notions, whether in-
trospective or descriptive of behavior, psychology could finally join the ranks
of legitimate natural sciences.
Psychology’s path toward scientific respectability has become the subject
of interesting debates in the history of science. According to mainstream
reconstructions, this maturation was brought to completion when neo-
behaviorists such as Tolman, Hull, and Skinner joined the methodological
positivism of Schlick, Carnap, Hempel, and other thinkers associated with
the “Vienna Circle.”10 On these traditional accounts, logical empiricism fur-
nished psychology with the conceptual tools to rid itself of untestable meta-
physical claims, thereby gaining rigor and objectivity. This was attempted by
ensuring that descriptions of behavior and its triggers be translatable—that
10 Classical accounts along these lines can be found in Boring (1950) and Leahey (1980).
is, reducible to—the language of physics along the lines of what eventually
become the classical model, discussed in section 2.3 of Chapter 2.
Over the last few decades, this standard narrative of American psychology’s
alliance with positivism has been challenged. One revisionist account is
founded upon the rejection of two assumptions presupposed, more or less ex-
plicitly, in the story rehearsed earlier.11 From this standpoint, a first mistake
is the belief that after psychology disenfranchised itself from philosophy, in
the 1890s, a state of hostility existed between these two disciplines, until the
rise of positivism reconciled them. A second misunderstanding lies in the
failure to appreciate that virtually all neo-behaviorist psychologists rejected
the main conceptual pillars of logical positivism, such as verificationism
and the analytic-synthetic distinction. This does not mean that philosophy
had no role to play in the development of behaviorism. Quite the contrary,
philosophers were involved from the very beginning, and such discussions
played a formative role in the maturation of Hull’s and, especially, Tolman’s
thought. This revisionist reconstruction rejects the common view that phi-
losophy of science only dawned in North America when positivists migrated,
following the surge of Nazism across Europe. Mainstream American phil-
osophical traditions—such as realism, neo-realism, and critical realism—
had long called for philosophy to take the sciences seriously. In short, from
this perspective, the actual relations between philosophy and psychology in
American academia were already reasonably amiable, promoting a fruitful
spirit of intellectual exchange.
Adjudicating between historical narratives lies beyond my interest and
professional competence. Be that as it may, let us fast-forward to the end of
the story and take a look at the outcome of this revolution in psychology.
The most developed form of mature behaviorism comes from the work
of B. F. Skinner, arguably the most influential psychologist to ever live and
work in the United States. Constructing psychology as the science of beha-
vior, in stark opposition to previous characterizations as the study of mind
or consciousness, Skinner followed the trail blazed by early behaviorists.
Especially influential on the young Skinner was Watson, who had diagnosed
the delayed growth of psychology as the effect of pervasive metaphysical and
epistemological vices. These attitudes were readily identified as a nefarious
legacy of Descartes’s substance dualism, which implied that minds are non-
physical, private, unobservable entities, essentially rendering any serious
11 The following reconstruction is due to Amundson (1983, 1986) and Smith (1986).
64 Black Boxes
scientific approach to psychology a chimera. Yet, as we shall see, Skinner’s

mature views were, in many important respects, more nuanced and sophisti-
cated than the ones of his illustrious behaviorist predecessors.
The attempt to place psychology on more secure methodological grounds
was undoubtedly a noble endeavor. However, in decisively rejecting any form
of mentalistic vocabulary and confining the study of the mind to talk of overt
conduct, early forms of radical behaviorism ended up trivializing psychology
by providing a drastically impoverished reconstruction.
In the initial stages of his long and productive career, a phase pervaded by
the intellectual milieu of operationalism, Skinner seems to have fallen prey
to precisely this same mistake. Yet, by the end of World War II, he had clearly
realized that an uncompromising psychological theory, which eschews as
meaningless all subjective, cognitive, and affective phenomena, would fall
short of a plausible, adequate, and comprehensive theory of mind.12 Like his
behaviorist precursors, Skinner set out to replace Cartesian dualism with a
programmatic materialist stance according to which all mental events are
physical and, therefore, they must obey the same laws that govern overt be-
havior. But, contrary to Watson’s relatively simplistic reductive materialism,
Skinner’s behavioristic psychology is much more nuanced and complex.
The fundamental difference between mental and physical phenomena, for
Skinner, is not a straightforward ontological matter. Behaviorism is an epi-
stemic stance. Allow me to elaborate.
To see what exactly is at stake, consider one of Skinner’s mature publications:
Verbal Behavior (1957). This volume, which systematizes the results of over
two decades of extensive experimental work, aims to provide a systematic and
comprehensive functional analysis of linguistic behavior. What Skinner is after
is a precise specification of the interactions between the variables that con-
trol human action and, specifically, those which determine particular verbal
responses. These variables ought to be described exclusively in terms of phys-
ical stimuli, reinforcement, deprivation, and other notions that Skinner and his
collaborators have painstakingly operationalized over numerous experimental
studies of animal conduct. As Chomsky (1959, p. 26) states in his scathing re-
view of this monograph, “the goal of this book is to provide a way to predict and
12 Skinner is often portrayed as advocating a naive “stimulus-response psychology,” analyzing be-

havior in terms of unconditioned and conditioned reflexes, along the lines of Pavlov and Watson.
This, however, is a mischaracterization, as witnessed by Skinner’s attempt to integrate traditional S-R
models with the concept of operant behavior.
control verbal behavior by observing and manipulating the physical environ-

ment of the speaker.”
What makes Skinner’s project so ambitious and controversial? Contrary
to common misconceptions, the answer is neither that it sets out to pro-
vide a functional analysis of behavior, nor that it restricts itself to studying
observables per se. The proposal is so radical, Chomsky notes, because of the
specific limitations on the observables under investigation and, in particular,
due to the simplistic description of the causes of behavior.
Skinner’s thesis is that external factors consisting of present stimulation

and the history of reinforcement (in particular the frequency, arrangement,
and withholding of reinforcing stimuli) are of overwhelming importance,
and that the general principles revealed in laboratory studies of these phe-
nomena provide the basis for understanding the complexities of verbal
behavior. He confidently and repeatedly voices his claim to have demon-
strated that the contribution of the speaker is quite trivial and elementary,
and that precise prediction of verbal behavior involves only specification
of the few external factors that he has isolated experimentally with lower
organisms. (1959, pp. 27–28).
One can perhaps paraphrase Chomsky’s insight along the following lines.
What makes Skinner’s project controversial is his deliberate choice to omit
any reference to structures “internal” to the organism, the structures which
govern the relation between inputs and outputs. Whether stimuli and envi-
ronment play a role in shaping behavior is not in question—of course they
do. And, in the absence of neuropsychological data, they might well be the
main or even the only kind of evidence.13 The striking move is the complete
rejection of abilities and capacities germane to the organism itself.
These considerations highlight how Skinner’s strategy mirrors the
approaches of Darwin and Mendel. All these authors effectively set aside the
details of the mechanisms producing the behavior in question. Many readers
will find this perplexing. Behaviorism might survive, in various revised
13 Interestingly, Chomsky is adamant on this score: “Insofar as independent neurophysiological

evidence is not available, it is obvious that inferences concerning the structure of the organism are
based on observation of behavior and outside events. Nevertheless, one’s estimate of the relative im-
portance of external factors and internal structure in the determination of behavior will have an im-
portant effect on the direction of research on linguistic (or any other) behavior, and on the kinds of
analogies from animal behavior studies that will be considered relevant or suggestive” (Chomsky
1959, p. 27).
66 Black Boxes
forms, as a methodological directive in psychology, for instance, in applied

behavioral analysis therapies. But is it not an obsolete theory of mind? How
can it be compared to some of the most successful theories in all of science?
Let me reassure everyone. I am not suggesting that these theories are on a
par. The parallel is simply meant to show that Skinner’s controversial move
cannot be his decision to construct a black box. What is problematic is the
nature and structure of the black box itself, that is, the specifics of what is in-
cluded in and what is omitted from the explanation.14
Time to tie up some loose ends. The behaviorist attempt to make psy-
chology respectable on scientific grounds involved a systematic attempt
to expunge mental states and other internal features from explanations of
human conduct. This, it is often claimed, was Skinner’s capital sin: black-
boxing mental states. This conclusion cannot be correct. Now, surely, “ana-
lytic” or “logical” behaviorism is dead, at least as a philosophical analysis of
the nature of mental states. Let it rest in peace. At the same time, as we saw,
Darwinian evolutionary theory and Mendelian genetics also contain black
boxes, and so do molecular biology and evo-devo. Since these theories are
far from discredited, black-boxing per se cannot be the root of the trouble
for behaviorism. What did Skinner do that Darwin and Mendel managed to
avoid? Where exactly did behaviorism take a wrong turn?
The key to answering these questions is to note that not all black boxes
are created equal. Not all processes and events are black-boxed for the same
reasons, and some attempts are more successful than others. Darwin and
Mendel were unable to accurately depict the mechanisms of inheritance
and variation. The completion of the Modern Synthesis had to await the
breakthroughs of evo-devo. Similarly, Skinner and his followers lacked sub-
stantive knowledge of the neuropsychology of behavior. Still, as Chomsky
lucidly points out, this is only part of the story and not the controversial one.
The behaviorist rejection of mental states is not merely an admission of ig-
norance like we saw in the biological cases. The black-boxing of internal psy-
chological variables is a deliberate methodological strategy to eschew their
relevance in the causal explanation of action. Talk of mental states should be
avoided, within a Skinnerian perspective, not because such entities do not
exist or because not enough is known about them. By reconstructing psy-
chology in such a way as to avoid any reference to mental events and neural
14 As Skinner (1974, p. 233) remarked in a later book: “The organism is, of course, not empty, and it
cannot be adequately treated simply as a black box, but we must carefully distinguish between what is
known about what is inside and what is simply inferred.”
processes, Skinner purported to dismiss—or, more charitably, to minimize—

their relevance in causal accounts of human conduct. In this sense, the mind
is completely absent from his psychological theory.
There are two general morals that can be drawn from this excursus into
the history of psychology. First, behaviorism provides another clear example
of the ubiquity of black-boxing across the sciences. Mental states are explic-
itly treated as black boxes: they are complex phenomena that, for the sake
of explaining human conduct, can be replaced by patterns of inputs and
outputs, by stimuli-and-reaction models. In this respect, Skinner’s treatment
of the mechanisms of behavior resembles Darwin and Mendel’s functional
description of the mechanisms of inheritance and variation.
Second, and more important, the methodological insulation of mental
states reveals an aspect of black-boxing that was sidelined in the biological
examples. Darwin and Mendel avoided any discussion of the mechanisms
of variation, heritability, and ontogeny because of ignorance and incorrect
knowledge. The behaviorist strategy, in contrast, eschews any direct, un-
mediated appeal to mental states because they were deemed unsuited to be
investigated scientifically. The difference could hardly be more prominent.
These observations raise a host of follow-ups. How was Darwin able to set
aside the heart of his theory without compromising the structure and suc-
cess of his explanations? How could Mendel found the field of genetics while
having virtually nothing to say about the processes by which traits are passed
on from parent to offspring? What was Skinner’s mistake in setting mental
states to the side? Could it have been avoided? These important issues require
a more systematic discussion of black-boxing. I will undertake this project
in the ensuing chapters. Before doing so, let me provide one final example,
which will unveil additional layers of complexity. After all, my immediate
aim is to convince you that black boxes can be much more diverse and sys-
tematic than contemporary scholars typically suppose.
§3.5. The Nature and Significance of Economics
What is economics? In Chapter One of his Essay On the Nature and Significance
of Economic Science, Lionel Robbins proposes one of the most influential
modern definitions: “Economics is the science which studies human behaviour
as a relationship between ends and scarce means which have alternative uses”
(1932, p. 16). The fundamental axioms of the enterprise of economics, Robbins
68 Black Boxes
maintains, are grounded in a theory of value which, in turn, is based on a theory

of preference. In this account, then, the whole of economics rests on the capacity
of agents to arrange their preferences in some order.
This construction of economics on mental foundations might seem puz-
zling to contemporary readers. Prima facie, Robbins’s definition turns eco-
nomics into a subfield of psychology, namely, the study of how human agents
rank available options, given constraints on limited temporal and material
resources. Robbins is quick to address the concern. Why and how creatures
like us attach particular values to specific objects of choice are substantial
empirical problems. Yet, he glosses, these issues do not concern economists.
They are questions for psychology. All that economists—qua economists15—
need to assume is the ranking of options itself.
Why the human animal attaches particular values in this sense to partic-
ular things, is a question which we do not discuss. That is quite properly a
question for psychologists or perhaps even physiologists. All that we need
to assume as economists is the obvious fact that different possibilities offer
different incentives, and that these incentives can be arranged in order of
their intensity. (Robbins 1932, p. 86)
This is an adage that most students of economics learn at an early stage. Still,
in the ensuing discussion, Robbins adds a remark that many of his colleagues
would now find disconcerting. No economic explanation, he claims, can be
considered adequate, let alone complete, without invoking elements of a sub-
jective or psychological nature, such as preference.16
This ineliminable subjective component in economic explanations raises
issues. From a historical perspective, it strikingly contrasts with the meth-
odological orthodoxy of the day—logical empiricism—which purported to
15 This disclaimer is required because nothing prevents the same individual or research team from
pursuing agendas in both economics and psychology. Still, for Robbins’s point to go through, it is
enough to distinguish psychological questions from economic ones.
16 “But even if we restrict the object of Economics to the explanation of such observable things as
prices, we shall find that in fact it is impossible to explain them unless we invoke elements of a sub-
jective or psychological nature. It is surely clear [ . . . ] that the most elementary process of price de-
termination must depend inter alia upon what people think is going to happen to prices in the future.
[ . . . ] It is quite easy to exhibit such anticipations as part of a general system of scales of preference.
[Footnote omitted.] But if we suppose that such a system takes account of observable data only we
deceive ourselves. How can we observe what a man thinks is going to happen? It follows, then, that
if we are to do our jobs as economics, if we are to provide a sufficient explanation of matters which
every definition of our subject-matter necessarily covers, we must include psychological elements”
(Robbins 1932, pp. 88–89).
reduce all synthetic knowledge to statements that are testable, at least in prin-
ciple. Mental states are not directly observable. Thus, they cannot be tested,
in practice or in theory. According to the verificationist semantics endorsed
by logical positivism, this turns them into meaningless gibberish.
The methodology of economics vis-à-vis its status as a science quickly be-
came a contentious matter requiring serious discussion. Prominent scholars,
such as Terence Hutchison (1938), maintained that economics, as a bona fide
science, must formulate falsifiable generalizations and subject them to rig-
orous test. This requirement might initially appear plausible, even truistic.
Upon further scrutiny, it poses strictures of all sorts. Conforming economic
methodology to the basic tenets of logical positivism turned out to be a real
challenge, independently of its appeal to psychological elements.
For one thing, several mainstream economic assumptions are heavily
hedged with ceteris paribus qualifications and, hence, not directly testable.
The reason is straightforward. An unqualified generalization such as “cutting
taxes is always followed by a raise in employment rates” can be corroborated
by ensuring that the consequent, a raise in employment rates, invariably
follows the antecedent, a cut in taxation rates. In contrast, the hypothesis that
“all things being equal, cutting taxes is followed by a raise in employment
rates” is harder to assess. When exactly should one deem it falsified? When
should things be considered equal and by what standards?
A second and more troubling issue emerged when empirical economic
claims were actually tested. Many turned out to be false. For instance, ac-
cording to a well-known theoretical hypothesis, economic agents operate
under the assumption of marginal cost pricing. However, a survey was
published revealing that real-world firms actually base their calculations on
full-cost pricing. A few years later, another study showed that firms do not
maximize profits by equating marginal costs and marginal revenues, another
core tenet of early neoclassical economics.17 In short, it soon became evident
that economic models provide an inaccurate depiction of business conduct.
If economists were supposed to behave like responsible scientists—which, in
the logico-positivist intellectual milieu of the early twentieth century, meant
responsible physicists—they were not doing a very good job.
This impasse triggered various kinds of responses. Economists associ-
ated with the Austrian School—such as Frank Knight, Carl Menger, Ludwig
von Mises, and Friedrich Hayek—willingly accepted the conclusion that the
17 These results are reported by Hall and Hitch (1939) and Lester (1946), respectively.
70 Black Boxes
standards of natural science simply do not apply to economics. This, how-

ever, is a hard bullet to bite. What methodology should economics adhere to,
if natural science is not the answer?
A different route involved showing that, despite superficial differences,
economics did meet the stringent positivist criteria. How? One option for
salvaging economics in the face of the falsity of core assumptions was to
treat its basic laws as truths, albeit inexact truths. Unfortunately, from the
standpoint of logical empiricism, this approach amounts to methodolog-
ical suicide. If the benchmark of rigorous scientific statements lies in their
testability, inexact truths are just as hard to verify as hedged ceteris paribus
generalizations. How is an inexact truth confirmed? When is it discarded,
and what should replace it? All these solutions appeared to be non-starters.
A more promising strategy was soon to be offered.
The way out of the dilemma was provided by Milton Friedman, in his cel-
ebrated 1953 article, “The Methodology of Positivist Economics.” Friedman
argued that the truth or falsity of economic assumptions is perfectly irrel-
evant to the acceptance or rejection of a positive scientific theory. The only
thing that matters is whether the hypothesis is predictive and explanatory
with respect to the class of phenomena that it is intended to cover. An ade-
quate theory, Friedman goes on to clarify, is one that explains much by little.
It must abstract the crucial factors from the mass of complex, idiosyncratic
circumstances surrounding the phenomena to be explained, and predict fu-
ture states on the basis of these core elements alone. Hence, the relevant ques-
tion to ask about the assumptions of a scientific theory is not whether they
are descriptively realistic. They never are. The appropriate issue is whether
our hypotheses constitute good enough approximations for the purposes at
hand, whether they yield accurate predictions.
Friedman illustrates his contention with simple and effective examples.
Consider the law of free-falling bodies in elementary physics: s = 1/2gt2,
where s is the distance in feet and t is time in seconds. Take the hypothesis
that leaves are positioned around a tree as if each leaf deliberately seeks to
maximize its exposure to sunlight, relative to the position of its neighbors.
Or the claim that expert billiard players make their shots as if they computed
the complicated mathematical formulas that specify the optimal direction of
travel. Taken literally, all these hypotheses are obviously false. The preceding
formula only captures the falling of bodies in a perfect vacuum. Leaves do not
seek to maximize anything, let alone sunlight exposure. And no human being
comes remotely close to computing in real time the equations underlying the
motion of billiard balls. Nevertheless, Friedman argues, these hypotheses are

commonly accepted. This is because they allow us to predict, with reasonable
accuracy, their target phenomenon: the falling of dense objects, the distri-
bution of leaves on trees, and the outcome of billiard shots. More precisely,
Friedman argues, these hypotheses are typically accepted because they strike
the right balance between precision and complexity. Sure, more predictive
generalizations are available. Yet, in most contexts, this greater predictive ac-
curacy is offset by the required amount of calculation. Could one not say the
same thing about economic tenets, such as the presupposition that individ-
uals and firms behave as if they sought to maximize their effective returns?
Economists accept or reject assumptions because of their predictive power,
not because of their (in)accuracy.
Few methodological directives have been as widely debated as the one just
presented. And yet, Friedman’s argument is often mischaracterized. First,
some have countered that realism is not detrimental to hypotheses.18 But,
on a charitable reading, this is not Friedman’s contention. The point is rather
that realism should not be a factor in theory choice. One can always prefer a
more accurate hypothesis to a less realistic one, without loss of explanatory
power, as long as the former has a better predictiveness-to-complexity bal-
ance than the latter. Regardless, the justification for doing so resides in the
greater predictive value of the hypothesis, not in its realism.
Second, Friedman’s strategy conforms economic methodology to the
standards of logical empiricism.19 Positivistic tenets such as the in-principle
testability of all scientific statements have long been rejected. Does this
not, ipso facto, make Friedman’s argument obsolete? The answer is nega-
tive. Friedman’s contention stands or falls independently of radical forms
of empiricism. In addition, this pioneering directive anticipates contempo-
rary discussions of the role of abstractions, idealizations, and other forms
of deliberate misrepresentation, which have now finally found their way
into mainstream philosophy of science.20 In short, Friedman’s strategy
18 See, for instance, Simon (1963).

19 The connection between Friedman and positivism are clearly seen in Nagel (1963), who
maintains that, while Friedman’s argument lacks cogency, it is nevertheless sound.
20 The use of fictions and distortions is commonly assumed to be a long-standing scientific prac-
tice. This, as argued by Chomsky (2012), is a misconception. The idea of taking a complex system and
transforming it into an artificial construct, which can be investigated in depth, was a huge step that
was only taken in relatively recent times. Galileo saw laws principally as calculating devices. It wasn’t
until the twentieth century that theoretical physics became a legitimate field in itself. To this day,
Chomsky notes, the idea of constructing idealized systems for the sake of computational tractability
is considered somewhat exotic outside of the hard-core natural sciences, in fields such as linguistics.
72 Black Boxes
is more powerful than most critics realize. No wonder it has been so

influential.21
The final ingredient required to reconcile Robbins’s dictum—the ine-
liminable presence of subjective psychological components in economic
explanations—with logical positivism is the revealed preference view of
utility. By assuming that choice and preference exhaustively determine each
other, effectively making the two notions interchangeable, preference may
continue to perform an essential role in economics without positing intan-
gible mental quantities. Identifying preference with choice, the former essen-
tially becomes as objectively measurable and testable as the latter. Economics
could abide to positivist demands, after all.
Since the 1950s, mainstream neoclassical economics has come under fire
from various standpoints. First, famous paradoxes due to the independent
work of Allais and Ellsberg have unveiled widespread and systematic
violations of the axioms of expected utility theory.22 Second, the “definitional
egoism” underlying rational decision-making has been critiqued as an over-
simplified view of human choice. Third, neoclassical economics is said to lack
normative force and also to give up on the intuitive assumption that mental
states explain behavior.23 I shall set all these critiques aside. My goal is not to
For an insightful philosophical discussion of idealization—which will also be discussed more sys-
tematically in Chapter 5 and Chapter 7—see Appiah (2017).
21 I should clarify that my sympathetic assessment of Friedman’s strategy deliberately focuses on

his general methodology and abstracts away from his substantive ideological baggage, political pro-
ject, and underlying agenda. As George DeMartino (2000) has convincingly argued, the crusade led
by Friedman and his disciples in defense of the free market depended heavily on the assumption
that the preferences emerging within a free market are legitimate. This, in turn, presupposes that
preferences are exogenous to economic activity. The rationale is simple. Within neoclassical eco-
nomics, the market is the primary means of preference satisfaction. But if the market also shapes,
influences, or otherwise warps the preference that it satisfies, then we cannot point to preference sat-
isfaction as a defense of free markets. Neoclassical theory sidesteps the problem by presuming exoge-
nous preferences. Similarly, neoclassical economists did not invariably, or even reliably, practice what
they preached. While holding up the test of prediction to validate their own unrealistic assumptions,
the profession also became intolerant of all other approaches. Not only were alternatives not pro-
moted. In addition, neoclassicists occasionally actively thwarted attempts to show that different
methodologies did better by their own standards of predictive performance. To wit, much econo-
metrics in the twentieth century was directed toward vindicating the claim that all would benefit
from free trade, that minimum wage actually harmed people more than it helped, and other items
on the agenda of neoliberalism. Doing justice to these considerations requires a longer—and
valuable—discussion that, however, transcends my aim and competence. Still, I stress that my praise
of Friedman’s methodological strategy does not imply any endorsement of sociopolitical agendas or
exploitation of double standards when it comes to choice and implementation of theory.
22 As we shall see, in greater detail, in section 6.5, expected utility—the utility of an entity or ag-
gregate economy over a future period of time, given unknowable circumstances—is standardly em-
ployed to evaluate decision-making under uncertainty.
23 A classic critique of “definition egoism” is Sen (1977). Philosophical discussions concerning
normativity (or lack thereof) are found in Hausman (1992) and Rosenberg (1992).
assess neoclassical economics tout court. Rather, I wish to bring to light some
methodological presuppositions of contemporary economics.
One observation is especially important from our perspective. As noted,
the combination of Friedman’s “as if ” approach with the revealed preference
view of utility effectively made economic theory kosher from a positivist
standpoint. Yet, in doing so, it also completely shielded economic models
from psychological influence and critique. The rationale is simple. In neo-
classical economics, preference is not something that is open to further anal-
ysis. The identification of preference with choice turns both into raw data
that explain the behavior of a rational—that is, a logically consistent—agent.
In a nutshell, choice, preference, and utility are all black boxes.
This screening off of psychology had a triple effect on twentieth-century
economics. First, identifying preference with choice provided a clear, simple,
and operational way to talk about reasons and preferences. Complications
arising from the idiosyncrasy of human psychology and the neural imple-
mentation of these processes, both of which, at that time, were far beyond
our ken, could be unabashedly swept under the rug. Second, if, from an eco-
nomic perspective, all there is to preference is revealed choice, then psycho-
neural discoveries about the human motivational system, in principle,
cannot shed any light on economics. We can effectively talk about behavior
without mentioning its causes. In addition, and more radically, specifying
such causes will leave our understanding of economic behavior untouched.
Third, the isolation of economics from psychology was a powerful and effec-
tive strategy to screen off the core assumptions of neoclassical theory from
potential attacks. Specifically, the superiority of free markets could not be
questioned by appealing to the psychological states of economic agents. One
could not argue, for instance, that the preferences that led to choice were
somehow normatively deficient.
The insulation of economics might have been relatively inconsequential
during the days of logical positivism, when psychology and neuroscience
were still relatively young, underdeveloped, and had little to contribute to
the study of economic phenomena. To be sure, the critical examination of
the relation between preference and behavior has a long history, harking
back to Plato and Aristotle. Still, until recently, we lacked any systematic
theory which might shed light on how these choice-mechanisms are actually
implemented in the human mind, allowing for more accurate predictions
and sharper explanations. Things have changed quite drastically over the last
74 Black Boxes
few decades. Following the cognitive revolution, proposals regarding the im-
pact of psychology and neuroscience on economics have emerged.
The claim underlying contemporary “psycho-neuro-economics” (PNE) is
that new insights into how minds and brains actually frame, compute, and re-
solve problems challenges fundamental assumptions regarding the behavior
of economic agents.24 For instance, behavioral economics has denounced
the lack of reference points in expected utility theory. In doing so, it has
emphasized how real humans, as opposed to economic agents (“econs”),
use heuristics and biases in making decisions, rather than calculating ex-
pected utilities. Neuroeconomics has also made some progress revealing
how and where mechanisms governing choice are implemented in the brain,
unveiling further inconsistencies with conventional wisdom. In short, the
story goes, classical economic notions, such as risk aversion, time preference,
and altruism, are due for a makeover. Is it finally time to open the black boxes
that were so skillfully crafted by Robbins, Friedman, and their colleagues?
Supporters of PNE answer affirmatively. This is clearly stated in one of the
founding manifestos of neuroeconomics (Camerer et al. 2005, p. 53):
Economics parted company from psychology in the early twentieth cen-

tury. Economists became skeptical that basic psychological forces could be
measured without inferring them from behavior (the same position “be-
haviorist” psychologists of the 1920s reached), which led to adoption of
the useful tautology between unobserved utilities and observed (revealed)
preferences. But remarkable advances in neuroscience now make direct
measurement of thoughts and feelings possible for the first time, opening
the “black box” which is the building block of any economic interaction
and system—the human mind.
It is worth stressing that, in calling classical economic notions “black boxes,”

the claim is not that these notions were unstructured. Von Neumann,
Morgenstern, Friedman, and their colleagues had much to say about the
nature of choice, preference, and utility. The point is, rather, that these ne-
oclassical economic constructs are black boxes relative to the underlying
24 For the sake of convenience, I introduce the moniker “psycho-neuro-economics” to lump

together various contemporary developments in behavioral economics and neuroeconomics.
Whenever the need for a finer-grained classification arises, I draw subtler distinctions between be-
havioral economics proper and its neuroeconomic counterpart.
neuro-psychological mechanisms, which played no role, in principle or in

practice, in determining and explaining their constitution.
Not everyone, however, is so enthusiastic. Naysayers vary with respect
to the scope and depth of critique. Some acknowledge the potential of
neuroeconomics and, yet, complain that, so far, it has delivered a mixture
of marketing bloopers, bad epistemology, and messy experimental prac-
tice.25 Others fear that outsourcing the study of preference to psychology
may alter the very nature of economic explanations, turning them into
questions about how the mind works. In what follows, I focus on a dif-
ferent worry.
A more trenchant objection, which echoes Friedman from a current
standpoint, maintains that psychology and neuroscience, in principle,
are irrelevant to economics.26 Economics and psychology, from this per-
spective, address distinct questions, have different objectives, and in-
troduce their own abstractions. Moreover, contrary to the therapeutic
aims of PNE, neoclassical theory identifies welfare with choice without
purporting to advance a theory of happiness. On this view, opening the
black box of mental states does nothing to economics because it is not part
of economics at all.
In conclusion, the history of economics illustrates another role for
black-boxing in science. By isolating a class of phenomena and insulating
it from further explanation, one can draw disciplinary boundaries between
subfields. These boundaries can then be used to advance a research project
in the face of epistemic limitations—it allows us to define and employ a
concept when not much is known about its underpinnings. Alternatively,
black-boxing can also become a strategic way to shore up and protect the
most vulnerable components of a theory against its rivals. A black box,
in this sense, acts as a protective belt. Perhaps classical and early neoclas-
sical economic theory segregated mental states because of ignorance or to
render their disciplines fully testable and, thereby, scientific. But no con-
temporary economist is unaware of the staggering discoveries regarding
psycho-neural mechanisms accomplished over the last few decades. No
scholar worth their salt would dismiss core studies of the mind and brain
as pseudo-scientific. It is the relevance of these discoveries for economics
25 Harrison (2008) humorously dubs this programmatic approach “MacGyver Econometrics.”

Indeed, even authors who are more sympathetic to PNE than Harrison, such as McCabe (2008), will
admit that the contribution so far has been somewhat limited.
26 The main reference here is an influential article by Gul and Pesendorfer (2008).
76 Black Boxes
that is at stake. PNE maintains that understanding these mechanisms is

advancing economic theories. This point of view is expressed clearly by
Camerer et al. (2005, p. 10): “The ‘as if ’ approach made good sense as long
as the brain remained substantially a black box. The development of eco-
nomics could not be held hostage to progress in other human sciences.” In
contrast, advocates of a “mindless” form of neoclassical economics, which
views economics and psychology as fundamentally distinct, question the
relevance of these findings for their own discipline: “Our conclusion is
that the neuroeconomic critique fails to refute any particular (standard)
economic model and offers no challenge to standard economics method-
ology” (Gul and Pesendorfer 2008, p. 7).
Who wins this battle? While I shall not take sides on this issue, I want to
stress that the current disagreement hinges, to a large extent, on the char-
acterization and methodological boundaries of economics. If economics,
by definition, presupposes the choices of agents, then psychology and neu-
roscience clearly have no relevance. The opposite conclusion can be drawn
on a broader conception, that defines economics as the study of a special
class of choices.27 We shall return to this debate in Chapter 6. The point, for
the time being, is that this duality is precisely what we should expect given
the situation discussed in section 2.6 of Chapter 2. PNE embodies the re-
ductionist tenet that breaking down economic explanations to more fun-
damental constituents increases their power. Neoclassical economics, in
contrast, embodies the antireductionist tenet of autonomy. However, as cur-
rently presented, the dispute is generally terminological. We need a different
approach to make the discussion more substantial. The key, I will argue in
later chapters, is understanding utility and other economic constructs as
black boxes.
27 To wit, an influential neuroeconomist, Camerer (2010, p. 43), offers a broad definition of eco-
nomics as “the study of the variables and institutions that influence economic choices, choices with
important consequences for health, wealth, and material possession and other sources of happi-
ness.” While Gul and Pesendorfer (2008) do not offer an explicit comprehensive definition of eco-
nomics, they narrow the field substantially with assertions such as “standard economics focuses on
revealed preference because economic data comes in this form” (p. 8). It should be evident that, on
the former definition, it becomes hard to dismiss the relevance of psychological discoveries on ec-
onomic choices. In contrast, the latter delimitation virtually dismisses the import of economics on
psychology by definition, by severely constraining what may and may not count as economic data.
§3.6. The Many Faces of Black Boxes
The practice of successfully explaining a phenomenon despite explicitly ap-

pealing to entities and processes the details of which are left unspecified or
mistakenly identified has a long and hallowed history. Newton’s treatment
of gravity—culminating in his famous dictum “Hypotheses non fingo”—is a
fascinating example. Yet, I focused our present discussion on four examples
from three special sciences: biology, psychology, and economics.
Our excursus into the history of science began by tracing its steps back
to the roots of contemporary evolutionary thinking. Darwin’s induction
ceremony into the Scientists’ Hall of Fame is sacrosanct. He is the principal
founding father of one of the great success stories in modern biology: the
theory of evolution by natural selection. At the same time, it comes as no
surprise that much of what we refer to as “evolutionary theory” today, in-
cluding what is standardly called “Darwinism,” is absent from Darwin’s orig-
inal writings. For instance, the great English naturalist is candid in the Origin
about his ignorance regarding the mechanisms of inheritance and variation.
And his subsequent speculative hypotheses purporting to fill in these blanks
were strikingly off the mark. Despite these evident shortcomings, Darwin’s
evolutionary explanations are far from deficient or misguided. How is this
possible? The key lies in Darwin’s deliberate decision to black-box the pro-
cesses underlying inheritance and variation. Their central role in evolu-
tionary theory was crystal clear. Yet, a systematic exploration of their nature
and structure had to be set to the side until much more was figured out.
My second illustration of black-boxing “in action” focused on Mendel’s
groundbreaking findings, which effectively treated genes as black boxes.
These results, rediscovered in 1900, were finally incorporated into main-
stream biology and brought to bear on phylogeny in the first half of the twen-
tieth century, with the forging of the Modern Synthesis. The grand union
of genetics and evolution effectively opened the black boxes of heredity
and variation, answering many questions left unanswered or inadequately
addressed by Darwin and Mendel. At the same time, it constructed new black
boxes, for instance, segregating the mechanisms of ontogeny. Issues set aside
by the architects of the Modern Synthesis have recently begun to be unrav-
eled, in the context of the so-called Developmental Synthesis. Along the
same lines, early molecular biology successfully opened Mendel’s black box,
the biochemical structure of genes, only to replace it with new placeholders.
With the steady growth of molecular techniques, the details underlying the
78 Black Boxes
regulation of gene expression and the principles governing the new “omics”
fields in biology, initially left untouched, started to emerge.
The third case history explored the history of psychology. Until the nine-
teenth century, psychologists typically considered mental states to be pri-
vate conscious phenomena not amenable to public scrutiny. In an attempt
to endow their discipline with a stricter scientific methodology, behaviorists
proposed that psychology restrict itself to seeking laws linking stimuli to be-
havior. The rationale was that only what is publicly observable is a fit subject
for science, essentially excluding mental states, as traditionally conceived,
from rigorous scientific examination. From our perspective, the black-
boxing of mental states enacted by behaviorism is interesting for two reasons.
First, it provides yet another clear and historically significant instance of the
widespread use of black boxes in scientific practice. Second, and more im-
portant, this methodological segregation of mental states highlights some
aspects of the general strategy that were not evident in the biological cases.
My final example came from the field of economics. In eighteenth-and
nineteenth-century Britain, economics and psychology were considered two
branches of a single subject: moral philosophy. Understanding the behavior
of markets presupposed a reasonably accurate psychological portrait of in-
dividual agents. It is thus unsurprising that seminal works by authors such
as Smith, Mill, Edgeworth, Marshall, and Pigou included discussions of how
mental states affect economic interactions. With the development of neoclas-
sical economics, however, feelings and other psychological states came to be
set aside, effectively isolating economics from fields such as psychology and,
later, neuroscience. Why did this occur? Simply put, feelings came to be per-
ceived as useless constructs, meant to predict behavior, but which could only
be inferred from behavior itself. In the 1940s, the concepts of ordinal utility
and revealed preference eliminated the superfluous intermediate step of pos-
iting unmeasurable feelings, equating unobserved preferences with observed
choices. The risk of circularity was avoided by requiring consistency in be-
havior. Once an agent reveals a preference for a over b, they ought not sub-
sequently choose b over a. This made the theory falsifiable, conforming
economics to positivist strictures. This “as if ” approach makes perfect sense
as long as the brain is considered a black box. Yet, following almost a century
of separation, economics has begun reimporting insights from psychology.
Behavioral economics is now a prominent fixture on the landscape and has
spawned applications to various areas of economics, such as game theory,
labor economics, public finance, law, and macroeconomics. Assessing
current debates regarding whether and how psychological and neuroscien-

tific discoveries can benefit economic theorizing requires understanding the
black-boxing of mental states during the neoclassical revolution. Are these
black boxes really in the process of being reopened? Should they be?
The case studies presented in this chapter play a dual role. First, they il-
lustrate the prominence of black-boxing across fields as diverse as evolu-
tionary theory, genetics, evo-devo, behavioral psychology, and economics.
One could say that the history of science is a history of black boxes, con-
tinually unwrapping some, while packing others (Figure 3.1). Second, they
show that, contrary to a common if tacit presupposition, black-boxing is not
a one-trick pony, a monolithic strategy that merely allows scientific research
to proceed in the face of our ignorance. Our excursus highlights that black
boxes can play various, subtly different roles across disciplines.
Darwin and Mendel set aside the principles of inheritance and variation
because they lacked accurate knowledge of the key mechanisms. This strategy
effectively postponed important issues that were intractable in their day and
age. Similarly, evo-devo’s ongoing attempt at cracking development requires
theoretical and experimental breakthroughs that had not yet emerged during
the forging of the Modern Synthesis. In all these cases, black-boxing is a way
of masking our ignorance for the sake of progress.
The maturation of psychology between the late 1800s and the late 1950s
tells a different tale. Behaviorism is an attempt to reconstruct an old and ven-
erable discipline on a more secure scientific foundation. This is a noble en-
deavor. The primary challenge that confronted Watson, Skinner, and their
BIOLOGY PSYCH & ECON
Figure 3.1. The history of science is a history of black boxes.

80 Black Boxes
followers was not overcoming lack of psychological knowledge. Their meth-

odological crusade was directed at a different target. They chiefly wanted to
show how it is possible to engage in an empirical study of the mind without
appealing to private, inaccessible, and unmeasurable properties. This require
bracketing mental states, as traditionally conceived. Contrary to Darwin
and Mendel, however, these black boxes were not meant to be set aside until
later times, when they could finally be opened. Behaviorism used the black-
boxing strategy as a trash can to dispose of concepts that would ultimately
muddle a rigorous, scientific study of the mind.
From a philosophical perspective, behaviorism can be viewed as an ap-
plication of Popper’s demarcation problem to a developing field, with the
aim of distilling a truly scientific methodology purged of pseudo-scientific
lures. The history of economics emphasizes how black-boxing can also serve
a different—but equally important—purpose: drawing boundaries between
disciplines, thereby protecting and insulating a field from “external” cri-
tique. Despite its flirtatious relation with behavioristic psychology and pos-
itivist philosophy of science, early neoclassical economics had little interest
in dismissing the study of mental properties as pseudo-scientific. The cen-
tral contention was that psychological states are not the appropriate subject
matter for economics. Allow me to illustrate the difference with a somewhat
colorful metaphor. Behaviorism treated the black box containing mental
states as a sort of Pandora’s Box that is better off hidden to avoid dangerous
temptations. Neoclassical economists, in contrast, considered that theo-
retical construction as a housewarming gift to be repackaged and shipped
across campus to their colleagues working in revamped or newly founded
departments of cognitive psychology and neuroscience. Incidentally, this
was also an effective and convenient strategy to protect one’s own favorite ap-
proach from rival theories and methodologies.
In conclusion, understanding the legacy of Darwin and Mendel in con-
temporary biology requires grasping why and how their black boxes were
constructed. The failure and consequent demise of behaviorism hinges on
the shortcomings of its attempt to dismiss mental states. Such conceptual
framework also sheds light on the prospects of psycho-neural economics
which, as noted, remain open and hotly debated issues.
One final remark. This chapter emphasized substantial differences in the
aim and structure of various episodes in the history of science. Given this
methodological hodgepodge, does it still make sense to talk about “the” prac-
tice of black-boxing in the singular? This entire book is based on the premise
that yes, it does make sense. Setting important divergences aside, all these re-
markable case studies from the history of science converge on the same core
phenomenon. This is the identification of some principle that, for various
reasons, is deemed dispensable from a set of explanations. It is now time to
delve in greater detail into the nature of this practice. How exactly do we con-
struct a black box? How do we determine what to include and what to leave
out? How does one judge whether the construction has achieved its intended
purpose?
The following chapters aim to provide a more general, precise, and sys-
tematic philosophical analysis of the practice underlying all the examples
described in the previous pages. A precise definition of black boxes will have
to wait until Chapter 5. First, we shall explore the foundations of an impor-
tant scientific construct: the concept of placeholder.
4
Placeholders
Mihi a docto doctore

Domandatur causam et rationem quare
Opium facit dormire.
A quoi respondeo,
Quia est in eo
Vertus dormitiva,
Cujus eat natura
Sensus assoupire.*
—Molière, Le Malade imaginaire
§4.1. Introduction
Where do we stand on our conceptual map? The main goal of this book is
to present and examine a form of explanation called “black-boxing.” Black
boxes, I contend, promise to reconcile the main insights of both reduc-
tionism and antireductionism, thereby revealing the productive role of ig-
norance in science. The historical excursus in Chapter 3 played a dual role.
First, it showed that the construction of black boxes is widespread across the
sciences. Second, it emphasized how this nuanced and complex practice may
accomplish a variety of functions. Darwin, Mendel, and the subsequent bi-
ological syntheses illustrate how black boxes effectively set aside questions
and problems that, for the time being, are intractable. Skinner and his fellow
radical behaviorists constructed black boxes to expunge entities, such as
mental states, that were deemed unsuitable for scientific investigation, in
spite of their hallowed place in traditional nineteenth-century psychology.
Modern economics tells a different story. Friedman and other neoclassical
* “The learned doctor asks me //The cause and reason why //Opium sends to sleep. To him I make
reply //Because there is in it //A virtue dormitive //The nature of which is //The senses to allay.”
Translated by Sir William Maddock Bayliss.
DOI: 10.1093/oso/9780190095482.003.0004
Placeholders 83
economists treated feelings, preferences, and other subjective values as an

inappropriate subject matter for their discipline. The rejection of mentalism
here becomes a fruitful strategy to draw boundaries, fine-tune the principles
of a particular field, and screen off rival methodologies. Setting these and
other significant differences to the side, all these endeavors are accomplished
by the same overarching strategy: the construction of black boxes.
Now that the stage is set, it is time to address more directly some issues
that have been lurking in the background. What is a black box? Under what
conditions is the introduction of a black box appropriate? How do these
constructs accomplish such a diverse set of tasks? Answering these questions
will involve a two-tiered approach. The present chapter provides some
foundations. The following Chapter 5 will apply these notions and mold
them into a general analysis, leading to an overarching definition.
Our starting point is a basic observation. The four historical illustrations
make clear that, at the most general level, black boxes are placeholders.
They are coarse- grained representations— pictorial depictions or verbal
descriptions—that stand in for entities whose nature and structure are either
unknown or deliberately set to the side. But what exactly are placeholders?
How are they introduced? What role do they play in science? This chapter
sharpens these questions and sketches some answers.
To keep the discussion focused, section 4.2 kicks off by introducing a con-
crete example: biological fitness. Specifically, I present two commonplace
theses. First, fitness is commonly defined as a dispositional property, namely,
the propensity of an organism or trait to survive in a specific environment.
Second, fitness attributions are coarse-grained characterizations of biolog-
ical properties that can be redescribed, more accurately and perspicuously,
at finer scales. Intuitive as they are, these two tenets are jointly in tension.
Section 4.3 spells out my argument, revealing its generality, and sketches a
way out of the impasse. The problem, I argue, is independent of fitness per
se. It affects any dispositional property that supervenes on a lower-level
basis. My solution, simply put, involves noting an ambiguity in the notion
of placeholder, which may replace two very different sets of entities. Section
4.4 clarifies my proposal with an array of illustrations, from mundane dis-
position ascriptions to scientific posits. Section 4.5 wraps up the discussion
by systematizing my distinction between two types of placeholders. Frames
stand in for a range of behaviors in need of explanation. Difference-makers
stand in for the mechanisms, broadly construed, that produce the behaviors
in question.
84 Black Boxes
Before getting on with it, a quick note to readers. We are now embarking
on a foundational analysis of black-boxing that will keep us busy until the
end of Chapter 5. Empirically oriented scholars with little interest in phil-
osophical discussions should consider glancing over the next two sections
and focus on the illustrations and systematization in the second half of the
chapter. Still, I should stress that abstract discussions of dispositional proper-
ties and explanations are crucial to fully comprehend the role of placeholders
in science.
§4.2. Two Theses about Fitness
Here is a motivating example, to get us started. My analysis of placeholders

begins with a short detour into contemporary biology. Fitness is a core ingre-
dient of Darwin’s framework and has remained a cornerstone of the theory of
evolution ever since. As we saw in section 3.2 of Chapter 3, for a population
to evolve by natural selection, it must contain heritable variation in fitness.
The reason is simple. Selection fuels evolution in two different ways. It may
affect an organism’s viability—its ability to survive—or its fertility, that is, its
capacity to reproduce. When members of a population differ along either
(or both) dimensions, and such differences are heritable, at least in part, that
population can evolve by natural selection. But what exactly is fitness? And
what role does fitness play in evolutionary explanations? I now take up both
questions, in turn.
Can we provide a precise definition of fitness? An intuitive way to ap-
proach this task is to ask: how does one determine the relative adaptedness of
organisms or traits in a population? Simply put, there are two main options.
One is to observe and compare actual frequencies. If individuals bearing
trait a survive to adulthood more often or typically leave behind more off-
spring than organisms bearing trait b, this is evidence that a is fitter than
b.1 Alternatively, one can examine the physical constitution of the organisms
under scrutiny. Occasionally, this merely requires educated common sense.
Noting that rabbits are preyed on by foxes, which are excellent runners,
suggests the (accurate) conjecture that speed and stamina contribute to the
1 Strictly speaking, talking about the “fitness” of x is ambiguous, as it may refer either to the fitness
of trait x or to the fitness of organisms bearing trait x. Nevertheless, for the sake of brevity, I shall often
make unqualified references to the “fitness of x,” leaving it up to the reader to determine the appro-
priate interpretation, depending on the context.
Placeholders 85
fitness of both species. In other cases, determining which anatomical charac-

teristics will make organisms “fitter” necessitates more sophisticated forms
of theorizing, such as mathematical modeling.
Discerning frequency patterns and studying physical makeup are effective
ways of measuring fitness, which provide useful proxies. Yet, a little reflection
shows that neither strategy constitutes an adequate definition of fitness. In
order to see why this is so, it is instructive to address the limitations of both
proposals.2
The futility of any general definition of fitness as a function of traits
should be fairly obvious. The fittest Bacteriophage λ, the fittest Drosophila
melanogaster, and the fittest Homo sapiens share no physical characteristic
that sets them apart from less-fit conspecifics. Survival and reproductive suc-
cess are heavily context-dependent. Different adaptations will be beneficial
across organisms and, even when we focus on a single species, one and the
same trait may be advantageous in some environments and detrimental in
others. In short, what all fittest organisms have in common is surely not any
feature of their physical makeup. What about the alternative route?
The shortcomings of defining fitness based on actual patterns of survival
and reproduction are subtler. One problem is that this encourages the charges
of circularity raised by foes of evolution at least since Spencer coined the—
somewhat misleading—expression “survival of the fittest.” The issue here is
that natural selection is often presented by the slogan that organisms who
survive are fitter. One can then legitimately ask: which individuals are the
fitter ones? If the reply is that fitter organisms are those who survive longer
and leave more offspring behind, we get stuck in a vicious circle. From this
standpoint, it cannot fail to be the case that the organisms presently on the
planet are the fitter ones, turning the principle of natural selection into an an-
alytic truth, as opposed to an empirical discovery.
A more troubling difficulty with the analysis presently under considera-
tion pertains directly to biological practice. Identifying fitness with actual
patterns of survival and reproduction fails to distinguish the consequences of
bona fide differences in viability and fertility from the outcome of spurious
accidents, like genetic drift, or even brute luck. To illustrate, assume that p-
organisms are better adapted to their environment than q-ones. For instance,
they are stronger, faster, energetically more efficient, and better at avoiding
2 The philosophical overview of fitness throughout this section draws from influential theoretical
discussions, such as Godfrey-Smith (2014) and, especially, Sober (1984, 2000).
86 Black Boxes
predation. For this reason, p-variants are more prevalent, in the envisioned
ecosystem, than q-types. Still, as a result of a wildfire, a large portion of the
p-population is wiped out. In this hypothetical scenario, more q-organisms
survive than p ones, and will therefore leave behind more offspring, at least
for a few generations. Yet, this is due to casualty and not—as the defini-
tion of fitness as actual survival and reproduction patterns implies—due to
differences in fitness between p-types and q-types.
The extent to which these theoretical considerations affect scientific prac-
tice is debatable. Working biologists often employ actual frequency patterns
as a proxy for fitness, and this seems to work reasonably well. Is there re-
ally a pressing need for change? At the field-work level, the answer may well
be negative. Things are different, however, from a pedagogical perspective.
Here, more accurate and perspicuous definitions of biological fitness, which
overcome the conceptual limitations just rehearsed, contribute to making
the content of evolutionary theory crisper and clearer.
So, what is fitness? Can we find more adequate definitions? Various
proposals have been developed in the literature. An influential one is the
“propensity interpretation of fitness.”3 On this view, fitness refers neither to
the actual number of offspring spawned by an individual or trait relative to a
reference point, nor to the physical constitution of organisms. Moreover, it is
not assumed as a primitive, undefined term of evolutionary theory. Rather,
in the words of its main proponents, “fitness can be regarded as a complex
dispositional property of organisms. Roughly speaking, the fitness of an or-
ganism is its propensity to survive and reproduce in a particularly specified
environment and population” (Mills and Beatty 1979, p. 270).
The propensity interpretation is not devoid of controversy. For one, pro-
viding precise mathematical characterizations of an organism or trait’s ex-
pected number of offspring is no trivial task. Furthermore, ascribing an
expected number of offspring to members of a population is not necessarily
a reliable predictor of evolutionary change. Nonetheless, let us ignore these
difficulties. Mills and Beatty’s dispositional interpretation of fitness does mit-
igate the problematic implications mentioned earlier. Readers who find this
analysis misguided are advised to skip to other examples in the following. As
we shall see, fitness is just an instance of a much broader phenomenon.
With all of this in mind, set definitions of fitness aside and focus on the
second issue raised at the outset. What role does fitness play in evolutionary
3 For a rival approach, which I shall not discuss here, see Rosenberg (1983).
Placeholders 87
theory? How does it contribute to biological explanations? To make these

questions more concrete, consider an example originally developed by
Sober. Envision a hypothetical scenario where, over the course of a year, two
chromosome inversions change frequency in a population of Drosophila
melanogaster, as a result of selection pressures. What could be going on?
The variation at the population level can be explained, at least provision-
ally, by positing that one type has greater fitness than the other. This, how-
ever, is hardly the end of the story. As researchers further inquire into the
physical basis of these differences in frequency, they might discover that
the chromosome in question produces a thicker thorax, which enhances
the fly’s insulation, making it more resistant to rigid temperatures. We now
have two different accounts of the evolution of the Drosophila population.
The higher-level depiction explains the change in frequency of the two chro-
mosome inversions by appealing to differences in fitness between the two
types. The lower-level characterization captures the evolutionary trajectory
by describing the physiological effects of the chromosome inversion. What is
the relation between these two explanations? Sober suggests:
Once [a detailed] physical characterization is obtained, we no longer need

to use the word “fitness” to explain why the traits changed frequency. The
fitness concept provided our initial explanation, but the physical details
provide a deeper one. This does not mean that the first account was en-
tirely unexplanatory. Fitness is not the empty idea of a dormitive virtue. The
point is that although fitness is explanatory, it seems to be a placeholder for
a deeper account that dispenses with the concept of fitness. (Sober 2000,
pp. 75–76)
What are we to make of this? Intuitively, there is an asymmetry between an

organism’s fitness and the underlying physical properties. (Here, the expres-
sion “physical property” should be interpreted quite generally, encompassing
not only the ontogenetic characteristics intrinsic of organisms, but also the
phylogenetic, environmental, and other relational conditions that are rele-
vant to the individual in question.) The asymmetry can now be stated along
the following lines. Physical properties, thus broadly construed, determine
how fit organisms are, in the sense that two individuals who are physically
identical must also be identical with respect to their fitness value. In contrast,
the converse relation typically does not hold. The viability or fertility of an
organism does not constrain its constitution. Individuals who share the same
88 Black Boxes
expected life span or an identical propensity to beget a particular number of

offspring need not, and usually do not, share all their ontogenetic and envi-
ronmental properties. We can express this asymmetry in contemporary phil-
osophical jargon by claiming that the fitness of an organism supervenes on its
physical base.
Let us return to our original question. What is the relation between
“higher” and “lower” explanatory levels? Sober provides an effective met-
aphor when he characterizes fitness as a placeholder. Fitness stands in for
ontogenetic, phylogenetic, and environmental processes that are responsible
for producing the appropriate kind of variation. In other words, talking about
fitness is a convenient way to refer to the physical properties that underlie
the viability and fertility of organisms. As noted in our discussion of Darwin
and Mendel, this can be done regardless of whether the detailed structure of
the underlying mechanisms is known and can be stated explicitly. Once the
nature of the underlying mechanisms is revealed, there remains an—open—
question of whether the placeholder should be discarded or retained.
Time to tie up some loose ends. The present section introduced two wide-
spread and relatively uncontroversial claims regarding biological fitness.
First, fitness is commonly defined as a dispositional property, namely, the
propensity of an organism or trait to survive and reproduce in a partic-
ular environment. The second tenet states that, since fitness supervenes on
the underlying physical base, it is a placeholder for a deeper analysis that
dispenses with the concept of fitness altogether. So far so good.
Here is the rub. Intuitive as they may seem, when considered separately,
these two theses, taken together, are in tension. Once we combine them,
they generate an apparent dilemma. Qua placeholder, fitness is explanatory.
It provides a rough causal explanation of relevant differences in viability
and fertility. In contrast, qua disposition, fitness does not convey any causal
knowledge. This is because dispositions explicate the underlying pattern of
behavior without causally explaining it.4 Section 4.3 lays out this argument in
full. I begin by clarifying the tension and showing that it is perfectly general,
affecting all dispositional or supervenient properties. Next, I suggest a solu-
tion that dispels the illusion of conflict and its paradoxical flavor.
4 Simply put, an explanation provides causal information, broadly construed, about how or why an
event, or pattern of events, occurs. An explication, in contrast, spells out a semantic analysis, a defini-
tion of the concept in question.
Placeholders 89
§4.3. Are Dispositions Explanatory?
Cashing out the argument presented at the end of the previous section
requires us to delve into the nature of a class of properties that has received
much attention in philosophy: dispositions. While, so far, I have appealed to
an intuitive understanding of dispositions, a succinct definition will make
the following discussion more rigorous. Dispositional properties express
the capacity, the ability, or the tendency of entities to act, under particular
circumstances. For instance, solubility captures how salt dissolves in water
and fragility underlies the tendency of glass to shatter when struck with
force. Mills and Beatty’s propensity analysis defines fitness along the same
lines. The fitness of organisms and traits, on their view, is a propensity to sur-
vive and reproduce, relative to specific environments. This section discusses
whether and how dispositions contribute to scientific explanations.
It is common to presuppose, more or less explicitly, that dispositions
causally explain the behavior of entities that satisfy them. Indeed, many
philosophers would consider this a truism to be taken for granted.5 Is the
solubility of salt not the reason that salt dissolves in water? Does a dropped
vase not shatter because it is fragile? From this perspective, it is hardly sur-
prising that, although Sober treats fitness as a placeholder standing in for,
and replaceable by, a deeper analysis, he distinguishes it from the idea of a
dormitive virtue, that is, an empty, question-begging circular analysis.
Things, however, are not quite that simple. How exactly does a disposition
account for the behavior of the entities to which it is ascribed? What kind of
knowledge does it provide, and how does it convey it? On what basis should
the explanatory power of solubility, fragility, and fitness be distinguished
from the empty idea of virtus dormitiva? For the sake of simplicity, let us
begin by focusing on mundane examples: salt, glass, and the like. In due time,
I shall generalize the argument to scientific dispositions like fitness.
In his celebrated masterpiece, The Imaginary Invalid, the French play-
wright Molière mocks a group of physicians who purport to uncover the
sleep-inducing quality of opium by ascribing to it a virtus dormitiva. What
triggers the humorous effect in this goofy attempt? The issue cannot be fal-
sity. Opium, after all, does have the power to put people to sleep. The problem
is rather vacuity, complete lack of informativeness. Saying that substance
5 The reason, I argue elsewhere, is an assumption typically presupposed without further motiva-
tion, namely, that dispositions are properties of entities (Nathan 2015a).
90 Black Boxes
x has a dormitive virtue cannot explain why x puts people to sleep because
ascribing a dormitive virtue to x is just a fancier way of saying that x induces
sleep. The entailment is analytic—or quasi-analytic6—turning the alleged ex-
planation into a restatement. Molière skillfully plays with this intuitive idea
that paraphrases are explicative, not explanatory.
None of this is especially novel or controversial. The aspect of Molière’s in-
sight that I want to stress is how broadly applicable it is. Having a dormitive
virtue is being able to induce sleep. Thus, virtus dormitiva cannot explain
sleep-induction because having a virtus dormitiva is synonymous with
having the power to induce sleep. This being so, is an ascription of solubility
to x not to say that x has the capacity to dissolve? Is the meaning of fragility
not expressible as a tendency to shatter when struck? Answering in the pos-
itive entails that solubility cannot account for the dissolving of salt and shat-
tering is not causally explained by fragility. Setting aside the wit—or lack
thereof—these examples are structurally analogous to the mockery staged
by Molière in The Imaginary Invalid. Counterintuitive as it may sound, it
appears that none of these dispositions is explanatory after all, at least, not on
any causal reading of explanation.
What about scientific dispositions such as biological fitness? Recall that,
on the propensity interpretation, stating that a-flies are “fitter” than b-flies
means that the former have a higher propensity to survive and reproduce
than the latter. Then, can fitness attributions causally explain the distribu-
tion of traits in the population of Drosophila? Is fitness any more informative
6 The unorthodox expression “quasi-analytic” calls for elucidation, which will take us on a slight
detour into analytic philosophy. Readers with little interest for such technicalities are advised to skip
this footnote entirely. In the central decades of the twentieth century a close connection was noted
between dispositions and subjunctive conditionals. Thus, the fragility of a glass entails that said ob-
ject would shatter, if struck. This observation was generalized into a “simple conditional analysis”
of dispositions (SCA): x is disposed to D when C iff x would D if it were the case that C. Basic as it
may sound, this thesis was endorsed by eminent philosophers, including Ryle, Goodman, Quine,
and Mackie. However, there is now a widespread consensus that this analysis is fatally flawed be-
cause the connection between disposition and entailed conditional breaks down in cases of “finkish”
dispositions, where objects temporarily lose or acquire dispositions, or when the manifestation
of a disposition is “masked” or “mimicked.” The philosophical community is still divided on what
should replace the SCA. Some responded by replacing the “simple” conditional with a more sophisti-
cated one. Others, such as myself, attempted to salvage the SCA (Nathan 2015a). Some explored the
prospects of non-conditional analyses. Others abandoned altogether the search for explication in
favor of a non-reductive account of dispositions. An assessment of these routes lies beyond the scope
of this work. The point is that most authors accept the existence of a connection between dispositions
and subjunctives. How is this connection to be cashed out? Advocates of conditional analyses, in
the original or revised forms, will presumably maintain that the relation between dispositions and
associated behaviors is analytic. Naysayers will likely resist identifying such connection with full-
fledged analyticity. What is it then? For lack of a better term, I call this weaker form of entailment
“quasi-analytic.”
Placeholders 91
than solubility or fragility? Is any disposition at all different from a virtus

dormitiva? It seems hard to escape a negative answer to all these questions.
We can now recast the puzzle raised at the end of section 4.2 in the form
of a general dilemma. We began with two theses about fitness, widely ac-
cepted both in biology and in philosophy. First, since the fitness of organisms
and traits supervenes on its physical properties, fitness is a placeholder that
stands in for a more detailed, lower-level causal explanation. Second, fitness
is a dispositional property, as prescribed by the propensity interpretation.
In the first part of this section, we saw that these two tenets lead to oppo-
site conclusions. The argument is simple. If fitness is a placeholder standing
in for the underlying mechanisms that produce the relevant variation in
populations, then it becomes clear how it can partake in causal explanations.
Fitness is an abstract characterization of the causal factors that produce var-
iation, where the nature and structure of these mechanisms are left unspec-
ified. At the same time, conceived as a disposition, fitness cannot explain.
This is because, as emphasized by Molière, dispositions explicate, but do not
causally explain their associated behavior. This implies that fitness is both
explanatory and non-explanatory, an overt contradiction. Something must
have gone wrong. What could it be? To answer this question, it might help to
visualize the argument in a schematic form:
(i) Fitness is a propensity, a disposition of organisms (Mills and Beatty).

(ii) Dispositions are explicative, but not causally explanatory (Molière).
(iii) Fitness is a placeholder for mechanisms (supervenience).
(iv) Unlike dormitive virtues, placeholders are causally explanatory
(Sober).
(v) ∴ Fitness is both explanatory and non-explanatory (⊥).
The remainder of this section dispels the apparent paradox. I will stick to
common wisdom and argue that both theses are ultimately correct. Qua
placeholder, fitness is perfectly explanatory, in a causal sense. But fitness can
also be analyzed as a dispositional property and, as such, it provides no causal
explanation. Reconciling these two prima facie incompatible claims will re-
quire disambiguating the notion of a placeholder. I will do so by drawing a
distinction between two kinds of placeholders in scientific explanations.
Before doing so, let us explore a different route. Attentive readers will have
surely noted that, to avoid the contradiction (v), it is sufficient to reject one
92 Black Boxes
of the previous four statements. Could any of these assumptions be dropped?

To answer this question, we must gauge our alternatives.
The first option is to drop assumption (i), which is tantamount to denying
that fitness expresses a dispositional property. At first blush, this might look
like a painless opt-out. Have I not myself acknowledged that the propensity
interpretation of fitness has counterintuitive implications and is not univer-
sally embraced? Yet, upon further reflection, this solution turns out to be a
non-starter. True, biologists often define fitness in non-dispositional terms.
As noted in section 4.2, actual frequency patterns of survival and reproduc-
tion are often used as handy proxies. Still, this hardly addresses the problem
at hand, namely, the explanatory power of fitness attributions. Actually, if an-
ything, it exacerbates it. Allow me to explain. The tension we are trying to
dispel is generated by the observation that, if fitness is defined as a propen-
sity to survive and reproduce, then one cannot explain why a is fitter than b
by noting that a is fitter than b. This being so, defining fitter traits as those
that are prevalent in a population is surely not to make the claim that a is
fitter than b any more explanatory. Au contraire, it makes it fully analytic and,
therefore, less explanatory from a causal perspective. In short, so far, no one
has been able to provide a definition that makes fitness causally explanatory,
without identifying fitness with propensities, actual frequencies, or specific
traits. To make things worse, the problem is does not pertain to fitness per se.
The preceding dilemma can be generated whenever we have a supervenient
property, many of which are bound to be dispositional in nature. This puts us
right back where we started.
These considerations motivate a more promising strategy: drop assump-
tion (ii) and concede that dispositions are indeed causally explanatory. I have
no knock-down refutation of this route—KO punches do not come easy, in
either science or philosophy. Indeed, I will ultimately embrace a solution
that attributes a specific explanatory value to dispositions. Still, for the time
being, let me stress that allowing disposition ascriptions to causally explain
the behavior associated with them (quasi) analytically is a hard bullet to bite.
Now, many readers will have little trouble accepting that fitness explains var-
iation, that fragility explains shattering, or that solubility explains the disso-
lution. But, then, why does dormitive virtue not explain the sedative power
of opium? Critics will retort that the comparison between fitness and virtus
dormitiva is simply preposterous. Fitness captures an important property
of biological populations, a constitutive ingredient of natural selection and
evolutionary explanations. Dormitive virtue, in contrast, is a made-up term
Placeholders 93
coined by a witty playwright to mock pompous scholasticism in medicine.

The two are not on a par. I agree. Still, on what grounds should we draw this
line? What distinguishes natural kinds from cooked-up impostors? Without
a principled criterion, one cannot justify the claim that some dispositions are
explanatory while others are merely explicative.
What about sacrificing (iii)? Is fitness really a placeholder that stands in
for variation-producing mechanisms? Some readers might quibble with—
or flat out reject—the specifics of my own account of placeholders sketched
in the sections and chapters to follow. Yet, that fitness is a placeholder is
harder to deny. The reason is simple. Most, arguably all, biological proper-
ties like fitness supervene on a lower-level basis, which, as noted, means that
any change in fitness entails a corresponding change in physical constitu-
tion, but not vice versa. Hence, talk about fitness can always be replaced with
richer, more precise characterizations of underlying physical features. From
this standpoint, it is hard to avoid the conclusion that fitness ascriptions are
placeholders standing in for deeper, micro-depictions.
This leaves premise (iv). The final strategy to avoid the—allegedly—
contradictory conclusion is to accept that properties such as fitness are
placeholders, but to deny that placeholders are causally explanatory. This
gets us out of the frying pan and into the fire. Even if one could successfully
show that fitness does not explain variation, the general problem is far from
solved. Basically all properties that supervene on a lower-level basis can be
treated as placeholders. Is no supervenient property explanatory? Are we
really willing to bite this bullet and deny that biology, neuroscience, psy-
chology, economics, and all other special sciences provide any explanation,
unless they posit non-supervenient properties? What would such properties
be? This radical and unpalatable form of fundamentalism flies right in the
face of centuries of explanatory successes of the special sciences.
In sum, all four assumptions appear sound. Rejecting any one of them
comes at a high cost. At the same time, these statements jointly imply that
supervenient properties, such as fitness, are both explanatory and non-
explanatory, an overt contradiction. How should we proceed, then?
My proposal boils down to a “Hempel-Davidson-style” dispelling of the
appearance of the paradox.7 Upon further scrutiny, there is no contradic-
tion; all four premises are correct. Specifically, I defend the tenability of
7 This is a nod to Hempel (1945) and Davidson (1970), who present apparent paradoxes and pur-
port to resolve them by reconciling seemingly incompatible assumptions.
94 Black Boxes
two propositions. First, dispositions are explicative but not causally ex-
planatory. Dispositions are like dormitive virtues: they capture behaviors
without explaining them. Second, dispositions are higher-level properties
that stand in for unspecified mechanisms without being identical to them.
As such, I shall argue, they may provide bona fide causal explanations.
My strategy to enact this reconciliation and dispel the paradoxical flavor
is to diagnose an ambiguity in the notion of placeholder. Dispositions like
“fitness” and other expressions denoting supervenient properties may
stand in for two different sets of entities. On the one hand, they may take
the place of mechanisms that produce specific patterns of behavior. When
this is the case, the placeholder functions as a difference-maker. On the
other hand, higher-level properties may stand in for the target range of
behaviors that one is attempting to explain, in which case the placeholder
functions as a frame. It explicates and lays out these behaviors but does not
causally explain them.
Section 4.4 illustrates this frame vs. difference- maker distinction
with examples that should by familiar to most readers. Next, section
4.5 provides a more systematic characterization of these two types of
placeholders, which will provide the key for our general analysis of black
boxes in Chapter 5.
Before moving on, let me stress again that expressions such as “mech-
anism” and “behavior” should be interpreted in the broadest and most ecu-
menical sense. In the present context they are essentially blanket terms that
designate objects, processes, structures, physical interactions, and many
other kinds of entities and activities that function either as explanantia or
explananda in science. Also, as noted in Chapter 1, the idea of a property or
description being “higher vs. lower level” or “macro vs. micro” is always rela-
tive to a choice of level of explanation.
§4.4. Placeholders in Scientific Theory and Practice
The previous section argued that macro-properties supervening on micro-

bases may function as placeholders in two different ways. Qua “frames” they
stand in for patterns of behaviors to be explained. Qua “difference-makers”
they take the place of mechanisms that produce these patterns. Frames
are explicative. Difference-makers are causally explanatory. This section
Placeholders 95
illustrates the distinction with examples, beginning with a familiar disposi-

tional property.
§4.4.1. Solubility
Imagine your five-year-old niece querying you about what happened to that
handful of salt crystals that you just poured into a boiling pot of water. You
reply by promptly noting that salt is soluble and, as such, it dissolves in water.
What kind of question did she ask? And what sort of answer did you provide?
Note that the issue, as just posed, is ambiguous, as your young interlocutor
could be asking either of two questions. On the one hand, she could be in-
quiring into the behavior of salt. On the other, she could be wondering about
what produces the behavior in question. Accordingly, your appeal to solu-
bility provides two different, but equally legitimate types of answers.
First, suppose that your niece is vaguely familiar with salt. She knows
what salt is. She has seen salt crystals in shakers and tasted salt in food.
Yet, she does not know much about its behavior. What happens when one
immerses salt in water? Does it float? Does it explode? Does it become invis-
ible? Does anything happen at all? Appealing to solubility effectively answers
this question: salt dissolves in water, a piece of information which could also
be obtained by observing salt in water. Here, the ascription of solubility to
salt is perfectly informative without being causally explanatory in the least.
Solubility indicates the relevant behavior, what happens to salt, without
saying anything at all about why or how the behavior itself occurs.
Next, consider a variant of this scenario where your young niece has an
altogether different question in mind. She has learned in school that certain
substances, such as salt and sugar, dissolve in water. She has also been shown
that not all substances are water-soluble. Glass, sand, and plastic, for in-
stance, are not. Intrigued by these remarks, she would like to understand this
difference in behavior. Note that both her question and your answer may be
phrased in exactly in the same terms. “What happened to salt in water? Salt
is water-soluble: it dissolved.” Yet, the implicit contrast class is completely
different. In the former case, she wants to know that salt dissolves in water, as
opposed to floating, exploding, changing color, remaining unaltered, etc. In
this latter case, she wonders in virtue of what salt and sugar typically dissolve
in water, whereas sand, glass, and plastic do not. Hence, pointing out that salt
dissolves will not cut it. We need to say something about why or how it does.
96 Black Boxes
A complete description of the sophisticated biochemistry of salt would be

inappropriate for a child. Yet, pointing out that salt is soluble, that is, it has
mechanisms that make it dissolve in water, is a much simpler, but still infor-
mative explanation.
In short, the same statement—the ascription of “solubility” to salt—may
provide an appropriate answer to two kinds of inquiry. First, it may specify
that salt dissolves in water. Second, it may explain how or why salt behaves
the way it does. In both cases, solubility is a placeholder. But it is a placeholder
of a very different kind. In the former case, “solubility” stands in for a pattern
of behavior: it describes salt as dissolving in water. In the latter case, “solu-
bility” stands in for the physical or chemical processes triggering the produc-
tion of said behavior: the mechanisms that make salt dissolve in water. Let us
apply the terminology introduced at the end of the previous section. When
solubility stands in for patterns of behavior, it functions as a frame. When it
stands in for mechanisms, it is a difference-maker, which provides a higher-
level causal explanation of the behavior in question.
Some readers might feel inclined to dismiss the use of solubility qua frame
as having marginal significance. This would be a mistake, fueled by the perni-
cious tendency among contemporary philosophers to privilege explanation
as the paramount scientific goal, at the expense of all others.8 Neither kind
of placeholder is primary. Sometimes, an inquiry is advanced by refining the
range of behaviors under scrutiny. Other times, what is sought is an expla-
nation of these patterns. Explanation is an important scientific goal. Yet, it is
hardly the only, or the most prominent one.
These considerations will be further developed in the ensuing chapters. In
the meantime, let us elucidate our distinction between placeholders—frames
vs. difference-makers—with some actual scientific examples.
§4.4.2. Fitness
We are now in a position to address the puzzle introduced in section 4.2. The
stage was set by outlining two widely accepted theses. First, fitness is typi-
cally defined as a dispositional property, namely, a propensity of organisms
8 Interestingly, prominent philosophers of science, such as Hempel (1965), explicitly warned

against the perils of this asymmetry. The disproportional attention currently devoted to explana-
tion at the expense of prediction, description, and other scientific aims witnesses the extent to which
Hempel’s warning has been left unheeded.
Placeholders 97
or traits to survive and reproduce in specific environments. Second, as Sober

aptly puts it, the uncontroversial observation that the fitness of an organism
or trait supervenes on its physical basis suggests that “although fitness is ex-
planatory, it seems to be a placeholder for a deeper account that dispenses
with the concept of fitness” (2000, p. 76). We noted how these two theses
are in tension. Prima facie, they imply that macro-properties such as fitness
are both explanatory and non-explanatory. I suggested that the contradic-
tion can be avoided by disambiguating two senses in which fitness is a place-
holder. On the one hand, it may stand in for patterns of variation within a
population, in which case it is explicative but not causally explanatory. On
the other hand, it may stand in for the mechanisms that produce the rele-
vant variation within a population, making appeals to fitness explanatory. It
is now time to elucidate the proposal in greater detail.
To make the discussion more concrete, recall Sober’s hypothetical sce-
nario, where two chromosome inversions change frequency in a population
of fruit flies over the course of a year, as a result of selection pressures. Here,
the statement “a is fitter than b” may address two different inquiries.
First, one may be looking into which of the two genotypes, a or b, is fitter.
In this case, the statement “a is fitter than b” can be read as claiming that flies
with genotype a have a tendency to survive longer and to beget more off-
spring than flies with genotype b. Thus understood, the difference in fitness
between the two genotypes does not explain how or why these distribution
patterns are observed, any more than virtus dormitiva explains the sedative
power of opium. Yet, this hardly makes such ascriptions trivial, vacuous, or
uninformative. On the contrary, pointing out that a-type flies are fitter than
b-type ones makes an important contribution. It is an effective way to set up
the explanandum, that is, to lay out the observed patterns of variation in the
population of fruit flies, which is in need of explanation. Still, this explica-
tion, in and of itself, lacks causal explanatory force.
The statement “a is fitter than b” may also address a second kind of inquiry.
Suppose that, a year into the study, the number of type-a flies has increased
dramatically, whereas type-b flies significantly decreased—literally dropping
like flies. One can explain this observation by pointing out a difference in
fitness between a and b. Once again, fitness acts as a placeholder. Yet, here, it
cannot stand in for patterns of variation. This would make the analysis unin-
formative, as these patterns are already presupposed in the question. Fitness
now stands in for the processes responsible for producing the variation in
question. In our example, these are the mechanisms, broadly construed, that
98 Black Boxes
allow flies with genotype a to produce a thicker thorax, providing better in-
sulation during colder winter months.
In sum, fitness can be defined dispositionally, as the propensity of
organisms and traits to survive and reproduce. Thus construed, fitness may
function as two distinct kinds of placeholders. First, it may stand in for dis-
tribution patterns. This lays out the explanandum by stating that traits vary
and by how much, without addressing why they do. Second, fitness can also
act as a placeholder standing in for the properties that underlie the produc-
tion of the traits in question. As noted by Sober, this shallower explanation
can be eliminated and replaced by a deeper account, which dispenses with
the concept of fitness altogether. The bottom line, once again, is that there are
two kinds of placeholders: frames and difference-makers. As we shall see in
Chapter 6, keeping them distinct is the key to understanding the role of fit-
ness in biological explanations, from Darwin to our day and age.
§4.4.3. Psychological States
Both examples so far involve either a textbook disposition (solubility) or a

concept that can be analyzed dispositionally (fitness). What about macro-
properties that do not obviously fall in either category? The distinction be-
tween kinds of placeholders— frames vs. difference- makers— applies to
supervenient properties of all sorts, regardless of whether or not they are dis-
positional. To establish this, I return to our discussion of psychological states.
What role do mental states play in psychology? Since the demise of be-
haviorism, many philosophers and theoretically minded psychologists have
rejoined forces with Descartes’s commonsensical stance in treating mental
states as the principal causes of behavior. Mary’s hunger plays an impor-
tant part in why she is going to the kitchen to fix herself a sandwich. And
Jack’s being upset with Jill is one of the reasons why he refuses to call her.
Now, to be sure, various theories of mind vehemently disagree over the na-
ture and structure of these psychological states. Functionalists, type-identity
theorists, token-identity theorists, and property dualists—to mention a few
influential candidates—offer very different analyses of the ontology and epis-
temology of mental states. Yet, contemporary psychology and philosophy of
mind agree that, contrary to the strictures of behaviorism discussed in sec-
tion 3.4 of Chapter 3, mental states contribute to triggering human conduct.
Placeholders 99
I have no quibbles with any of this. Still, I do want to emphasize that this is
only part of the story. Mental states play two distinct and equally important
roles within psychological investigations. On the one hand, as just noted,
mental states may function as difference-makers in the production of beha-
vior. On the other hand, mental states may be frames standing in for patterns
of behaviors to be explained. Allow me to clarify.
Imagine that Taylor and Dana spend a lot of time together. They make
dinner plans. They go out dancing. They talk on the phone daily. And they
often blush in the presence of one another. All these behaviors—which note,
constitute a wildly heterogeneous bunch—can be explained, at least provi-
sionally, by positing that Taylor and Dana are in love with each other. Could
one provide lower-level explanations of these phenomena? Of course! For in-
stance, we could provide a detailed account of how, in the presence of Taylor,
Dana’s body releases adrenaline, speeding up heart rate and dilating blood
vessels, improving blood flow and oxygen delivery in facial veins. We now
have two explanations of Dana’s blushing: a higher-level one and a lower-
level one. Does this story have a familiar ring to it? If so, it is likely because it
mirrors our discussion of fitness from section 4.2.
The precise relation between these two explanations is a controversial
matter. On the one hand, reductionists will argue that once a more detailed
physiological explanation is obtained, there is no need to use the expres-
sion “being in love” to account for blushing. The mental state can be reduced
to a deeper, lower-level analysis, thereby eliminating it. Antireductionists
will retort that no such reduction can be enacted. We shall return to the re-
ductionism vs. antireductionism debate in Chapter 10. For the time being,
the issue is that both stances can be reconciled with the claim that mental
states are placeholders, difference-makers standing in for mechanisms pro-
ducing behavior. The point of contention is the nature and character of
the mechanisms explaining the behavior. Reductionists will characterize
them at lower levels—brain processes, environmental stimuli, and the like.
Antireductionists will provide higher-level characterizations, appealing to
functional descriptions, qualia, or other macro-descriptions. But the mental-
states-qua-placeholder thesis is perfectly consistent with both stances, in-
cluding radical forms of materialism and a hard-core dualism rejecting
mind-body supervenience.
There is also, however, a different way in which “being in love” is a place-
holder. It can function as a frame, that is, it may stand in for a range of
behaviors that we are trying to explain. To illustrate, imagine that a team of
100 Black Boxes
psychologists sets out to investigate Taylor’s psychological profile. As any

ordinary human agent, Taylor engages in a vast array of behaviors, ranging
from breathing, to eating, to thinking, to working, etc. How do we restrict the
range of actions that our research team is going to study? An effective way
to do so is to focus on the behaviors displayed by Taylor in virtue of being
in love with Dana. In this sense, observing that Taylor is in love with Dana
does not explain why they are out dancing tonight, any more than ascribing
solubility to salt explains why salt dissolves in water rather than exploding,
or why the dormitive virtue explains why opium puts people to sleep as op-
posed to killing them. Nevertheless, loving-qua-frame plays a prominent,
non-explanatory role in psychological inquiries. It captures, in a clear, pre-
cise, and succinct fashion, the range of actions that we are trying to explain,
that is, how the agent in question behaves.
In conclusion, mental states are not dispositions—at least, they are not ob-
viously so. Yet, they are placeholders which may be used either as frames, to
capture how an agent behaves, or as difference-makers to provide a higher-
level causal explanation of the conduct itself. These considerations suggest
a more charitable reinterpretation of behaviorism. Pace Ryle and (early)
Wittgenstein, the manifestation of behavior, no matter how nuanced, is
neither necessary nor sufficient for mental states. Pace Skinner, the causal
explanation of behavioral patterns cannot be simply reduced to stimulus
conditions supplemented by operant behavior. Mental states are higher-level
causes of behavior. But this is only part of the story. And behaviorism was
largely correct about the other half. The evident connection between mind
and conduct allows us to use mental states as frames that capture an agent’s
complex behavioral patterns in need of explanation.
§4.4.4. Economic Utility
The distinction between frames and difference-makers also sheds light on

the role of utility in economic theory. Suppose that, on several occasions,
I observe Bill choosing apples over oranges, regardless of price fluctuations
between the two items. Economists from various traditions capture this data
by positing that Bill prefers apples to oranges in the following sense. With
diminishing marginal utility, the utility Bill receives from his first unit of
apples consumed is greater than the utility he receives from his first orange
unit consumed. His second unit of apples will bring him less marginal utility
Placeholders 101
than his first. (Whether this amount of utility is less than the utility he receives
from his first unit of oranges is a separate question, that we shall set aside.)
In general, it is assumed that Bill will consume a mixture of these fruits, and
he will do so until MUx/px = MUy/py for all x and y to maximize total utility
obtained. Without getting into technicalities, we can simplify this by positing
that Bill assigns a higher utility to apples than oranges: u(a) > u(o), and these
utilities can only be ranked, as opposed to quantified precisely. Note that this
claim is affected by the same kind of ambiguity that characterizes all previous
examples. On the one hand, the proposition that u(a) > u(o) may function as
a frame. In this former sense, it is a placeholder for the relevant pattern of be-
havior in need of explanation. On the other hand, that same proposition may
act as a difference-maker. In this latter sense, it is a placeholder standing in
for mechanisms that underlie and produce these observable patterns of pos-
sible and actual behaviors. How does this work?
First, imagine that the goal is to explain why Bill consistently prefers apples
to oranges. What is going on in Bill’s head when he is presented with this
basic choice? The notion of utility provides a preliminary answer: Bill assigns
a higher utility to apples than oranges. Note that, in the expression u(a) >
u(o), “u” is a placeholder. It stands in for whatever psychological mechanism
underlies the choice at hand. Just as we saw in the cases of solubility, fitness,
and mental states, this coarse-grained depiction can be replaced by a richer
micro-explanation that makes no reference to utility, but only describes the
psycho-neural processes that trigger the behavior in question. In a nutshell,
in this first instance, utility is a difference-maker.
Next, consider a slightly different scenario. We embark in a systematic
study of Bill’s fruit-related behavior. On day one, Bill is presented with a
choice between apples and oranges, and he selects the former. The following
morning, confronted with a similar range of options, he still picks apples
over oranges. On day three, the store is out of apples, and Bill picks pears over
oranges. On day four, the store is out of oranges, and Bill selects apples over
pears. It should be obvious that this depiction is somewhat simplified. Real-
life agents face choice selections that are exponentially more complex and
involve numerous other variables. Nonetheless, this basic scenario should
suffice to establish the main point. Can we systematize all these behaviors
in a coherent fashion? Utility provides a simple and convenient way of doing
so. Consider the proposition that u(a) > u(p) > u(o), which states that Bill
assigns a higher marginal utility to apples than pears and that his utility for
pears is higher than the value assigned to oranges. This simple claim allows
102 Black Boxes
us to make lots of predictions and retrodictions of Bill’s behavior pertaining

to fruit choice. Importantly, this is not a causal explanation of behavior. Yet,
this frame for the behavior to be explained is an important endeavor, none-
theless. It answers the that but not the how or the why.
Before moving on, a historical clarification is in order. Friedman, Savage,
and other early neoclassical economists who worked within the milieu of
logical positivism endorsed a covering-law (C-L), or deductive-nomological
(D-N) approach to explanation. On this model—which identified explana-
tion with logical derivation from a set of laws and initial conditions, along
the lines discussed in section 2.3 of Chapter 2—prediction and explanation
are two sides of the same coin. If Bill’s behavior can be accurately predicted
from the preceding ranking of utility, then it has also thereby been explained.
Contemporary philosophy of science rejects the C-L model of explanation,
for well-known reasons briefly rehearsed in Chapter 5. Yet, virtually all con-
temporary accounts explicitly eschew the identification of prediction and
explanation. As a result, they relinquish the idea that assignments of utility
provide an explanation and, especially, a causal explanation of the resulting
behavior.
The distinction between these two notions of utility—frame vs. difference-
maker—can be used to characterize, more precisely, the relation between dif-
ferent approaches to economics. Mainstream neoclassical economics (NCE)
treats utility, marginal or total, as an abstract mathematical construct: an or-
dinal ranking fully revealed by preference, with no cardinal counterpart. In
contrast, much contemporary psycho-neural economics (PNE) views utility
as an actual neural value, instantiated in the brain and quantitatively meas-
urable, at least in principle. Despite sharing the name “utility,” these two
concepts are importantly distinct, as they presuppose different assumptions
and methodological approaches. These considerations present an obvious
quandary. Given these evident discrepancies, why would neoclassical and
psycho-neural economists view these two notions of utility as competing
alternatives?
My suggestion is simple. NCE and PNE disagree on how choice should be
explained. Neoclassicists believe that the actual process underlying decision-
making is irrelevant to economics. Psycho-neural economists consider these
same mechanisms paramount, relative to their epistemic goals. Thus, utility
is a different explanans across the two theories. They posit distinct difference-
makers. At the same time, at a very broad level of description, NCE and PNE
Placeholders 103
may converge in their use of utility as a frame, that is, on the kind of behavior
that falls within the domain of economics.
On this note, what counts as “economic” behavior? How does one pick
out the set of phenomena to be covered? The central aim of economics is
to explore the consequences of choice. Which choices? Whenever options
are characterized by differences in utility, such choices can be studied
from an economic standpoint. At this general level of description, NCE
and PNE converge on Robbins’s definition, presented in section 3.5 of
Chapter 3: “Economics is the science which studies human behaviour as a
relationship between ends and scarce means which have alternative uses.”
Utility frames the behavior to be explained by both approaches. How are
these patterns to be accounted for? Here the two theories diverge. One side
opts for an idealized “as if ” mathematical model. The other side cashes out
a mechanistic model. In sum, the two approaches may be subsumed under
the same frame, the same choice of object of explanation. But the underlying
selections of difference-makers are clearly distinct.
§4.4.5. Phlogiston
All the examples discussed so far involve supervenient properties—some

dispositional, others categorical—which are widely employed in everyday
life, as well as in scientific theory and practice. Our final illustration draws
attention to a different set of issues, by focusing on a superseded notion that
has been purged from our ontology: the concept of phlogiston.
Phlogiston theory was introduced in 1667 by J. J. Becher. Initially devel-
oped by G. E. Stahl and his student J. H. Pott, it received its most advanced
formulation toward the end of the eighteenth century in the work of Joseph
Priestley and Henry Cavendish. Such theory purports to provide a systematic
explanation of important chemical reactions, most prominently, the process
of combustion. Simply put, the term “phlogiston” was introduced to pick out
the substance, the “principle,” emitted in all cases of combustion. As it turns
out, there is no single substance that is emitted in all cases of combustion.
Consequently, the term “phlogiston” fails to refer—although, as will become
clear in the course of our discussion, this initial characterization is a crass
oversimplification. These well-known observations raise some interesting
problems for philosophical theories of reference and related issues con-
cerning the advancement of scientific theories. Specifically, if “phlogiston” is
104 Black Boxes
non-referential, then what explains the celebrated successes of Priestley and

his colleagues?
The issue of progress will be addressed, in detail, in Chapter 9. For the
time being, my goal is to show that superseded scientific concepts, such as
phlogiston, raise a kind of puzzle similar to the one presented in section 4.3.
From the standpoint of phlogiston theory, the expressions “combustible”
and “rich in phlogiston” are, for all intents and purposes, synonymous. The
reason should be evident. Since phlogiston is defined as the substance that
is released in the air during combustion, it follows that all and only com-
bustible substances contain phlogiston. Consequently, phlogiston looks like
solubility and fitness. While these dispositional properties explicate the un-
derlying concepts, they appear to have all the explanatory force of a virtus
dormitiva. This, once again, seems counterintuitive. Phlogiston is the epon-
ymous explanatory core of phlogiston theory, not a definition. Sure, con-
trary to fitness or solubility, it has been discarded. But how could an analytic
entailment—the stipulative identification of phlogiston-rich substances with
combustible ones—turn out to be empirically false?
Attentive readers will probably anticipate my response to this quandary,
which is analogous to my solution to the general puzzle. “Phlogiston” is a
placeholder. And there are two types of placeholders, depending on the kinds
of entities they stand in for.
To begin, recall that the chief goal of phlogiston theory is to explain the
process of combustion. Combustion generates familiar patterns, where
a substance is gradually transformed into another as a result of a chemical
reaction. Thus, burning transforms wood into ashes, while simultaneously
altering the composition of the surrounding air. Granite, in contrast, is not
combustible because burning it does not produce any equivalent reaction.
What role does the concept of phlogiston play in these explanations?
Here is the punchline. The statement “substance x is rich in phlogiston” is
ambiguous, in the sense that it provides an answer to two different questions.
On the one hand, it may specify the kind of behavior to be expected when
substance x is burned, namely, that it will combust. In this sense, “rich in
phlogiston” is synonymous with “combustible” in the same way that “sol-
uble” and “tends to dissolve in water” mean exactly the same thing. Just like
solubility, thus understood, cannot causally explain why salt dissolves in
water, the claim that wood is rich in phlogiston cannot capture why wood
combusts. The reason has nothing to do with the shortcomings of phlogiston
theory per se. It is because explications are not causally explanatory. In this
Placeholders 105
former sense, “phlogiston” is a frame. It picks out the class of events that the
theory is supposed to account for and, as such, it is explicative without being
explanatory.
On the other hand, “phlogiston” may refer to the principle, substance, or
process that causes and underlies the reaction itself. The theory posits the ex-
istence of a single principle, phlogiston, that is emitted in all cases of combus-
tion. In this second sense, “phlogiston” is employed as a difference-maker. It
picks out whatever it is that makes a difference to why and how the process
of combustion occurs. As such, it is perfectly explanatory. It just so happens
that it provides an inaccurate, superseded causal explanation.
In sum, qua frame, phlogiston theory provides valid explananda; we just
no longer use the terms of this old theory to describe them. Qua difference-
maker, it provides discarded explanantia, replaced by atomic chemistry.
Incidentally, why exactly has phlogiston theory been discarded? If “rich
in phlogiston” is synonymous with “combustible,” and combustion is a real
phenomenon in nature, then why was the concept of phlogiston eventually
eliminated and replaced by atomic chemistry? The obvious reason is that talk
about “phlogiston” is not perspicuous enough. In particular, it introduces
a nonexistent element, phlogiston. In doing so, it fails to address a number
of important distinctions. By positing a single substance that is emitted in
all cases of combustion, phlogiston theory clashes into the same category
of events which should be kept distinct, like the burning of wood and the
heating of metals. Unsurprisingly, modern atomic chemistry fares much
better on this score. By differentiating between, say, the combustion of wood
and coal and the oxidation of mercury, iron, and other metals, it provides
more precise causal-mechanistic explanations and picks out a more perspic-
uous range of behaviors. Still, these observations raise further issues. Why is
the transition from phlogiston theory to atomic chemistry a progressive one?
Addressing this and related questions requires a more systematic discussion
of black boxes. We will return to it in Chapter 9. Now, it is time to wrap up
our discussion of placeholders.
§4.5. Two Types of Placeholders
Time to tie up some loose ends. This chapter began by presenting the notion
of biological fitness. I outlined two widespread theses. First, fitness is com-
monly defined as a dispositional property. It is the propensity of an organism
106 Black Boxes
or trait to survive and reproduce in a particular environment. Second, since

fitness supervenes on its underlying physical properties, it is a placeholder
for a deeper account that dispenses with the concept of fitness altogether.
Plausible as they are, these two theses are in tension. Qua placeholder, fit-
ness is explanatory. Qua disposition, it explicates but cannot causally explain
the associated behavior. The rub is that dispositions are placeholders. Thus,
taken at face value, our intuitive premises generate an overt contradiction. In
the second part of the chapter, I suggested a way out of this impasse. My solu-
tion, simply put, involves disambiguating the notion of placeholder. On the
one hand, a placeholder may stand in for the range of events to be accounted
for. In this case, the placeholder functions as a frame: it spells out an expla-
nandum, a behavior, or range of behaviors in need of explanation. On the
other hand, a placeholder may stand in for the mechanisms, broadly con-
strued, which bring about the patterns of behavior specified by the frame, the
explanandum, regardless of how well their nature and structure are under-
stood. When this occurs, the placeholder becomes a difference-maker. These
types of placeholders can be visualized in Table 4.1, which distinguishes
them along two orthogonal axes: their role as explananda or explanantia,
and whether they stand in for mechanisms or behaviors. By drawing these
simple distinctions, all four premises in the argument from section 4.3 can be
maintained consistently.
We are finally in a position to resolve the general paradox introduced in
section 4.3. Are appeals to solubility, fitness, mental states, and utility caus-
ally explanatory? Short answer: it depends. As frames, these constructs
are not explanatory; they require explanation. As difference-makers, they
are perfectly explanatory. Thus, Sober is correct in noting that appealing
to fitness provides an explanation of changes in frequency patterns across
populations. At the same time, his attempt to contrast fitness with virtus
dormitiva is potentially misleading. Sure, when dormitive virtue stands in
for a range of behaviors—putting people to sleep—then it hardly explains
the powers of opium, or anything else. This confusion is what generates the
Table 4.1 Two Types of Placeholders in Science
Explanatory Role Placeholder Type Stands in for
Explanans Difference-maker Mechanisms

Explanandum Frame Behavior
Placeholders 107
comic effect skillfully crafted by Molière. Yet, when virtus dormitiva stands in
for the mechanisms that sedate people, appeals to dormitive virtue become
unproblematically explanatory. Sure, they might not be very enlightening.
Virtus dormitiva may be closer to solubility than fitness, mental states, or
utility. But a vague, general explanation is an explanation, nonetheless.
It is worth stressing that my notion of placeholder is quite broad, and in-
tentionally so. Specifically, placeholders may range from having much to very
little content. To wit, the higher-level properties discussed in this chapter—
solubility, fitness, mental states, utility, phlogiston, and the like—provide
a minimal specification of the structure of the mechanisms represented.
Nevertheless, if all supervenient properties are placeholders, it is also pos-
sible to provide much richer, more detailed, characterizations of the un-
derlying properties. This anticipates an important point to be developed in
Chapter 5. My definition of a black box, based on the dual role of placeholders
introduced in the present chapter, encompasses not only stereotypical “black
boxes,” but less opaque concepts, such as gray boxes, and quasi-transparent
boxes too, as long as these supervene on lower, physical levels.
In conclusion, how does all of this fit into the philosophy of science? What
implications does our analysis of placeholders have vis-à-vis the role and con-
tribution of higher-level properties to scientific inquiries? Many authors have
embedded these considerations into the broader context of the reductionism
vs. antireductionism debate. Sober, for one, generalizes his discussion of bio-
logical fitness to the conclusion that all properties in the special sciences su-
pervene on physical properties. This, in turn, raises an overarching question
about the nature of scientific explanation. If the special sciences supervene
on physics, is there a fundamental physical explanation for any phenomenon
explained by other sciences? If higher level properties are placeholders that
can be replaced by more detailed lower-level descriptions, are these micro-
level explanations always deeper than their macro-counterparts?9 These are
precisely the type of issues that we have previously encountered in Chapters 1
9 Sober explores two kinds of answers. First, he suggests that, while all non- fundamental
explanations retain practical value, in principle, they are all disposable. This seems like a modest
form of reductionism, since it acknowledges that, while physics is not yet able to explain everything,
in principle, it could. Next, he goes on to provide a second answer, which falls in line with sophisti-
cated antireductionism. Even if it would be possible to best explain all individual token events at the
most fundamental level, explanations of the type sort cannot be paraphrased at lower levels, either
in practice or in principle. For a similar argument, from the standpoint of physics, see Batterman
(2002).
108 Black Boxes
and 2. Once again, we are hopelessly haunted by the Homeric hazard. What is
our doom? Are we going to get devoured by Scylla or drowned by Charybdis?
By now, readers will not be surprised to hear that I intend to explore a dif-
ferent route. There is no need to pick your poison. We can have the cake and
eat it too. Specifically, reductionists are absolutely correct that it is always pos-
sible to replace a causal-mechanistic explanation of how and why a certain
phenomenon occurs with a more detailed, lower-level, micro-depiction. At
the same time, this does not invalidate the antireductionist tenet that many
explanations in the special sciences are “autonomous,” in the sense that they
stand alone perfectly well, without the need to enrich them with additional
micro-details. Showing how these seemingly incompatible claims can be rec-
onciled is the ultimate task of this book. Before getting there, we have ways to
go. Since my positive proposal will focus on black-boxing, the first move is to
elucidate this strategy. With this target in sight, the following chapter identi-
fies and discusses the constitutive stages of this form of explanation.
5
Black-Boxing 101
Oh, won’t you

Gimme three steps, gimme three steps, mister
Gimme three steps toward the door?
Gimme three steps, gimme three steps, mister
And you’ll never see me no more
—Lynyrd Skynyrd, “Gimme Three Steps”
§5.1. Introduction
Chapter 4 began a more systematic investigation of the black boxes first iden-
tified by our historical excursus in Chapter 3. Specifically, I introduced a dis-
tinction between two kinds of placeholders in science. Frames stand in for
explananda, patterns of behavior in need of explanation. Difference-makers
stand in for explanantia, mechanisms that produce and explain the patterns
in question. Both kinds of placeholders will play a pivotal role in my anal-
ysis of black boxes, developed with an eye on the following issues. What is a
black box and how is it constructed? How does a single concept accomplish
so many tasks, across a variety of fields? Can black boxes explain the inter-
play between productive ignorance and knowledge?
Here is the punchline. This chapter sets out to decompose the black-boxing
strategy into three constitutive phases. The first step involves sharpening the
explanandum by placing the object of explanation in the appropriate context.
This is typically accomplished by constructing a frame, a placeholder that
stands in for the pattern(s) to be covered. For this reason, I refer to this phase
as the framing stage. The second step, the difference-making stage, provides
a causal explanation of the explanandum, now appropriately framed. This,
simply put, involves identifying the relevant difference makers, that is,
placeholders standing in for the mechanisms that produce the patterns under
scrutiny. The third step, the representation stage, determines how these differ-
ence makers should be characterized, which features of the mechanism are to
DOI: 10.1093/oso/9780190095482.003.0005
110 Black Boxes
be explicitly represented and which can be idealized or abstracted away. The

outcome of this process is a model of the explanandum, a representation of
the target system or relevant portions of the world.
Taken together, these three steps provide the general definition we have
been looking for. A black box is a placeholder—frame or difference-maker—
in a causal explanation represented in a model. The goal of the sections
to follow is to elucidate this compact statement by breaking it down to its
main constituents. How is an explanandum framed (§5.2)? How is it sub-
sumed under a causal explanation (§5.3)? How is this causal explanation
represented in a model (§5.4)? With this definition on the table, our next
task, in Chapter 6, will be to apply the present analysis to our case histories,
before exploring, more systematically, its philosophical implications.
Before moving on, allow me a clarification and a word of caution.
Beginning with the former, I should stress from the outset that all three
stages are individually necessary and jointly sufficient for the effective con-
struction of a black box. Now, surely, it is not uncommon in science for one
of these phases to be performed in isolation from the others. To wit, so-called
exploratory experiments often complete the first two steps of the procedure
described earlier, without attempting to provide a mechanistic model of the
phenomena in question. Nevertheless, I maintain that the specification of a
black box requires all three steps. To emphasize, there is nothing inherently
wrong with research that only engages with some of these phases without
bringing all three of them to completion. Still, from the present standpoint,
the outcome of these inquiries falls short of a full-fledged black box. At the
same time, it would be a mistake to insist that these steps must invariably
be taken in any specific order. Scientific research is complex and messy, and
practitioners are often pushed into a project in medias res. Hence, my nar-
rative should not be understood as a realistic portrayal of laboratory life. It
is intended as a regulative ideal, a rational reconstruction of actual empir-
ical work.
Finally, here is my cautionary warning. The present chapter is admittedly
dense, rich in somewhat technical analysis, and it covers a vast philosophical
literature. There is a reason for this. The topics discussed here—explanatory
relativity, causal explanation, and modeling—have a hallowed place in con-
temporary philosophy. Yet, my rationale for inserting them in the appro-
priate context transcends my scholarly due diligence. These long-lasting
debates reveal why problems, commonly glossed over in much scientific lit-
erature, should not be taken for granted. A proper philosophical analysis of
Black-Boxing 101 111
science must pay attention to both its scientific and philosophical precursors.
Having said this, frustrating my audience is neither kind nor wise on my
part. Empirically minded readers with little interest in philosophical debates
may want to consider glossing over the detailed discussion of the three stages
of black-boxing, focus on the succinct summary in section 5.5, and dig into
the real-life scientific examples and applications in Chapter 6.
§5.2. Step One: The Framing Stage
Requests for explanation are seldom, if ever, unqualified. To wit, consider

a mundane question such as: why did the match light? What is the appro-
priate reply? It depends. Typically, it suffices to remark that the match lit be-
cause it was struck. Still, there are situations where this response misses the
mark. Imagine a legal setting where an attorney is reconstructing the chain of
events that led to a gas station blowing up. It has been established that the ex-
plosion was triggered by the burning of a match. While interrogating a sus-
pect, the attorney asks why the match lit. Replying that the match lit because
it was struck would be inappropriate, teetering on contempt of court. What
is in question here, presumably, is whether the match was struck deliber-
ately, with the intention to detonate, or whether it was struck accidentally by
a careless, non-malicious agent. Or think of a fictional NASA probe striking
a match while examining the chemical atmosphere of a distant planet. Here,
the appropriate answer to “why did the match light?” could be the sufficient
quantity of breathable oxygen in the atmosphere, an observation that would
be uninformative, even insulting, in the previous cases. The first scenario is
more usual. The other two are admittedly far-fetched. Still, they establish an
important point. Explanations come equipped, more or less explicitly, with
information that provides the context in which they should be understood,
processed, and assessed.
These basic remarks are hardly original. Both vintage and contempo-
rary discussions acknowledge this relativization of explanation and related
pitfalls. Before delving into the complexities of scientific framing, let us ex-
amine how instances of this phenomenon have been analyzed in philosophy.
A well-known illustration of explanatory relativity comes from a 1973
paper by Fred Dretske. Alex, after being fired, is short on cash. His friend
Clyde lends him $300. Dretske notes that there are (at least) three questions
that may be expressed by the string of words: “why did Clyde lend Alex
112 Black Boxes
$300?” Accordingly, three distinct answers may be appropriate. First, why

did Clyde lend Alex $300, as opposed to, say, $100 or $500? Second, why did
Clyde lend Alex $300, as opposed to gifting him the cash? Third, why did
Clyde lend Alex the cash, as opposed to their common friend Scrooge, who
is rolling in dough? Dretske notes that the difference in stress is a pragmatic
one, posing serious challenges to purely syntactic or semantic theories of
explanation. This is true. But the issue remains: how should this pragmatic
form of explanatory relativity be characterized and addressed?
An influential development of this pragmatic dimension of explanation is
provided by van Fraassen in his 1980 book, The Scientific Image, a contempo-
rary classic in the philosophy of science. The backdrop of van Fraassen’s anal-
ysis is constituted by the “problem of asymmetry,” the most widely debated
counterexample to Hempel’s “deductive-nomological” (D-N) model of ex-
planation, common among logical positivists.1 Imagine a tower 100 feet tall
casting a shadow 75 feet long. According to the D-N model, we can explain
the length of the shadow by deducing it from the height of the tower, in con-
junction with assumptions on the angle of elevation of the sun, basic princi-
ples of trigonometry, and light traveling in straight lines. The problem is that
these same background conditions equally allow one to deduce the height
of the tower from the length of the shadow. Now, no one seriously wants to
question that the tower’s height explains the shadow. But does the shadow’s
length explain the tower? Most philosophers answer in the negative, gener-
ating a counterexample to the D-N model, which is committed to an affirm-
ative answer.
All of this was well known by the 1980s. An explanation of the height of
the tower appealing to the length of the shadow would typically be rejected.
But does the shadow’s length never explain the tower’s height? Van Fraassen
casts doubt on this stronger conclusion. His memorable fable, “The Tower
and the Shadow,” skillfully crafts a scenario where the height of the tower is,
indeed, explained by the length of the shadow because of the builder’s de-
sire for the shadow to reach a specific spot at the right time of day. These
considerations are molded into a pragmatic analysis, according to which an
1 From this D-N perspective, also known as the “covering-law model,” the explanation of an event
consists in a formal derivation of an explanandum from a set of laws and initial conditions. Thus,
one can explain why the match lit by observing that, given initial conditions, such as the presence of
oxygen in the air, the structure of the match, and the striking of the match, together with the relevant
laws of nature, the explanandum event was logically bound to happen. The problem of asymmetry
was noted and presented, through a variety of equivalent examples, by various authors, especially
Michael Scriven and Sylvain Bromberger.
explanation is an answer to a question Q of the form “Why Pk?” Specifically,

explanatory requests are identified with ordered triplets <Pk, X, R>, where Pk
is the “topic” of the question, X = {P1 . . . , Pk, . . .} is its contrast class, and R is
the relevance relation of the explanation.
Is van Fraassen sketching a pragmatic theory of explanation or a theory
of the pragmatics of explanation? On the former, stronger reading—
characterized as a theory purporting to fully account for scientific explana-
tion in terms of pragmatically determined contextual features—the view is
notoriously subject to trivialization worries. Still, even critics who note these
shortcomings acknowledge the value of van Fraassen’s contribution when
understood as a theory of the pragmatics of explanation.2 The basic insight
can be paraphrased as noting the context-relativity of explanation when it
comes to assessing its intended target and, consequently, its success.
Yet another notable attempt to characterize this explanatory relativity can
be found in Garfinkel’s 1981 monograph, Forms of Explanation. When the
infamous bank robber Willie Sutton was in prison, he was allegedly visited by
a priest, who asked him why he robbed banks. “Well,” Sutton replied, “that’s
where the money is!” What is going on in this exchange? Simply put, the
bandit and the clergyman are talking past each other. This is because they are
tacitly presupposing different objects of explanation. The priest is wondering
about why Sutton would choose to rob at all, as opposed to conducting an
honest life. Sutton, in contrast, is addressing a different question. Given
his decision to rob, why should the target be a bank, as opposed to, say, a
public library or a department store? The two interlocutors have different
conceptions of the relevant options. They take the same string of words—
“why do you, Sutton, rob banks?”—but embed this statement in different
spaces of alternatives, effectively changing the explanandum. This triggers
the humorous effect. Incidentally, Garfinkel notes, this simple recipe has pro-
duced countless jokes for generations. Why do birds fly south for the winter?
Because it is too far to walk!
Setting jest aside, the philosophical point—explanatory relativity—is deep
and important. Explanations always take place relative to a background of
alternatives, a contrast space that determines the nature of the explanandum
2 As Kitcher and Salmon (1987, p. 316) put it in their scathing review of van Fraassen’s theory of
explanation, “Failing to appreciate that arguments are explanations [ . . . ] only relative to context, we
assess the explanatory merits of the derivations by tacitly supposing contexts that occur in everyday
life. With a little imagination, we can see that there are alternative contexts in which the argument we
dismiss would count as explanatory.”
114 Black Boxes
and whether the explanans is ultimately successful.3 In some cases, the sa-
lient alternatives may be stated explicitly. Why did you finish the pizza but
not the broccoli? More typically, the underlying contrast class is presupposed
at a tacit level. Dretske’s example is a case in point. Whether the question is
why Clyde lent Alex $300, why Clyde lent Alex $300, or why Clyde lent Alex
$300 is unspecified. It must be inferred from the context. Similarly, Sutton’s
remark that banks are where the money lies is a perfectly good explanation
of why Sutton robs banks, but only provided that his commitment to robbing
is assumed. It should be clear from the context that the priest is not willing to
take that premise for granted.
Time to take stock. All requests for explanation are subject to interpre-
tation. Even simple, ordinary, everyday questions—Why did the match
light? Why did Clyde lend Alex $300? Why does Sutton rob banks?—are
tacitly equipped with assumptions, “given that” clauses, and pragmatic
presuppositions that place them in the appropriate context so that they
can be parsed, processed, and addressed. In other words, the object of ex-
planation is never an event simpliciter. The explanandum is always an event
together with a contrast class, that is, a set of alternatives C1 . . . Cn that deter-
mine what counts as an acceptable explanation of E.4
3 What exactly is a contrast space? Garfinkel (1981, p. 40) describes its basic structure as follows. If
Q is some state of affairs, a contrast space for Q is a set of states [Qa] such that (i) Q is one of the Qa;
(ii) every Qa is incompatible with every other Qb; (iii) at least one element of the set must be true;
and (iv) all of the Qa have a common presupposition, that is, there is a P such that, for every Qa, Qa
entails P.
4 Why must explanations thus limit our options? The answer, Garfinkel suggests, “lies in our need
to have a limited negation, a determinate sense of what will count as the consequent’s “not” hap-
pening” (1981, p. 30). This point is elucidated with the help of an example. Suppose that one morning,
I go out for a drive. I am going over 110 mph when I round a bend where a truck has stalled. Unable to
stop in time, I crash into the truck. Later, you chastise me: if you hadn’t been speeding, you would not
have crashed. That is true, I retort. But then, had I not had breakfast, I would have reached that spot
before the truck. So, had I not eaten breakfast, I would not have crashed. Why do you not blame me
for having breakfast? What makes my reply fishy? Here is Garfinkel’s answer (1981, pp. 30–31): “My
claim is based on the assertion that if something (eating breakfast) had not happened, the accident
would not have happened. The problem is, what is going to count as that accident’s not happening? If
‘that accident’ means, as it must if my statement is going to be true, ‘that very accident,’ that concrete
particular, then everything about the situation is going to be necessary for it: the shirt I was wearing,
the kind of truck I hit, and so forth, since if any one of them had not occurred, it would not have
been that accident. But this is absurd, and in order to escape this absurdity and not have everything
be necessary for the accident, we must recognize that the real object of explanation is not my having
had that accident. [ . . . ] [T]‌he real object of explanation is an equivalence class under [a] relation
[determined by a set of ‘irrelevant’ perturbations]. The equivalence relation determines what is going
to count as the event’s not happening.” As Michael Strevens has brought to my attention, appealing
to blameworthiness as a test for causal or explanatory relevance is problematic, as these features do
not invariably go hand in hand. Causally explanatory factors like icy or oily roads may not attract
blame. Vice versa, non-explainers such as negligence sans adverse effects can be morally relevant.
Causal relevance will be discussed in detail in section 5.3. The important point, for the time being, is
Explanatory relativity is an old adage, widely noted and discussed in the

philosophical literature. Nevertheless, these considerations raise some in-
teresting theoretical issues for science, whose significance is frequently
overlooked. Spelling out the background assumptions, the contrast class, of
everyday explanations is daunting enough. It requires interpreting various
contextual cues and parsing tacit presuppositions. And, yet, the complexity
of lit matches, towers and shadows, and friends borrowing cash pales in com-
parison to actual scientific inquiries. Where does one even begin contextual-
izing explanations which presuppose evolution by natural selection, classical
genetics, behavioristic psychology, or neoclassical economics? Borrowing a
Firestein comment, originally directed to the constant growth of scientific
knowledge, and adapting it to a slightly different context: “How does anyone
even get started being a scientist?” (Firestein 2012, pp. 14–15).
To fathom the real depth of the problem, let us take a brief detour into the
nature of science itself. As noted in Chapter 1, textbooks have a tendency
to present theories as systematic arrays of laws and principles, obtained
through an organized accumulation of data sets, and a rational discussion of
their implications. This image is a drastic oversimplification, and hardly an
inconsequential one. Things in actual laboratories are much messier, more
chaotic, and way more interesting. Scientific fields are characterized by core
problems, methods, and techniques. These rich theoretical thickets deter-
mine the range of explananda and the acceptable explanantia.
To complicate things even further, scientific fields neither provide re-
search questions from scratch, nor do they stand alone. What look like new
domains actually import assumptions, results, methodologies, and experi-
mental techniques from predecessors and neighboring disciplines. Here is
a simple albeit paradigmatic illustration. The concept of gene did not origi-
nate in molecular biology. As we saw in section 3.3, the existence of “factors,”
their relevance to heredity, and even their relation to chromosomes were
postulated by classical geneticists who, in turn, inherited them from Mendel.
And their roots arguably dig even deeper. All of this goes to show that, to
spell out the relevant context of a scientific hypothesis, such as the biochem-
ical explanation of meiosis, it is not even sufficient to provide an exhaustive
characterization of cytology. Many concepts, assumptions, methodological
simply that explanatory relevance depends on the contrast class. This is independent of the specifics
of Garfinkel’s example.
116 Black Boxes
standards, problems, and presuppositions have been inherited from previous

paradigms or were transferred “horizontally” from neighboring fields.5
In sum, contextualizing the lighting of matches and the borrowing of
cash is difficult enough. Where do we begin capturing the contrast class
of questions that presuppose a host of intra-field and inter-field scientific
hypotheses, that is, inquiries that span within and across fields? Impossible
as it may seem, students of science learn this quickly and relatively pain-
lessly. How? The key is that, indispensable as they are, scientific questions
are never introduced directly. The framing of an inquiry is always mediated
by the specification of a model. Models are presented and discussed in detail
in section 5.4. In the meantime, we can preliminarily characterize them as
abstract, idealized systems that represent parts of the world on the basis of
similarity. The claim that scientific explananda are contextualized by models
is unlikely to raise many eyebrows. Still, this apparently unremarkable obser-
vation conceals dangerous pitfalls.
Intuitively, the framing of an explanandum must occur prior to the con-
struction of a model. First, we decide what we are trying to explain. Next, we
can construct a model to capture it and explain it. However, if the sharpening
of the explanandum presupposes a model, we run into a classic “chicken and
egg” problem. Posing the right explanandum in the appropriate theoretical
context requires a model, and the model presupposes a phenomenon to be
modeled. How does any of this get us off the ground?
An obvious solution is to posit that both endeavors must be addressed and
developed simultaneously. What looks like two independent processes—
framing and modeling—are really parts of a single endeavor. Framing an in-
quiry is part and parcel of the construction of a model. There is something
true to this: models are goal-directed, they are conceived with an explanatory
target in mind. At the same time, as just presented, this solution is a non-
starter, for the simple reason that models are costly. Questions—scientific
and ordinary ones alike—have many possible readings. Each interpretation,
on the view under present scrutiny, is represented by a different model, and
pragmatic considerations dictate which one is most appropriate in the pre-
sent context. This might work with relatively simple and intuitive scenarios,
like the lighting of matches or Clyde lending cash to Alex. Yet, it would be
preposterously inefficient to ask, say, a team of biologists to provide an array
of models to frame various interpretations of the structure of a gene so that
5 For a classic discussion of these “interfield theories,” see Darden and Maull (1977).
the most appropriate one may be selected. Genetic models do not come easy!
A viable solution must recognize the need for a cheaper, tentative scaffolding
that provides an initial sketchy characterization of what we are trying to ex-
plain, informative enough to get the inquiry going, without making it overly
expensive. How is this done without overburdening researchers with endless
requests for alternative models?
This problem has a simple way out. Recall from Chapter 4 the concept of
frame: a placeholder that stands in for a range of behaviors in need of expla-
nation. Frames are the key to solving our conundrum. To specify an expla-
nandum, one need not spell out an entire model. It is sufficient to provide
the relevant “frame,” that is, a shorthand description of the class of behaviors
that the model will then set out to explain. This preliminary scaffolding,
this coarse-grained characterization of the explanandum, is what kickstarts
an exploration, suggesting how to move the inquiry further. Are these
placeholders permanent, or are they progressively eliminated as science
advances? This will be addressed in Chapter 10. For the time being, let me
illustrate the main idea with an intuitive example. More realistic applications
to actual scientific research are postponed until Chapter 6.
Why did Clyde lend Alex $300? Dretske’s question is subject to interpre-
tation. How does one determine whether the object of explanation is why
Clyde lent Alex $300, why Clyde lent Alex $300, or why Clyde lent Alex $300?
Garfinkel proposes an effective strategy to clarify the explanandum: specify
a contrast class. Do we want to know why Clyde lent Alex $300 as opposed
to $100 or $500, why Clyde lent Alex $300 as opposed to gifting the money,
or why Clyde as opposed to Scrooge lent Alex $300? These coarse-grained
characterizations are frames: placeholders that stand in for patterns of beha-
vior, thereby sharpening our explananda.
Three brief comments: First, providing the appropriate frame, that is,
specifying the contrast class, in and of itself, does not provide an answer to
the question at hand. All it does is pinpoint which question we are trying to
answer, which events are to be explained. Second, identifying the relevant
contrast class clearly falls short of providing a complete model. A full-fledged
model of why Clyde lent Alex $300 as opposed to $100 or $500 will specify
why Alex needs the money, the relation between Alex and Clyde, how Clyde
feels about Alex’s financial situation, and much else. Yet, framing the ques-
tion by specifying a contrast class is the first, preliminary step. It indicates the
object of explanation, indirectly clarifying how the explanandum ought to be
addressed and many other implicit features of the model. Third, the relation
118 Black Boxes
between placeholders and behaviors is best understood as one of representa-

tion. Patterns of behavior are concrete events in the world. In asserting that
frames stand in for a range of behaviors, I am claiming that the placeholder
provides a coarse-grained description—linguistic, pictorial, or otherwise—
that characterizes what all these concrete events have in common, how they
can be subsumed under a single causal explanation. In this representational
sense, frames are placeholders which subsume actual or potential patterns of
behavior in need of explanation.
With all of this in mind, we are finally in a position to present, more pre-
cisely, the first stage of our strategy. The construction of a black box begins
by packaging and contextualizing the object of the explanation. Making the
target fully explicit requires the development of a model, which is no trivial
endeavor. Fortunately, in order to get an inquiry going, we do not need an
exhaustive specification of the complete model. It is sufficient to provide a
frame, that is, a preliminary coarse-grained description or general explora-
tory proto-model that stands in for the range of behavior in need of explana-
tion. These placeholders accomplish what would otherwise appear to be an
impossibly arduous task. They provide a preliminary scaffolding that allows
us to present the explanandum quickly and efficiently.
§5.3. Step Two: The Difference-Making Stage
The first step toward the construction of a black box involves clarifying the
structure of the explanandum. As we have seen in section 5.2, this is a com-
plex endeavor, digging deep into the pragmatics of languages and theories.
Fortunately, this aim can be achieved, in a fairly painless fashion, by pro-
viding a frame, a placeholder that stands in for the patterns to be covered.
Once the inquiry is in focus, the next task is spelling out the explanation
itself.
The topic of explanation has spawned a hefty literature that cannot be
adequately reviewed here, let alone critically discussed. Given our present
concerns, I shall focus on a prominent form of scientific explanation: causal
explanation. How does one provide a causal explanation? The idea, simply
put, is to explain an event by identifying the set of causes that produce it.6
6 Here is David Lewis’s (1986, p. 217) influential formulation: “to explain an event is to provide
some information about its causal history.” As the quote suggests, Lewis’s contention goes further than
mine. He maintains that all explanation is causal. I wish to remain agnostic on this controversial
This requires determining, first, which entities, structures, and processes

make a difference to the target and, second, providing an apportionment of
etiological responsibility. This is the essence of the next constitutive phase of
the black-boxing strategy: the difference-making stage.
Let us begin with the nature of causal explanation. Recognizing that one
way to explain an event is to provide information about its causal history is
only the first step in a long hike. Any happening, even the most mundane, has
a rich and dense history. This raises a question that Strevens (2008) calls the
“problem of relevance.” How do we determine which among myriad causes
of an event should be included in its explanation?
One strategy to tackle the problem of relevance is to flatly deny that there
is any principled distinction to be drawn between explanatory causes and
non-explanatory ones. From this perspective—which Strevens calls the
“minimalist account of causal explanation,” or “minimalism,” for short—
there simply is no selection principle. Every causal influence is explanatorily
relevant to the event brought about, no matter how distant in space or time.
A little reflection will convince most readers that minimalism, thus
crudely stated, will not do. Every event is affected by a gargantuan net of
causal influences literally tracing its roots to the dawn of time and extending
to the corners of the universe. Is it not preposterous to allege that, say, the
gravitational mass of Pluto and the Big Bang are part of the explanation of the
event of me sipping coffee while typing these words?
Sophisticated forms of minimalism acknowledge this objection and at-
tempt a response. Causal influence and, accordingly, explanatory relevance
come in degrees. The current wintery weather and the aroma of freshly
brewed coffee diffusing from the kitchen play the decisive role in prompting
me to opt for this beverage. Pragmatic convenience dictates the omission of
other causal factors, which are only marginally relevant, from abbreviated
explanations. Nevertheless, the minimalist story goes, as the explanations get
increasingly detailed and comprehensive, additional factors may be progres-
sively included, encompassing Pluto and the Big Bang as limiting cases.
Prominent philosophers, such as Railton (1981) and Salmon (1984), have
offered accounts of causal explanation along the lines of this sophisticated
minimalism. Yet, Strevens maintains—correctly, in my opinion—that min-
imalism is too minimal. Specifically, it fails to deliver the intuitively correct
tenet. I assume the platitude that causal explanation is one form of explanation, without presup-
posing that all explanation is causal.
120 Black Boxes
results whenever we have multiple potential causes of an event, but only one
of them should be identified as the explanation of the event itself.7 Any viable
analysis of causal explanation must provide a principled distinction between
explanatorily relevant factors and irrelevant ones.
How does one supplement minimalism with an appropriate selection
principle? Strevens distinguishes two families of strategies. One-factor ap-
proaches develop a more nuanced concept of cause. From this standpoint,
minimalism is correct that all causes explain. Still, not all influences are gen-
uine causes. The cold weather and the pungent aroma are causes of my sip-
ping coffee. Pluto and the Big Bang are not. A different path is followed by
two-factor approaches, which agree with minimalism that all influences are
causal, but stress that not all causes are explanatory. Pluto and the Big Bang
might well be causes of me sipping coffee, or so the suggestion runs. Yet, an
independent selection rule, a principle of causal relevance, will indicate that
these causes play no role in the explanation in question.
Which theory of causal explanation best serves our purposes? Several
options are available in the philosophical literature. The following is a
non- comprehensive list. Regularity theories follow Hume in analyzing
causal relations in terms of uniformities in nature. Statistical theories treat
difference-making causes as factors that alter the probability of the expla-
nandum. Counterfactual theories and manipulability theories view causes
as conditions sine qua non for their effects: had the cause not occurred, the
effect would not have occurred either. Finally, process theories and mecha-
nistic theories view causes as physical connections, understood in terms of
their capacity to transmit marks, exchange conserved quantities, or produce
particular effects. Each strategy can be adapted to explanation in either the
one-factor or two-factor guise. On the former reading, the theory specifies
what counts as a cause of the explanandum event. On the latter interpreta-
tion, it suggests which causes are relevant to the explanation.
Strevens finds none of these accounts of relevance fully satisfactory. His
own theory of explanation is founded on a different criterion for assessing
the significance of causal factors. This is the “kairetic” recipe which, simply
put, goes like this. Begin by taking a complete causal model of the production
7 Strevens illustrates this with the story of Rasputin. Rasputin’s assassins first attempted to poison
him and failed. Next, they shot him and failed again. Finally, they successfully drowned him in a
frozen river. Strevens considers several minimalist responses but argues that they all ultimately
miss the mark. I concur and refer to him for further discussion. For structurally analogous scientific
examples, see Nathan (2014).
of the explanandum, that is, the enormous description of all the causal
influences leading up to the events to be explained. Next, make the descrip-
tion as abstract as possible without violating the following two conditions.
First, it is important not to undermine the explanans’ entailment of the ex-
planandum, that is, one should not make the depiction so general that the
explanandum-event no longer follows from the specification of causes and
initial conditions. Second, the model must remain a causal model so that,
for example, one cannot abstract away by replacing everything with a prop-
osition to the effect that the explanandum occurred. If this process is applied
correctly, what remains in the model after this gradual process of abstraction
is completed are all and only the factors that actually made a difference to the
occurrence of the event that we are trying to explain.
A simple example should help drive the point home. Consider a simple
causal explanation: a piece of butter melts because it is heated. Note that a lot
of properties will influence how the butter melts: its shape, weight, outside
temperature, humidity, and many other factors. Yet, none of these features
makes a difference to the explanandum, that the butter, in fact, melts.
Therefore, they can be removed from the description without invalidating
its status as a causal explanation. In contrast, the claim that the frying pan is
heated above the melting point of butter cannot be abstracted away because it
does make a difference to the occurrence of the effect under scrutiny.
So, which criterion or criteria of causal relevance should we employ to
provide and assess causal explanations? Shall we opt for a regularity or sta-
tistical approach? A counterfactual or manipulability account? A process-
mechanistic analysis? While each theory has its virtues and limitations, all of
them will get the job done, one way or another. Fortunately, we are not forced
to choose. I will not commit to any specific theory of causation or causal
explanation. For the sake of illustration, I often borrow Strevens’s kairetic
account, that—setting aside the issue of whether and how it is possible, in
practice, to perform the required operations of abstraction—provides a clear,
simple, and compelling perspective. Nonetheless, I stress that nothing hinges
on this particular choice. My analysis of black boxes can be paraphrased,
mutatis mutandis, in terms of any theory of causal explanation, as long
as it provides a recipe for identifying factors that make a difference to the
outcome.
Next, what kind of entity produces the explanandum event? Depending
on the specifics of the system under investigation, it may be a set of objects,
events, processes, actions, or something else altogether. To avoid the tedious
122 Black Boxes
task of distinguishing and categorizing all these different phenomena, allow

me to introduce a blanket term intended to cover all types and varieties of
difference-makers. This is the concept of causal mechanism, or mechanism,
for short. Mechanisms will be discussed in detail in Chapter 7. For now, note
that they encompass a heterogeneous class of entities, which may occur at
various scales, relative to the choice of explanandum.8
How does all of this relate to the construction of black boxes? Envision
the situation from the perspective of Strevens’s kairetic strategy.
A difference-maker, on his view, is the result of a process of abstraction,
which removes details from the description of the causal mechanisms that
produce the explanandum in question. After all is said and done, what re-
mains is a functional input-output description of the factors that play an
ineliminable, crucial role in the occurrence of the effect. Had these factors
been absent, the framed event would not have occurred. Thus described,
difference-makers are placeholders. Specifically, they correspond to the
type of placeholders introduced in Chapter 4 as—wait for it—“difference-
makers”! They are abstract depictions that represent, in a coarse-grained
fashion, the detailed mechanisms producing a specific behavior or pattern
thereof.
It might be useful to further discuss the nature of these difference-makers.
First and foremost, the amount of detail in these descriptions may vary. In
some cases, the coarse-grained representation may specify a lot of detail
concerning the structure of the underlying mechanism. Still, as long as it
contains some abstraction, it constitutes a placeholder. At the other extreme,
mechanisms can be referred to even when next to nothing is known about
them. As Strevens (2008, §5.3) notes, even a virtually unidentified mech-
anism may be picked out as “the mechanism responsible for the behavior
of a specific component of the system.” This omission of detail—which
Strevens aptly calls “black-boxing”—may occur for two different reasons
and depending on the specific explanatory aims at play. Let us address them
in turn.
8 In some cases, causes and effects occur roughly at the same level, as when we say that the match
lit because it was struck. But the causal mechanisms producing an event may also occur at finer or
coarser levels. Consider accounts of the lighting of the match which appeal to chemical reactions trig-
gered when phosphorus on the head reacts with potassium chlorate mix on the side of the matchbox.
A precise characterization of levels of explanation is no trivial endeavor (Craver 2007). I appeal to an
intuitive characterization of levels, on the assumption that causes must be “commensurate” to their
effects.
On the one hand, the nature of the mechanism could be known. Nevertheless,
one may choose to leave out details for the sake of pragmatic convenience.
Suppose that you see fragments of ceramic scattered across my office floor. You
ask what happened and I reply that I dropped my mug. I just provided you with
a rudimental causal explanation. Pointing to the pieces of ceramic provides the
frame. The dropping of the mug is the difference-maker. Could I say more? Of
course. I could present both explanans and explanandum in greater detail by
clarifying that it was my favorite mug, or that I dropped it accidentally on my
way to class. With the help of some elementary physics, I could also tell you
more about why and how the mug broke. Still, most interlocutors will be sat-
isfied with my preliminary explanation. Additional information about the tra-
jectory and the force of the mug hitting the ground, or my exact intentions, can
be omitted for the sake of convenience, without affecting the adequacy of this
coarse account.
Note that this is precisely the situation in Sober’s Drosophila scenario. We
can describe the relation between type-a and type-b flies in terms of fitness,
although we could also provide a deeper explanation of the evolution of the
population that dispenses with the concept of fitness altogether.
On the other hand, there are cases where the precise nature of the mech-
anism is unknown. It must therefore be left out as a matter of necessity.
Consider sport-related discussions, common in bars across the globe. Your
friend Jim is offering an unsolicited reconstruction of the last Superbowl,
passionately explaining how the outcome was determined by coaching strat-
egies. The winning coach’s wisdom, the panegyric goes, was the difference-
maker of the game. Jim can provide some evidence. He can pinpoint a few
successful plays and offer a broad-brushed sketch of the strategy itself. Yet,
the precise mechanisms that govern the amazingly complex relation between
tactics and game results cannot be laid out precisely, by Jim or by anyone
else. Here, the details are omitted not by choice, but by necessity. Does this
mean that Jim’s explanation is wanting? Not really. Assuming that Jim is, in
fact, correct, about the difference-makers of the game, the explanandum has
been causally explained. Sure, there are myriad other factors that have not
been covered. Precisely which aspects of the coaching strategy are respon-
sible for the outcome? How did the coach’s decisions affect the game? Could
the overall strategy be improved? These, however, are altogether different
explananda. These questions are framed very differently, and thus require an
altogether different causal explanation.
124 Black Boxes
Scientific analogs of this latter situation can be found in the work of Darwin
and Mendel, who were in no position to accurately describe the mechanisms
underlying their explanations. More contemporary examples involve neuro-
degenerative diseases, such as Huntington or Alzheimer’s, and certain types of
cancer, whose molecular basis is only partially understood.
Time to take stock. All explanations need to be contextualized. Providing the
appropriate setup is no easy task, especially in the case of scientific explanations
that presuppose a host of inter-field and intra-field relations. Pragmatic con-
venience dictates that we employ a shortcut, developing some preliminary
scaffolding that specifies the object of explanation in a fast and frugal fashion,
while omitting lots of detail. This can be typically achieved by constructing a
frame, a placeholder standing in for a range of behaviors in need of explanation.
This is the first step of the black-boxing strategy, the “framing stage.” The next
step involves spelling out the causal explanation, a specification of the variables
that make a difference to the target explanandum, thus contextualized. For this
reason, I refer to this second phase as the “difference-making stage.” The un-
derlying causal mechanisms that produce the behavior in question are typically
quite complex. The specifics of how mechanisms are implemented are often
irrelevant. They can be omitted, because of either ignorance or convenience.
The causal explanation provides abstract depictions of these mechanisms. I
refer to such descriptions as difference-makers: placeholders standing in for the
mechanisms that produce the behavior in question, as noted in Chapter 4.
We are now in a position to fully appreciate the importance of the framing
stage in the causal-explanatory process. Consider the lighting of the match.
Which factors should we enlist among the difference-makers, relative to this
choice of explanandum? I will follow the kairetic recipe. Alternatively, feel free
to pick your difference-making theory of choice. Begin with a complete spec-
ification of all the causal influences leading to the event in question. This web
of influence will be enormous, including all sorts of minor disturbances. Next,
make the description as abstract as possible, removing all the factors that do not
affect the occurrence of the explanandum. Counterfactual reasoning dictates
that, say, the gravitational pull of Jupiter can be omitted because, had Jupiter not
been there, the match would still light. Similarly, we can take out the color of the
matchbox, and many other irrelevant features. In contrast, the striking cannot
be removed because, intuitively, had the match not been struck, it would not
have lit. This shows that the striking is, indeed, a cause of the lighting.
Wait. Sure, had the match not been struck, it would not have lit. But had no
oxygen been present in the atmosphere, the match would not have lit either.
And had the match been faulty, or made of plastic, it would also not have
lit. This is Goodman’s problem of counterfactuals.9 Why do we treat striking
as the cause of lighting, but not oxygen or the absence of manufacturing
defects? The answer, once again, lies in the framing process, which will influ-
ence the selection of causes, the structure of the model, feeding back on the
nature of the explanandum.
My central contention—the relativization of all causal statements to a
frame of reference—is hardly novel. In his classic study, The Cement of the
Universe, J. L. Mackie (1974, pp. 34–35) notes that causal statements “are
commonly made in some context against a background which includes the
assumption of some causal field.”10 Strictly speaking, then, what is caused
is not an event simpliciter, but an event relative to a causal field. This has
implications for our understanding of causal explanation. In assessing what
caused event E relative to field F, some conditions that play a necessary role
in the production of E can be dismissed as having no genuine causal role.
These are the conditions that are part of F. But, if we evaluated the same state-
ment relative to a different field, call it “G,” then parts of F may now be said
to cause E and, vice versa, some of the causes of E relative to F might become
part of G, fading into the background.
Analogous considerations have been revamped by Strevens (2008).
Echoing Mackie, Strevens treats the “given that” clauses which constitute the
framework as part of a fixed portion of the explanation against which spe-
cific difference-makers are evaluated. More precisely, a state of affairs s is part
of the framework F of explanandum E if all causal explanations of E must
9 In Fact, Fiction, and Forecast, Goodman introduces the “problem of counterfactuals” as the
task of defining “the circumstances under which a given counterfactual holds, while the opposing
conditional with the contradictory consequent fails to hold” (1955, p. 4). Simple as this may ap-
pear, Goodman is quick to point out two formidable difficulties. A first issue, the “problem of law,”
challenges us to describe the nature of the systematic connection between the antecedent and the
consequent of a subjunctive, given that such connection will typically not be a matter of logic but a
natural, physical, or causal law. This endeavor is rendered thorny by the threat of circularity. Laws are
counterfactual-supporting; accidents are not. Appealing to counterfactuals to identify laws and, si-
multaneously, employing laws to analyze counterfactuals generates a vicious circle or infinite regress.
Goodman’s second difficulty, more directly pertinent to our concerns, is known as the “problem
of relevant conditions,” or, alternatively, the “problem of co-tenability.” The connection between
antecedent and consequent in a counterfactual presupposes the occurrence of stable background
conditions, often left implicit. “Had the match been struck, it would have lit” only holds provided that
the match is well made and dry, oxygen is present, wind is absent, etc. Specifying which features must
be taken in conjunction with the antecedent to infer the consequent is a long-standing philosophical
problem. The connection between the context-dependency of explanation and subjunctives is noted
by van Fraassen (1980).
10 Mackie attributes the introduction of causal fields to John Anderson’s “The Problem of
Causation” (1938) and draws a connection to Russell’s (1913) “causal environment.”
126 Black Boxes
contain s, because s makes a difference to the occurrence of E. Just as Mackie

disqualifies any part of the causal field as a genuine cause, Strevens claims
that s is not a difference-maker for E, as evidenced by the observation that s
need not be mentioned explicitly in the causal explanation of E. In essence, in
the context of a causal explanation of E, the status of s as “difference-maker”
is suspended. It becomes a background condition.11
Technical as this may sound, the bottom line is a simple observation. All
explanations, causal or otherwise, must be understood relative to a frame-
work. And, although this is not invariably stated explicitly, the choice of the
framework is always relative to the choice of explanandum. When I claim
that the lighting of the match is caused by striking, I can dismiss the causal
relevance of oxygen and other background conditions by including them
in the framework. At the same time, when I reframe the question by asking
why the match lit on a distant planet, the presence of oxygen becomes the
difference-making factor and the striking event can be relegated to the back-
ground. This is why the framing stage is so important and why it is crucial to
keep it distinct from the difference-making stage, where the causal explana-
tion itself is spelled out.
With all of this in mind, we are ready to move on to the final step of black-
boxing, the representation stage, where models re-enter the picture.
§5.4. Step Three: The Representation Stage
Are we done? Have we concluded our characterization of the black-boxing

strategy? Not quite. After completing the first two steps, what we have is a
causal explanation relativized to a frame. To obtain a black box, our framed
causal explanation needs to be represented. We need to determine which
features of our causal explanation should be explicitly described, and which
can be omitted. This is done by embedding the contextualized causal story
within a model. What is a model? How does it capture explanations? Section
11 Two points are worth stressing. First, Strevens is more explicit than Mackie that frameworks
can be introduced in discourse in a number of ways. It may be presupposed, implied, stated overtly,
etc. Second, for Strevens, the introduction of a framework is optional. Many explanatory claims, he
argues, are not relative to any framework and, therefore, they specify absolute as opposed to relative
relations of difference-making. I respectfully disagree. I maintain that all explanations, no matter
how simple or apparently unambiguous, presuppose a host of background assumptions. At the same
time, my argument does not require this stronger claim. The main point, for present purposes, is the
weaker observation that all scientific explanations presuppose a framework.
5.2 provided a general characterization of models as abstract and idealized

systems representing features of the world. We now need to put some flesh on
these bare bones and spell out a more developed account. This will provide
the key to the final phase of black-boxing: the representation stage.
A good place to begin our discussion of models and representation is the
influential work of Ron Giere. Giere’s starting point is the observation that
standard vehicles of scientific knowledge are not accurate descriptions of the
world. What we find in textbooks and research articles across the sciences
are highly abstract and idealized systems with no claim to reality beyond
that conferred to them by the relevant community of researchers. Giere calls
these constructs “theoretical models,” or simply “models,” for short.12
What is the relation between a scientific theory and the system it purports
to represent? Traditionally, philosophers characterize the relation between
theory and reality as a sort of correspondence. On this view, scientific the-
orizing provides “true” descriptions which accurately depict the corre-
sponding portions of the world. Giere is adamant in rejecting this. For him,
there is no direct connection between the world and its representations. The
only relation is an indirect one, mediated by a theoretical model. Thus, what
one finds in textbooks, what is generally called a “scientific theory,” is best
described as a cluster of abstract, idealized models, together with various
hypotheses linking these constructs to real-world systems.13
Empirical hypotheses of the kind routinely found in science posit a rela-
tion of similarity between an abstract model and a physical portion of the
world. Yet, since anything can be deemed similar to anything else, claims
of similarity are vacuous without some more or less explicit specification
of the relevant respects and degrees. To make the world-model connection
more precise, Giere introduces the notion of “theoretical hypotheses,” that
is, statements of the form: “system S is similar to model M in the indicated
12 Giere develops these views in his classic Explaining Science: A Cognitive Approach (1988). A sim-
ilar perspective has been defended by Cartwright (1983), who maintains that the fundamental laws
of modern physics, such as Schrödinger’s equation, are not true. In her celebrated phrase, they “lie.”
Giere agrees with the content of Cartwright’s proposal but opts for a different reformulation. For him,
the general laws of physics cannot tell lies about the world because they are not statements about
the world at all. They are, as Cartwright herself sometimes suggests, parts of the characterization of
theoretical models, which may represent real systems. Here, I focus on Giere’s presentation because
his stance on models is more developed and explicit than Cartwright’s. Be that as it may, the relevant
issue, for our present purposes, is not much whether or not models “lie,” but that they do not provide
accurate representations.
13 Contrary to the positivist conception of theories as interpreted axiomatic systems, discussed
in Chapter 2, for Giere theories are not well-defined entities. No set of necessary and sufficient
conditions determines which models and hypotheses belong to a theory.
128 Black Boxes
respects and degrees.” Unlike models, theoretical hypotheses are true or false,
depending on whether or not the asserted relation actually holds. However,
as Giere (1988, p. 81) is quick to point out, “that theoretical hypotheses can
be true or false turns out to be of little consequence. To claim a hypothesis
is true is to claim no more or less than that an indicated type and degree of
similarity exists between a model and a real system. We can therefore forget
about truth and focus on the details of the similarity. A ‘theory of truth’ is not
a prerequisite for an adequate theory of science.”
In sum, Giere characterizes models as abstract and idealized systems,
vehicles for representing the world based on similarity. Giere was neither
the first nor the last to recognize the importance of models in scientific
practice. His main insight was replacing the old relation between model
and reality in terms of correspondence with a relation of stipulation.
Models are defined—created by their description. Consequently, whether
a model sufficiently resembles the world is not a brute fact, but a collec-
tive decision of the members of the appropriate portion of the scientific
community.
With this in mind, we can now consider, in greater detail, the issue of sci-
entific representation. How do models represent their target systems? And
how do we assess whether or not they succeed in doing so? Answering these
questions will require us to discuss the anatomy of models.
Giere’s account, originally developed with an eye to mathematics and
physics, is difficult to apply to less formal branches of sciences, such as parts
of biology, neuroscience, and psychology. For this reason, I will adopt a
broader view of representation, which generalizes Giere’s proposal while re-
taining its spirit. Weisberg (2013) characterizes models as interpreted struc-
tures. Unpacking this definition entails decomposing models into their two
basic constituents: structures and interpretations.
Weisberg distinguishes three types of models—concrete, mathematical,
and computational—based on the underlying structure. Concrete models
are real physical entities that stand in some representational relation to some
real or imagined system, the model’s “intended target.” Mathematical models
employ formalized structures to represent states and relations between states.
Finally, computational models represent causal properties of their targets by
relating these causes to procedures. Setting these differences to the side, at
the basic ontological level, all models are structures.
The second constitutive feature of models is their interpretation. Weisberg
breaks down the interpretation of a model—what he calls the model’s
“construal”—into four parts: an assignment, the modeler’s intended scope,

and two kinds of fidelity criteria, “dynamical” and “representational.”
The first two components of interpretation—assignment and scope—
determine the relation between the model and real phenomena. As Weisberg
(2013, p. 39) puts it, “Assignments are explicit specifications of how parts of
real or imagined target systems are to be mapped onto parts of the model.”
This explicit coordination is important for two reasons. First, although parts
of the model may map “naturally” onto specific portions of the world, this is
often not the case. Second, and relatedly, it is not uncommon for assignments
to be left implicit in discussions of models. The intended scope of the model,
in contrast, specifies which aspects of potential target phenomena are
represented by the structure and which are omitted.
A simple example may help clarify things. How do we use a concrete
structure, beads on a string, to represent genes on a chromosome along the
lines of what Ernst Mayr dubbed “beanbag genetics”(Figure 5.1)? The first
step involves interpreting the structure. We have to specify, implicitly or ex-
plicitly, which aspect of the beads correspond to which properties of genes.
Accordingly, an assignment must tell us whether each bead corresponds to
a nucleotide or to a base-pair. It might further use the string to represent
the backbone of phosphate and sugar residues keeping molecules together.
Second, the scope tells us what is represented and what is not. Are the beads
pattern-coded, that is, is there a signature mark for each of the four kinds of
bases? Do the beads represent functional differences between sequences of
nucleotides, such as coding sequences, operons, and promoters? Does the
GENE
DNA
NUCLEOTIDE
Figure 5.1. A simple model of genes: beads on a string.

130 Black Boxes
model represent single strands of DNA or double helices? Are ribosomes in-
cluded? Which properties convey information (number of beads, proximity)
and which should not be taken literally (size, shape, texture)?
The third and fourth components of construal are its fidelity criteria. These
determine how similar the model must be to the world, and in what respects,
for the representation to be considered “adequate.” Weisberg distinguishes
two types of criteria. Dynamical fidelity criteria specify how close the output
of the model, the values of dependent variables in the model and in the world,
must be to the input of real-world phenomena. This is the “error tolerance” of
the system, how far off predictions made on the basis of our beads on a string
are from the behavior of real genes. The second family of fidelity criteria are
representational.14 Typically, these provide standards for evaluating how well
the causal structure of the model maps onto the structure of the world, for
the representation to be adequate. Do our large, colored, spherical beads pro-
vide a good enough approximation of real biochemical molecules? Can this
simple structure capture the complexities of gene replication? What are the
salient analogies and differences between the causal structure of the world
and our representation of it?15
We are now in a position to state the final step of the black-boxing strategy.
The representation stage determines which features of causal explanations
should be portrayed and which can be idealized and abstracted away.
Was this not already done at the difference-making stage? Not quite. The
goal of difference-making is, first, to identify the factors that make a differ-
ence to the occurrence of an event of choice and, second, to distinguish be-
tween genuine causes and conditions that may fade into the background.
Once these causes have been pinpointed, there is still a question of how they
14 As Strevens and DeMartino (personal communication) have independently brought to my at-

tention, Weisberg’s terminology may be slightly misleading, as dynamical fidelity can also be under-
stood as fundamentally “representational.” Since nothing of substance here hinges on the distinction
between different kinds of “fidelity criteria,” I shall stick to the original nomenclature. At the same
time, I should stress that, regardless of how Weisberg’s approach in interpreted, on my view, models
should be understood in terms of similarity with target systems, not correspondence.
15 Given this emphasis on interpretation, can any structure at all be a model? Can anything model
anything else? To a first approximation, Weisberg answers in the positive. Yet, he recognizes that this
might be misleading. While it is, strictly speaking, true that anything can be a model of anything else,
concrete objects with very simple structures and few states have a “low representational capacity,”
that is, they are not able to represent many systems, especially ones of a very different type. Moreover,
when dealing with concrete models, researchers should pay attention to the overall similarity of a
model to the system being studied. It is difficult to represent one system with a structure very much
unlike it. In this respect, while one can represent genes as beads on a string, a more complex double-
helical structure will do a much better job. In conclusion, as Weisberg puts it, “The art of good mod-
eling involves not only choosing an appropriate construal for a model, but also choosing structures
with representational capacities appropriate to the target or the theoretical task at hand” (2013, p. 45).
should be represented. Precisely what goes in the model, and in what ways?
Representation is achieved by constructing a model, an interpreted structure
that represents parts of the world based on relations of similarity.
Following Weisberg, I distinguished three kinds of structures—physical,
mathematical, and computational— and four aspects of interpretation.
Assignment and scope map the structure onto the model. Dynamical and
representational fidelity criteria determine whether the representation
provides an adequate, informative characterization of the system under
scrutiny.
This is all admittedly vague. Invoking the notion of similarity, in and of it-
self, does not resolve the issue of representation, which constitutes one of the
thorniest and most resilient open problems in the philosophy of science.16
Similarly, appealing to models does not, ipso facto, tell us how the expla-
nandum should be represented and explained. Unfortunately, constructing
and assessing models cannot be done from the proverbial armchair. It
requires a painstaking combination of theoretical and empirical work. Still,
the work of Giere and Weisberg provides an effective depiction of the repre-
sentation stage. Metaphorically speaking, it involves constructing the frame-
work that offers the biggest “bang for the buck” relative to the explanatory
purposes at hand.17
Before moving on, a brief comment concerning the role of models in sci-
entific inquiry. Models have been the object of much discussion in the phi-
losophy of science.18 Yet, traditionally, scholars have focused on a particular
aspect of models, namely, their role as explanantia and explananda.19 While
models are a worthwhile object of investigation, and their contribution to the
explanation of a phenomenon or range of phenomena is certainly significant,
this is not the entire story. As we have seen throughout the chapter, models
play a variety of other roles. For one, models frame the explanandum, making
it more precise and perspicuous. Second, models provide the “toolbox” of
explanantia. They specify the array of all the entities, laws, mechanisms, and
16 For a comprehensive book-length discussion, see van Fraassen (2008).

17 To the best of my knowledge, this colorful and effective metaphor was first used by President
Eisenhower’s Secretary of Defense, Charles E. Wilson, in 1954, to describe the “New Look” policy
of depending on nuclear weapons, as opposed to a regular army, to keep the Soviet Union in check.
More recently, it has been developed, in the context of scientific explanation, by Franklin-Hall (forth-
coming), from which I borrow it.
18 Suppes (1960) and van Fraassen (1980) are pioneers. More recent developments can be found in
Cartwright (1983); Morgan and Morrison (1999); and Wimsatt (2007).
19 This trend is slowly changing due, in no small part, to the pioneering work of Morgan and
Morrison (1999) which emphasized the multifaceted role of models in science.
132 Black Boxes
processes that can be employed in the construction of the explanans. Third,

and finally, models provide the appropriate methodological standards. They
tell us how the explanation at hand is supposed to operate and how to as-
sess it. In essence, models are what allows researchers to address the chal-
lenge first posed by Goodman and subsequently developed by Mackie,
Dretske, van Fraassen, Garfinkel, Strevens, and many others. Models effec-
tively specify the background assumptions implicitly presupposed by an ex-
planatory question and contextualize the inquiry. This framework does not
precede the explanation. It is not implied by a theory. The framework is a
constitutive part of the theory itself.
The radical philosophical implications of these commonplace observations
will be explored, more systematically, in Chapter 7. For the time being, with
all three steps in place, we have the conceptual resources to spell out the ge-
neral definition of black boxes we have been seeking.
§5.5. What Is a Black Box? A Three-Part Recipe
This chapter has dissected the practice of black-boxing into three phases.
The first step, the framing stage, involves sharpening the object of expla-
nation. Specifying the explanandum in detail presupposes a full-fledged
model, which is not easy to obtain, especially in actual scientific practice.
Fortunately, to get the inquiry going, it is sufficient to employ a frame, that
is, a coarse-grained placeholder that stands in for patterns of behaviors in
need of explanation which, in principle, could be described at a finer scale.
The second step, the difference-making stage, provides a causal explanation
of the target by specifying which features of the explanans make a difference
to the occurrence of the explanandum.20 While there are various strategies
for doing so, I borrowed Strevens’s effective kairetic approach. Many of these
difference-makers may be left unpacked. In some cases, the micro-structure
is omitted because of mere convenience, to draw the boundaries of a field, or
to insulate a concept from empirical refutation. Other times, the decision is
dictated by ignorance, as these details are actually unknown. The third and
20 As discussed at length, I do not have any novel account of causal explanation to offer and I wish
to not take a stance on the question of whether all explanation is truly “causal.” Still, it is worth
stressing that the notion of difference-making adopted here is weak: it does not presuppose any phys-
ical interaction. Hence, my notion of black box can be applied to mathematical, statistical, and other
forms of abstract explanation.
final step, the representation stage, determines how these difference-makers

should be portrayed, that is, how much abstraction and idealization will
produce the optimal explanatory “bang for the buck,” given how the expla-
nandum was framed. This is done by constructing a model, an interpreted
structure that represents bits and pieces of the world based on a relation of
similarity.
Our analysis began with a series of questions. What is a black box? How
does it work? How is a black box constructed? What kind of entity is able
to accomplish such a diverse set of tasks? We now finally have a skeleton of
an answer. The three stages discussed in this chapter lead us to the general
definition we were seeking. A black box is a placeholder—frame or difference-
maker—in a causal explanation represented in a model.
A simple illustration is Putnam’s geometrical explanation of the square-
peg-round- hole scenario. The first step involves specifying the expla-
nandum. The simple question “why does the square peg not go through the
round hole?” introduces a frame. Contextualizing the object of the expla-
nation makes the physical properties determining the rigidity of the board
irrelevant. They can be assumed as part of the framework and, therefore,
readily dismissed. The difference-makers are the geometrical properties of
the system, mentioned in Putnam’s higher-level explanation, which gener-
alize to any rigid system satisfying the initial conditions. Still, reframe the
explanandum—for instance, ask what makes the board and peg rigid—and
the difference-makers will have to include some of these details.
Three comments. First, is this really it? Could we not have given this def-
inition at the end of our historical excursus, sparing us two long chapters of
dense philosophical analysis? We surely could have. But now we are in a posi-
tion to fully appreciate the facets of the definition in question. At the most ge-
neral level, a black box is a placeholder, a coarse-grained representation that
stands in for entities whose nature and structure are either unknown or set to
the side. As we saw in Chapter 4, there are two kinds of placeholders: frames
and difference-makers. Correspondingly, there are two types of black boxes,
depending on whether the representation in question spells out an explanans
or an explanandum, that is, whether the description stands in for patterns
of behaviors or for mechanisms. The causal explanation should be under-
stood as a specification of factors that make a difference to the occurrence
of the explanandum. The selection of these difference-makers is determined
by the contextualization of the target-event itself, which is initially sig-
naled by a frame and ultimately presupposes a scientific model, an abstract
134 Black Boxes
and idealized interpreted structure. The model also determines how these
difference-makers should be represented—which aspects of the underlying
mechanisms must be explicitly depicted, and which ones can fade into the
background. When detail is omitted, this is done on the assumption that it
does not affect the autonomy of higher-level explanations. No wonder it took
us so long to cover all this ground. Phew!
Second, my characterization places both frames and difference-makers
on their own, independent, and mutually exclusive spectrum. These
placeholders range from general, coarse-grained depictions—”the mech-
anism responsible for genetic inheritance” or “the range of behaviors
typically described as ‘falling in love’ ”—to detailed descriptions of the un-
derlying entities and activities. Hence, my general definition of a “black box”
encompasses also less-opaque constructs, sometimes dubbed “gray boxes”
or “semitransparent boxes,” which shall be discussed, at greater length, in
Chapter 7. What all these placeholders have in common, from my perspec-
tive, is that they result from a process of abstraction and idealization.
Third, as anticipated in section 5.1, these stages are necessary and jointly
sufficient for the construction of a black box. Not all valuable scientific work
requires all three phases. But all black boxes do.21 Yet, these steps should
not be viewed as chronologically ordered. Sure, there is a natural progres-
sion which begins by framing the explanandum, then provides the explana-
tion, which is finally represented in a model. Elegant as it is, this conception
of “well-ordered science” is just a regulative ideal. Actual practice is messy.
Research teams often begin by hypothesizing a causal explanation, which is
then found wanting, leading to the revision of the explanandum. Or a re-
vised representation of difference-makers might lead to a revisitation of
the explanandum, which, in turn, changes the nature of the causal explana-
tion. In short, the implementation of these steps may involve permutations,
repetitions, and feedback loops. All of this is captured by Firestein’s (2012,
p. 19) gloss: “ ‘Let’s get the data and then we can figure out the hypothesis,’
I have said to many a student worrying too much about how to plan an
21 Some readers might accept that all three steps are required to construct a black box, understood
as a difference-maker. But why does the construction of a black box qua frame require difference-
making and representation? Is the framing process not sufficient? My reason for answering in the
negative is simple. While a frame, per se, is a preliminary exploration, a black box is always relativized
to an entire model. And this requires the processes of difference-making and representation, in addi-
tion to framing.
experiment”—incidentally, an example of data mining which, like black-

boxing, is praised by some scholars while frowned upon by others.22 Still, all
three phases are required to characterize how black boxes are constructed,
used, and assessed in science. Speaking of which. . . .
We will get back to these questions and related issues of explanatory re-
duction in Chapter 10, where we shall also address the question of whether
scientific explanation will invariably incorporate some black boxes, or if all
these constructs can be progressively opened and eliminated, at least in prin-
ciple. Meanwhile, so much for square pegs, matches, cash lenders, bank rob-
bers, and other toy examples. Does our general definition capture the use
of black boxes in real historical cases? What implications does this have for
philosophy at large? It is time to put the recipe to work. Chapter 6 applies our
three-step recipe to Darwin, Mendel, Skinner, and Friedman. Next, we shall
offer some philosophical payoff of the strategy at hand.
22 Broadly speaking, data mining can be characterized as the practice of examining large databases
in order to generate new information. Data mining is controversial because it tends to seek out large-
scale correlations and set aside any causal-mechanistic underpinning.
6
History of Science, Black-Boxing Style
The history of science, after all, does not just consist of facts and
conclusions drawn from facts. It also contains ideas, interpretations of
facts, problems created by conflicting interpretations, mistakes and so
on. [ . . . ] This being the case, the history of science will be as complex,
chaotic, full of mistakes, and entertaining as the ideas it contains and
these ideas in turn will be as complex, chaotic, full of mistakes, and en-
tertaining as the minds of those who invented them.
—Paul Feyerabend, Against Method, p. 3
§6.1. Introduction
The previous chapter provided an analysis of the black-boxing strategy by

breaking it down to three constitutive steps. First, the framing stage sharpens
the explanandum by contextualizing the object of explanation. This gets
the inquiry going by constructing a frame, a placeholder that stands in for
patterns of behavior to be covered. The second step, the difference-making
stage, provides a causal explanation by identifying the factors that make a
difference to the occurrence of the explanandum: its difference-makers.
The nature and structure of the mechanisms producing the phenomenon
under scrutiny are often quite complex. But not all details are required or
even useful for the explanation which, when appropriately framed, may well
be perfectly autonomous. Which aspects must be explicitly represented?
Are there features that may be idealized or abstracted away? What are the
standards for the explanation? Which assumptions can be presupposed, and
which need to be introduced and defended explicitly? These questions are
addressed in the third, and final, representation stage, which aims at pro-
ducing a model of the explanandum. The significance of this step is often
neglected in discussions of causation and causal explanation. It is common
to talk as if causal attributions are absolute or context-independent. This is
not the case. As our discussion made clear, explanations—and, especially,
DOI: 10.1093/oso/9780190095482.003.0006
History of Science, Black-Boxing Style 137
scientific explanations— are always relativized to a frame of reference.

Consequently, causal claims are mediated by models that determine how the
explanandum should be represented, explained, and ultimately assessed.
These three steps jointly suggest and support a general definition. A black
box is a placeholder in a causal explanation represented in a model.
Chapter 4 distinguished two types of placeholders: frames and difference-
makers. Accordingly, we have two types of black boxes, depending on whether
the placeholder is an explanandum standing in for patterns of behaviors or
an explanans representing the mechanisms producing the behaviors in ques-
tion. The causal explanation spells out the difference-makers, which are then
represented in a model, an interpreted structure that stands in a relation of
similarity with bits and pieces of the real world. The nature and structure of
these black boxes—which may also be “gray” or “semi-transparent”—vary
substantially, depending on the amount of detail included. But they all have
in common the omission of content, idealized or abstracted away for prag-
matic convenience or necessity, on the assumption that this does not affect
the autonomy of the higher-level explanation.
Time to put this strategy to work and address the following questions. Is
our analysis adequate? Is it accurate? Does it capture the essence of black-
boxing? What are its limitations? At the end of the previous chapter, the def-
inition was illustrated by a toy example: Putnam’s square-peg-round-hole
system. Now, let us look at some real scientific applications.
At a general level, the second half of this book focuses on two broad is-
sues. First, does our three-step recipe for constructing black boxes fit in with
the vagaries of the case histories first introduced in Chapter 3? Now, given
that our three-step recipe for black-boxing was motivated and set up by
discussions of evolutionary theory, classical Mendelian genetics, behavior-
istic psychology, and neoclassical economics, it may not come as a shock to
see, in the present chapter, that the analysis in question successfully captures
the same episodes that inspired it. Nevertheless, our discussion of black
boxes does not merely strive to vindicate the initial choice of examples. It
puts them in a different light. Second, what theoretical implications does all
of this have? Presenting and assessing the philosophical payoffs of the black-
boxing strategy will be the aim of the chapters to follow.
Before moving on, I should stress, once again, that my goal is not to pro-
vide a comprehensive narrative of these momentous events in the history of
science. This substantial endeavor lies beyond the scope of this work. More
modestly, my objective is, first, to provide an accurate rational reconstruction
138 Black Boxes
of the role and significance of black boxes across scientific practice and,
second, to discuss its philosophical applications and implications.
§6.2. Darwin’s Black Boxes
Darwin’s theory of evolution by natural selection provides a powerful appa-

ratus for explaining the distribution of organisms and traits across the globe.
On the Origin of Species, can be viewed as one long argument organized into
three main parts. The first four chapters of his masterpiece present the funda-
mental tenets of his evolutionary theory. These are the ingredients skillfully
blended by Darwin into the main course: the principle of natural selection.
Chapters 5–8 provide evidence for natural selection, anticipate objections,
and respond to them. Chapters 9–13 put the account to work, revealing what
can be accomplished by applying these insights to problems in biology. The
final chapter summarizes and wraps up his discussion.
Let us focus on the third section of the Origin, where the breadth and
depth of evolutionary explanations truly come to life. Specifically, consider
Chapters 11 and 12, which address issues of biogeography. There, Darwin
begins by considering three striking facts regarding distribution patterns.
First, neither similarities nor dissimilarities among organisms can be fully
accounted for by environmental features alone. The reason is evident.
Species inhabiting distant lands differ vastly and, yet, continents across the
globe display similar climatic conditions. A second observation concerns
how geographic barriers, as well as other obstacles to free migration, deeply
affect variation among organisms and ecosystems. A clear illustration is the
difference in marine flora and fauna inhabiting eastern and western shores
of Central America, attributable to the narrow albeit (for many species) im-
passable isthmus of Panama. Third, there are unquestionable analogies be-
tween species inhabiting similar environments. Organisms from the same
continent or ocean share features that set them apart from species in other
areas of the planet, as shown by the uniqueness of Australasian marsupials.
These three facts, Darwin remarks, call for an explanation.
Given the remarkable scope of the explanandum—worldwide patterns of
geographical distribution—one could legitimately expect Darwin’s analysis
to be rich in detail and structure. This expectation is misplaced. Readers of
the Origin are presented with a surprisingly minimal account:
We see in these facts some deep organic bond, prevailing throughout space
and time, over the same areas of land and water, and independent of their
physical conditions. The naturalist must feel little curiosity, who is not led
to inquire what this bond is. This bond, on my theory, is simply inheritance,
that cause which alone, as far as we positively know, produces organisms
quite like, or, as we see in the cases of varieties, nearly like each other. The
dissimilarity of the inhabitants of different regions may be attributed to
modification through natural selection, and in a quite subordinate de-
gree to the direct influence of different physical conditions. (1859 [2008],
pp. 257–258)
Thus, in a few short statements, Darwin provides the key to one of the great
mysteries of science: the question of biogeography. How is this possible?
Some readers might feel inclined to downplay the brevity of Darwin’s
analysis by appealing to its obviousness. From a historical perspective, this
would be a mistake. Darwin’s view, deeply rooted in contemporary bio-
logical thinking, might sound like a truism to a modern audience. Yet, one
should not ignore the originality of the proposal and the stark opposition
confronting evolutionary theory since its inception, and which continues
inexorably to this day, at least in some circles. Darwin’s insight challenged
some of the most basic and entrenched, but still widely debated, credence of
its age. These include the belief that organic diversity is the result of divine
creation, that the Earth is much younger than it is, that natural phenomena
must be explained via teleological concepts like design, and an anthropocen-
trism placing humans at the pinnacle of the scala naturae. In addition to his
contribution to discrediting these hallowed ideas, Darwin also bolstered a
number of key concepts, such as the replacement of essentialism with popu-
lation thinking, the principle of natural selection, geographic speciation, and
the understanding of evolution as a process. Some will surely quibble with
Mayr’s somewhat hyperbolic conclusion that “no other philosopher or sci-
entist has had as great an impact on the thinking of modern man as Darwin”
(1988, p. 194). Yet, it is hard to overstate how evolution by natural selection
constitutes a traumatic rupture with the worldview of orthodox Christians,
natural theologians, laypeople, as well as many philosophers and scientists
born and raised in the nineteenth-century intellectual milieu.
These considerations bring us back to our original question. Given this
complex thicket of conceptual innovation and intellectual battles, how is it
140 Black Boxes
possible for Darwin’s evolutionary analysis to be so beautifully simple? How

can such a succinct explanation do so much theoretical heavy lifting?
The answer, in a nutshell, is that Darwin achieves such an elegant explana-
tion through his effective construction of black boxes. Allow me to elaborate
by decomposing Darwin’s endeavor into our three key steps.
To begin, Darwin’s target is not biogeography simpliciter. He is after a
highly hedged and contextualized set of patterns. To wit, myriad distributions
of organisms can be observed on the planet. Not all of them count as an ap-
propriate explanandum for evolution by natural selection. Darwin’s analysis
can account for differences between, say, cave-dwelling animals on different
continents and anatomical similarities between Australasian marsupials. Yet,
it does not apply to distributions that are the effect of luck, such as the out-
come of natural catastrophes. It does not cover the production of traits that
do not differ with respect to fitness, or the origins of life.1 In short, Darwin’s
explananda are far from absolute, self-standing events. They are specific
problems, relativized to contexts and frameworks.
What is Darwin’s explanatory model? Simply put, it encompasses his
“theory” of evolution by natural selection. This corresponds to the apparatus
patiently spelled out in the initial chapters of the Origin, that we succinctly
reviewed in section 3.2 of Chapter 3. The existence of a process of natural
selection can be directly inferred from the struggle for existence described
by Malthus, together with the ubiquitous presence of variation in nature,
some of which affects fitness and can be inherited across generations. Such
textbook reconstructions should not be taken too literally. Breaking the
theory of evolution down to a handful of axiomatic principles fails to do
justice to its groundbreaking contribution. What Darwin describes as “one
long argument” also involves a number of methodological directives, such
as the adequacy conditions that he borrows from Herschel, Lyell, and other
precursors. It includes painstaking observations—some novel, others well
known—like the paucity of the geological record or comparisons between
organisms and environments across the planet. Finally, it encompasses more
or less implicit assumptions concerning what he deemed obvious, prepos-
terous, or worthy of exploration—Grene and Depew (2004) call these “epi-
stemic presuppositions”—together with objections and his responses. This
1 To be clear, I am not arguing that contemporary evolutionary theory has nothing to say about
chance or the origins of life. It does. My point is that Darwin was is no position to answer these
questions. In this sense, the title of his masterpiece is misleading. He offers lots of insights on the evo-
lution of species, but not much about their “origin.”
hodgepodge of assumptions, observations, anecdotes, side remarks, and

hypotheses provides the backdrop against which Darwin’s research must
be understood, processed, and assessed. In a nutshell, they are his model.
Parts of this evolutionary theory are stated explicitly. Still, if Darwin were to
spell out in full the host of intra-field and inter-field connections that lurk
in the background, his inquiry would never take off the ground. Many, in-
deed most, premises are implicit. Thus, when the English naturalist states
that he shall consider “the distribution of organic beings over the globe,” this
coarse-grained description is a frame: a placeholder standing in for specific,
highly hedged distribution patterns in need of explanation. These frames are
what allow his inquiry to be presented in such a simple, succinct, and elegant
fashion.
The framing of the explanandum concludes the first stage of black-
boxing. Next is the difference-making phase, where the causal explanation
is laid out. Which variables make a difference to the production of the
patterns of geographical distribution presented in the Origin? My rational
reconstruction borrows from Strevens’s kairetic recipe.2 As noted in sec-
tion 5.3 of Chapter 5, similar results can be obtained by following various
alternative strategies. Begin with a detailed description of a specific dis-
tribution subsumed under the frame discussed earlier. For instance, con-
sider some differences in the marine flora and fauna along the eastern and
western shorelines of the isthmus of Panama. Next, make the depiction
as abstract as possible without undermining the entailment of the expla-
nandum, that is, ensuring that the distribution in question still follows
from the relevant initial conditions and law-like generalizations. What are
the decisive factors in producing the patterns under scrutiny? The short
answer is descent with modification.
Obvious as all of this may sound to contemporary readers, it was quite
controversial in the mid-nineteenth century. Prominent intellectuals still
believed that the only plausible answers would appeal to independent cre-
ation or intelligent design. Other influential minds considered this ques-
tion impossible to answer—Lyell’s “mystery of mysteries.” Darwin had a
different solution in mind, as we saw in the passage quoted earlier: “The
dissimilarity of the inhabitants of different regions may be attributed to
2 I refer to this as a “rational reconstruction” because, to the best of my knowledge, Darwin him-
self never explicitly identifies these three steps. Still, reformulating the explanation in contemporary
terms allows us to identify and characterize its distinctive aspects.
142 Black Boxes
modification through natural selection, and in a quite subordinate de-

gree to the direct influence of different physical conditions.” Ecce Darwin’s
causal explanation.
The abstract nature of the difference-makers renders the analysis perfectly
general. Hardly specific to the Panama coastlines, it applies, mutatis mutandis,
to organisms and environments across the globe. Why are native inhabitants
of the Galapagos similar to the fauna of South America and different from
species living on islands and mainland elsewhere? Darwin answers in terms
of a history of migration from the continent with different adaptation to local
environments explained by gradual inheritable differences in fitness. Why
are the forelimbs of whales and apes functionally distinct but structurally
similar? Again, descent with modification is key.
All this is now commonplace. Still, Darwin had numerous hurdles to
overcome, in order to convince skeptical audiences. How could plants
and other less-than-mobile species cross large bodies of water to colo-
nize distant lands? How does one explain discontinuous distributions of
organisms, such as trees flourishing on mountaintops, but absent in be-
tween? Darwin’s willingness to acknowledge these and similar problems
displays his intellectual honesty. His ingenious solutions reveal his talent
for natural history.
Is that it? Is this all there is to Darwin’s contributions to evolutionary
theory? Not quite. The reason is that Darwin’s causal explanation, as just
presented, is too coarse-grained to be persuasive. For one thing, it is not
perspicuous enough. Let us grant him that natural selection is the key
difference-maker. What is descent with modification? How does it account
for geographical distribution? What are the main forces and mechanisms at
play? Unless we provide some answer to these questions, Darwin’s proposal
cannot be distinguished from other proto-evolutionary accounts, such as the
theories offered by Lamarck and by his own grandfather, Erasmus Darwin.
In short, appealing to an unqualified notion of “natural selection” was not
going to cut it. And Darwin was well aware of this.
Clearly, not all aspects could be described in detail. As discussed, there
was much that Darwin did not know. He was not in a position to pinpoint
the mechanisms of inheritance, variation, fitness, and many other core
features of his own theory. Even if he could, that would have made the ex-
planation cumbersome and overly specific. This raises important questions.
Which aspects needed to be made explicit? Which could be left tacit? How
does one denote mechanisms whose nature and structure are unknown or
too complex for the explanation at hand? Solving these problems is the task
of the representation phase of the black-boxing strategy.
The key to Darwin’s representation is his preliminary characterization
of the theory. In the initial chapters of Origin, the English naturalist breaks
down evolution by natural selection to its basic constituents: competition,
variation, fitness, and heritability. These are all placeholders in a causal ex-
planation represented in a model. These are Darwin’s black boxes.3
To illustrate, recall from section 4.2 of Chapter 4 that fitness plays a dual
role in evolution. On the one hand, it is a frame that captures distributions
of organisms and traits. On the other hand, it is a difference-maker that
stands in for mechanisms producing the distribution in question. Analogous
considerations apply to competition, variation, heritability, and other
key constituents of evolutionary theory. All of them figure prominently as
placeholders, frames, and difference- makers, in Darwin’s evolutionary
explanations.
In conclusion, are Darwin’s explanations successful? Generations of
biologists agree on a positive answer and it is not hard to see why. Darwin
framed the right questions about the evolution of species, identified the
correct difference-makers, and appropriately represented their salient
aspects in his model. Sure, there is much that Darwin did not know or
even got plainly wrong. At the time Origin was published, Darwin was
ignorant about the mechanisms of variation and inheritance, and his sub-
sequent speculative theory strikingly misses the mark. Nevertheless, igno-
rance and mistakes do not affect his main accomplishments. How can this
be so?
We now have a sketch of an answer. The model does the work ex-
pected from a scientific theory: framing, explaining, and representing
phenomena. Two kinds of placeholders allow Darwin to provide an ex-
planation that is simple, elegant, and concise. Frames pick out patterns
of behavior needing to be covered. Difference- makers stand in for
mechanisms that produce these patterns, regardless of whether these pro-
cesses are known, unknown, or partly identified. These placeholders are
Darwin’s black boxes.
3 In choosing my words, I am perfectly aware that the expression “Darwin’s black box” is some-
times used as a scornful dismissal or undermining of Darwinian explanations (Behe 1996). My
perspective is diametrically opposite. As Mayr promptly recognized, Darwin’s black boxes are what
enhance his explanations, in the face of what the great naturalist could not know.
144 Black Boxes
§6.3. Mendel’s Black Boxes
At the turn of the twentieth century, biologists noted some interesting

patterns emerging in experiments involving plant and animal breeding. The
phenomena in question can be observed when two varieties of organisms
displaying some observable phenotypic characteristics are crossed to pro-
duce viable offspring. For instance, consider a red-eyed fruit fly mating with
a white-eyed fly, or a yellow-seed pea plant crossed with a green-seed plant.
In the first generation, one of the two parental characteristics disappears,
with all the offspring consistently showing the other one. Specifically, in the
examples at hand, all fruit flies will be red-eyed, and all pea plants will be
yellow-seeded. However, if these first-generation offspring are then crossed
with one another, the second generation will display both characters present
in the two original varieties. Moreover, these traits will reappear in a specific
predictable ratio, namely, 3:1. In our examples, three-quarters of the second-
generation flies will have red eyes and the remaining quarter will have white
eyes. Similarly, three-quarters of the second-generation pea plants will be
yellow-seeded, and one quarter will be green-seeded.
There is a compelling explanation of these and similar phenotypic patterns.
Each organism contains two “factors” that determine which observable trait
will be displayed. One factor comes from each parent and, if an organism
inherits two different factors, one is always expressed preferentially over the
other. These characters remain within it for life and are passed on virtually
unchanged to the next generation. Today, grade school kids are familiar with
these factors, which are now called “genes,” a term introduced by the Danish
geneticist Wilhelm Johannsen in the early 1900s.
The preceding patterns were first observed and explained, along the lines
just described, by Mendel in a paper published in 1866. Hence, we can refer
to them as “Mendelian ratios.”4 Contrary to common wisdom, these results
were published in a respectable journal and were fairly well known among
scientists working on related problems. Yet, these observations assumed a
much broader significance at the dawn of the new century. Early geneticists
such as Bateson and Johannsen soon realized that Mendel had provided the
key to unlock Darwin’s mystery: the basic principles of heredity.
4 I borrow this expression, and much else, from Griffiths and Stotz (2013). Excellent discussions of
classical genetics can also be found in Allen (1975, 1978); Darden (1991); Morange (1998); Waters
(2004, 2006); Weber (2005); Falk (2009); and Dupré (2012).
As we saw in section 3.3 of Chapter 3, Mendel and his twentieth-century

followers—who are now called “classical geneticists”—had a better grasp
of biological inheritance compared to most naturalists of the time. Still,
this deeper understanding boils down to more accurate laws and law-like
generalizations, which constrain the structure and behavior of phenotypic
patterns. From a mechanistic standpoint, the Bohemian geneticist black-
boxed the underlying entities and processes responsible for these patterns no
less than his illustrious English colleague. This leads us back to our original
question. How did classical genetics advance the study of inheritance while
lacking a substantial grasp of the underlying cytological workings? The an-
swer emerges as soon as we break down Mendel’s use of black boxes to our
three steps.
To begin, Mendelian ratios are not universal. For one, they only apply
to inherited traits and not, for instance, to acquired, learned, or otherwise
environmentally induced characters, like sunburns or language learning.
More importantly, not all innate phenotypic features display the patterns
in question. Height and size do not follow the same rules. How fortunate of
Mendel to randomly stumble upon the right kind of traits in the right kind of
organisms. . . . Actually, the success of Mendel’s experiments was hardly the
product of luck at all. Due to his profound knowledge of botany and related
areas of biology, the Bohemian monk was perfectly aware of the importance
of selecting the appropriate plants for his inbreeding experiments.5
In his seminal 1866 article reporting his original findings, Mendel was
explicit that, to avoid the possibility of jeopardizing the chances of success
from the very beginning, the experimental plants must possess the following
features. First, they must display constant differing traits. Second, their
hybrids must be protected from the influence of all foreign pollen during the
flowering period, or easily lend themselves to such protection. Third, there
should be no marked reduction in the fertility of the hybrids and their off-
spring in successive generations. And finding the appropriate plants was only
the first step. Next, Mendel had to choose the right variety and, finally, select
suitable traits. This painstaking research is described well by Mayr:
5 Some scholars have raised doubts about Mendel’s results being “too good to be true.” Assessing
claims of Mendel’s data being “fabricated” transcends my historical competence. Fortunately, from
our present standpoint, settling this issue is not necessary to establish my main philosophical point.
146 Black Boxes
Mendel procured 34 more or less distinct varieties of peas from several

seed dealers and subjected them to two years of testing. Of these varieties,
22 remained constant when self-fertilized, and these he planted annually
throughout the entire experimental period. In these 22 varieties, seven pairs of
contrasting traits were chosen for experimental testing. Two plants differing in
a given pair of traits were hybridized, and the behavior of the trait was followed
in the ensuing generations. The 22 varieties differed from each other by far
more than the seven selected traits, but Mendel found the other traits unsuit-
able because either they produced continuous or quantitative variation not
suitable for the study of the clear-cut segregation that he was interested in, or
else they did not segregate independently. (Mayr 1982, p. 714)
This shows that, when Mendel talks about “experiments in plant hybridiza-
tion,” the expression is what I call a “frame.” It is a coarse-grained placeholder
that stands in for a particular range of behaviors in need of coverage. The ob-
ject of explanation is not inheritance simpliciter. The range of explananda is
carefully selected and embedded in a specific framework.
This point can be reinforced by looking at how Mendel’s own work was
subsequently developed in the twentieth century. Mendelian ratios are typ-
ically presented and explained by appealing to two of Mendel’s laws: the
law of segregation and the law of independent assortment.6 This reconstruc-
tion, however, is not quite accurate, from a historical perspective. This is
because, right after the “rediscovery” of Mendel’s work, the two laws were
not considered conceptually distinct. Initially, the focus was on segregation.
Independent assortment was viewed as a simple extension of segregation to
two-character cases or “dihybrid crosses.” This view was eventually aban-
doned when exceptions to independent assortment began to appear.
In 1905, Bateson, Saunders, and Punnett were able to confirm Mendel’s
law of segregation for various characters in sweet peas. At the same time,
when they tried to perform two-character crossings with purple vs. red
flowers and long vs. dry pollen, they encountered deviations from the 9:3:3:1
ratios. Simply put, the problem was that the traits “purple” and “long” dis-
played a tendency to be inherited together, and so did “red” and “round.”
6 As noted in section 3.3 of Chapter 3, paraphrased in contemporary terms, the law of segregation
states that each parent passes on only one allele to each offspring. The law of independent assortment
says that which allele an organism gets from a parent in one locus has no effect on which allele it
gets in another locus. Mendel’s third law, the law of dominance, tells us that, for every pair of alleles,
one allele is “dominant,” the other “recessive.” Organisms who have a copy of each will look just like
organisms that have two copies of the dominant allele.
The explanation of this partial coupling was eventually provided by T. H.

Morgan, who treated it as a mechanical result of the location of the factors of
the chromosomes—an idea that fueled the genetic technique now known as
“linkage mapping.”
Setting details aside, the point to be appreciated is the significance of the
choice of explananda. What we are trying to explain is just as crucial as how.
All these explanations are relativized to a context. Mendel’s background
model is constituted by his unique blend of the mathematics, physics, and
botany of his time. The model sketched by Bateson, de Vries, Johannsen,
Morgan, and other twentieth-century Mendelians was eventually molded
into the “theory” of classical genetics. Such choices of explananda can only be
understood against these theoretical and experimental backdrops, much of
which is left implicit. This is why the framing stage is so important.
We now have the resources to move on to the other two stages of black-
boxing: difference-making and representation. Beginning with the former,
Mendel does not attempt a mechanistic explanation, in the contemporary
sense of the term. As noted in section 3.3 of Chapter 3, he does not provide a
detailed description of the process of heredity. There are two reasons for this.
On the one hand, he was not in a position to describe the phenomena in ques-
tion. Furthermore, he had an altogether different goal in mind. Influenced
by his training in mathematics and physics, Mendel sought a characteriza-
tion of the laws, not the mechanisms of variation. From this perspective, we
can view Mendel’s laws as his causal explanation of heredity, framed along
the lines discussed earlier. The nature of the main difference-makers should
now become clear. What makes a difference in heredity are those entities
which Mendel called “factors” and Johannsen dubbed “genes.” This is the
backbone of the explanation provided by Mendel and subsequent classical
genetics. Genes, thus construed, are black boxes of the second type. They are
difference-makers.
This leads us to the third stage. What is there to be represented? The ex-
planation of heredity in terms of Mendel’s laws has become entrenched in
contemporary thinking. Thus, it is easy to overlook how innovative and con-
troversial it was and how much it differed from competing accounts. A brief
overview of the main players on the field will reveal the significance of the
framing and representation of these causal explanations.
One strategy to explain inheritance was to postulate some physical mech-
anism underlying and governing the process. We have already encountered a
proposal along these lines. Darwin’s “provisional hypothesis of pangenesis,”
148 Black Boxes
outlined in section 3.2 of Chapter 3, followed precisely this route. Gemmules

were introduced as the particles responsible for the transmission of herit-
able qualities. A more sophisticated version of this approach was offered,
a few years later, by the German embryologist August Weismann, who
approached inheritance from the perspective of developmental physiology.
Specifically, he postulated an elaborate hierarchy of hereditary units that
control ontogeny, ranging—in ascending order of size—from “biophores,” to
“determinants,” to “ids.” This was based on the insight that all genetic ma-
terial must be contained in the nucleus. Incidentally, this rejection of “soft”
inheritance of acquired characters is what conclusively discredited Darwin’s
pangenesis.
An altogether different approach came from biometrics. Biometricians,
such as Darwin’s half- cousin Francis Galton, believed that the aim of
studying heredity was to discover the statistical relation between features
of an organism and features of its offspring. Along these lines, Galton’s “law
of ancestral heredity” stated that one-half of the characters of an individual
is explained by the characters of its parents. One-quarter is explained by
the grandparents, one-eighth by the great-grandparents, and so forth. Karl
Pearson, another eminent biometrician, went further, explicitly stating that
Galton’s law was not a biological “law” at all. Ancestral heredity, Pearson
maintained, is a statistical measure that could be used to predict any variable
from a set of variables with which it is correlated. Galton’s weights are thus
only estimates. Actual values must be determined empirically. Influenced by
Mach’s instrumentalism, Pearson viewed the aim of science as the systemati-
zation of observation. Any attempt to go beyond what is directly observable
diverts us away from science, delving into the realm of metaphysics. From
this standpoint, Pearson viewed Darwin, Weismann, and Mendel as guilty
of the same mistake: introducing unobservable mechanisms of heredity as
objects of research.
Classical genetics, early in the twentieth century, followed a third route,
distancing itself from both previous proposals. Mendelians rejected Darwin
and Weismann’s focus on material particles as overly speculative. At the time,
there was no direct, tangible evidence for such entities. In turn, they dismissed
Galton’s and Pearson’s focus on observable patterns as simply wrongheaded.
In Johannsen’s terms, what organisms inherit are their “genotypes,” not their
“phenotypes.” This being so, the relation between the phenotypes of parent
and offspring is not the real phenomenon of heredity. It merely provides ev-
idence that can be used to investigate the relation between the genotypes of
the two organisms. Johannsen clarified this by introducing the term “gene,”
referring to the entities which constitute the organism’s genotype. Each gene
comes in a variety of alternative forms (“alleles”) and an organism contains a
number of places (“loci”), one for each gene. Thus, a genotype is the combi-
nation of all the pairs of alleles at that locus.
So, what does all of this mean? The heart of Mendelian genetics—the
gene—has a distinctive status. It is not observable. But it is more than a mere
unobservable posit to explain data. Mendel’s factors are a tool for predicting
and explaining Mendelian ratios in breeding patterns. They may stand in for
these patterns of behavior or, alternatively, for the underlying mechanisms
that make a difference to these variants. It was only natural for many
geneticists to hope that the gene, thus construed, would eventually be shown
to exist. Yet, as T. H. Morgan noted in his 1934 Nobel lecture, the centrality of
genes in early genetics did not depend, in any significant way, on their status
as physical particles. This, in short, is the theoretical role of genes, their rep-
resentation in the model of classical genetics.7
A comprehensive historical reconstruction lies beyond the scope of this
work. From our perspective, the important point is that genes are essentially
placeholders in a causal explanation represented in a model. Genes have a
multifaceted nature. They may stand in for various mechanisms or patterns
of behavior. They can be frames or difference-makers. Accordingly, these
details may be integrated in different ways or appropriately omitted. This is
why they play so many different roles. Genes are Mendel’s black boxes.
§6.4. Skinner’s Black Boxes
Behaviorism is often presented as an uncompromising attempt to trans-

form psychology from the pre-scientific study of consciousness, into a sys-
tematic empirical inquiry meant to explain, predict, and control behavior.
7 The historian Raphael Falk (2009) sums up the situation by saying that the gene of Mendelian
genetics has two separate identities: as a hypothetical material entity and as an instrumental entity.
The future development of genetics was the result of the interplay between these roles. Griffiths and
Stotz suggest a different reading: “Most recent scholars agree that the real achievement of classical
Mendelian genetics was not a theory centered on a few principles of high generality, but rather an
experimental tradition in which the practice of hybridising organisms and making inferences from
patterns of inheritance was used to investigate a wide range of biological questions” (2013, p. 17).
“Classical genetics was not a theory under test, or a theory that was simply applied to produce pre-
dictable results. It was a method of expanding biological knowledge” (2013, p. 19). Similar points
have been raised by Darden (1991); Waters (2004, 2006); and Weber (2005).
150 Black Boxes
This is clearly stated in Watson’s (1913, p. 158) manifesto, “Psychology as the

Behaviorist Views It,” which opens with the following scathing remarks:
Psychology as the behaviorist views it is a purely objective branch of nat-

ural science. Its theoretical goal is the prediction and control of behavior.
Introspection forms no essential part of its methods, nor is the scientific
value of its data dependent on the readiness with which they lend them-
selves to interpretation in terms of consciousness. The behaviorist, in his
efforts to get a unitary scheme of animal response, recognizes no dividing
line between man and brute. The behavior of man, with all of its refine-
ment and complexity, forms only a part of the behaviorist’s total scheme of
investigation.
This programmatic presentation is an oversimplification, at best (cf. §3.4,

Chapter 3). Various strands of mainstream behaviorism presuppose different
methodological directives, objectives, and goals. Sure, by the end of World
War I, the general redefinition of psychology as the scientific study of be-
havior had become widely accepted. Yet, Watson’s uncompromising—and
somewhat naïve—plan to reduce mentalistic psychology to neurophysiology
was hardly the only or even the main player in the game. Tolman’s “purposive
behaviorism” and Hull’s “mechanistic behaviorism” substantially differed
from Watson’s pioneering proposal. The same can be said for Skinner’s “rad-
ical behaviorism,” which is the object of our present focus.
Inspired by Watson, Skinner proposed a drastic overhaul of psychology
by removing the causes of behavior from organisms and placing them in the
outside environment. But, contrary to his predecessor, Skinner did not pur-
port to reduce all mentalistic concepts to neurophysiological processes. In
his academic maturity, Skinner acknowledged that a psychology dismissing
all subjective, cognitive, and affective phenomena as meaningless would fall
short of an adequate and comprehensive analysis of the human mind.
How did Skinner envision and enact his black-boxing of mental states?
In what respects was he successful? What were his main shortcomings? This
section addresses these questions by applying our three-step strategy.
Begin by focusing on the framing stage. Skinner’s explanandum is the be-
havior of organisms. This, however, is a frame, a placeholder that stands in
for a specific set of behavioral patterns, understood in a particular context
that determines what should be explained and how. The context in question
is specified in full by a model, the theory of radical behaviorism.
To appreciate the nuances and subtle complexity of the explanandum,

consider what is being rejected. Prima facie, the answer seems obvious.
As a behaviorist, Skinner had a strong aversion to traditional Cartesian
characterizations and explanations of human conduct. Throughout his
publications, he mentions various reasons for eschewing mentalism from
psychology. So, what exactly is wrong with it? Answering this question, as
Dennett (1981) notes, is not as simple as one may initially suppose. Skinner
is undoubtedly familiar with classic philosophical problems with interac-
tionism, fueled by the assumption that minds are non-physical substances.
However, this worry is often downplayed. On a related note, Skinner is also
aware that, by treating mental states as inherently private, Descartes places
them out of reach for scientific inquiry. At the same time, cognizant of the
insurmountable difficulties faced by Watson’s naïve behaviorism, Skinner
recognizes the need for science to posit some unobservable or unmeasur-
able properties. A third traditional objection to Cartesian dualism is that
it considers mental states as “internal.” While Skinner, like all behaviorists,
finds this suspicious, in his better moments he sees that there is nothing obvi-
ously wrong with positing the existence of internal mediating states.
In sum, Skinner is aware of the long-standing philosophical qualms
with Descartes’s mentalism, and he is adamant in rejecting it. Nevertheless,
pinpointing the decisive objections and, accordingly, the appropriate strategy
to overcome them is no simple matter. This makes perfect sense from our
perspective. The target is not behavior simpliciter. After all, virtually all
psychologists are explaining behavior, Cartesian mentalists and Watsonian
reductionists included. The real object of explanation is a set of behaviors
framed by a model, the theory of radical behaviorism. Talk about “explaining
behavior” is accurate. Still, we should recognize that it is a frame, a place-
holder standing in for specific patterns of behavior to be covered.
With this in mind, let us move on to Skinner’s causal explanations, which
pertain to the second phase of his black-boxing strategy. From the standpoint
of radical behaviorism, what are the difference-makers in the production of
human conduct? Within Skinner’s framework, the relevant set of behaviors
is adequately accounted for when the investigator has identified all of the
influences of which the behavior is a function. Consequently, the behavior
itself is under experimental control. The antecedent influences acting on a
behavior are the “independent” variables, whereas the behaviors which are a
function of them are the “dependent” variables.
152 Black Boxes
What exactly are these difference-makers? How should they be represented

in the model? These questions bring us to the representation stage.
To appreciate the distinctive character of Skinner’s proposal, it is useful
to compare it with a different influential perspective, equally at odds with
Cartesianism. In his article, “Critique of Psychoanalytic Concepts and
Theories,” Skinner (1956) contrasts his behaviorism with Freud’s psychoan-
alytic approach. Skinner begins by praising Freud’s general contribution to
psychology, namely, “the application of the principle of cause and effect to
human behavior. Freud demonstrated that many features of behavior hith-
erto unexplained—and often dismissed as hopelessly complex or obscure—
could be shown to be the product of circumstances in the history of the
individual” (1956, p. 77). From a behavioristic standpoint, however, Freud
was guilty of a mistake. “Freud conceived of some realm of the mind, not
necessarily having physical extent, but nevertheless capable of topographic
description and of subdivision into regions of the conscious, co-conscious,
and unconscious. Within this space, various mental events—ideas, wishes,
memories, emotions, instinctive tendencies, and so on—interacted and com-
bined in many complex ways” (1956, pp. 77–78). Freud failed to see, Skinner
maintains, how this mental apparatus is a construction, which ends up being
neither necessary nor helpful. The link between environment and behavior,
he insists, is a direct one.
This point deserves elucidation. Following Flanagan (1991), let us represent
a typical psychological causal chain in terms of three variables: environmental
events (E), mental events (M), and behavioral events (B): E → M → B. To illustrate,
I borrow an example from Leahey (2018, pp. 361–362). Consider a student who
displays neurotic subservience to their teachers. A Cartesian will rationalize
this behavior by appealing to the student’s mental states. A Freudian would add
an extra step. For instance, they might explain this by observing that, say, the
student’s father was a punitive perfectionist, who demanded unquestioned obe-
dience. This image of a stern father has been internalized by the child, thereby
affecting their behavior in the presence of authority figures. Radical behav-
iorism praises the Freudian move of recognizing how the conduct in question
has an external cause. The key to explaining the student’s actions lies in the rela-
tionship to their father. Skinner, however, takes this one step further by insisting
that the link be direct. The inference to the unconscious image of an authori-
tative father in the superego explains nothing that cannot be explained better
by reference to past behavior. The mention of this mental state adds nothing.
Actually, it unnecessarily complicates the picture, as the nature of the mental
state and its relation to environmental and behavioral variables must now be
explained. And this is problematic. Mental states cannot be observed directly
and are therefore underdetermined by evidence. This lack of depth is evidenced
by Freud’s metaphorical language. Speculative mythological stories depicting
the mysteries of the mind are so enthralling that, regardless of their accuracy
or plausibility, they end up stealing the show. This, Skinner argues, distracts
from the difference-making factors that truly cause behavior and that can thus
be used to effectively manipulate conduct. Teaching parents to raise well-func-
tioning adults requires focusing on actual child-rearing practice, not fanciful
tales about their mental life. In short, behavioral and environmental stimuli are
the difference-makers responsible for behavior and, as such, they are the key to
efficacious interventions.
We are now in a position to fully appreciate the radical behaviorist stance
in action. Skinner’s dismissal of mentalistic psychology, of both Cartesian
and Freudian ilk, rests on a general methodological argument known as the
theoretician’s dilemma.8 Return to our psychological causal chain of the E → M
→ B form. What is the relation between these relata? Logic dictates that either
mental states link environmental stimuli and conduct deterministically, or they
must do so in a non-deterministic fashion. In the first case, Skinner argues,
one can fully explain B without mentioning M at all. Internal mental states, in-
cluding neurophysiological processes, turn out to be completely unnecessary
for explaining behavior. Alternatively, if the M → B relation is indeterministic,
then M itself becomes useless and thus, Skinner claims, it should be elimi-
nated. Either way, references to mental states do not contribute to psychological
explanations and should be omitted.9 Environmental stimuli are all we need to
account for behavior.
In sum, no mental properties mediate the relation between independent
and dependent variables. References to consciousness and other superfluous
mental states are eliminated once the controlling independent variables are
properly understood. All mechanisms producing behavior are characterized
in terms of environmental stimuli—E in our schema.
8 The moniker “theoretician’s dilemma” might sound puzzling to some readers. The rationale un-
derlying this terminological choice is that, abiding to the conventions of logical empiricism, Skinner
employs the expression “theoretical term” to refer to any unobservable object, law, or process. Thus,
the name “theoreticians’ dilemma” comes from its application to non-observable entities in science.
9 As Flanagan (1991) notes, the argument, as stated, is neutral with respect to whether to eschew
E or M. The fortified version of the dilemma thus runs as follows. One ought to avoid referring to
mental events whenever possible because such references are logically eliminable and because they
are epistemically problematic and useless in practice.
154 Black Boxes
What is the nature of these difference-makers? How are they represented?

Watson maintained that the causes of behavior should be expressed as phys-
ical processes. Skinner agrees that, in principle, physiology could re-describe
these mechanisms. Yet, Skinner rejects Watson’s naïve reductionism by
making behavioral explanations autonomous from physiology.10
Mental and psychological states are Skinner’s black boxes. Nonetheless,
from his perspective, unpacking the frames and difference-makers posited
by radical behaviorism is not an endeavor that will advance psychology.
Do Skinner’s black boxes hold up to critical scrutiny? Does radical be-
haviorism provide an adequate account of human conduct and, specifically,
of our mental capacities? My negative answer will likely not surprise many
readers. Behaviorism has long been discarded in psychology, at least as a ge-
neral theory of mind, and has been replaced by various cognitive approaches.
The question that I want to raise is: what exactly was Skinner’s capital sin?
Skinner’s mistake, common to all forms of behaviorism, lies in his stub-
born insistence on expunging mental states from psychology. Both his pars
destruens and his pars construens turned out to be problematic. Beginning
with the former, Skinner’s general argument against appeals to mentality
in psychology—the theoretician’s dilemma—is fundamentally flawed. One
horn of the dilemma assumes that, if links for the form E → M and M → B are
deterministic, then appeals to M become utterly unnecessary. This is contro-
versial, to say the least. But the other horn—if such links are indeterministic,
appeals to M are useless—is completely off the mark. Non-deterministic
generalizations can be perfectly explanatory, as witnessed by their constant
and ubiquitous use across the sciences, from quantum physics to evolu-
tionary biology, from neuroscience to economics.
Moving on to the constructive portion of the proposal, behavioristic
explanations work reasonably well when applied to non-human animals,
such as rats and pigeons, working in experimental laboratory settings. This
is because, in these confined conditions, identifying the relevant stimuli and
behaviors is relatively straightforward. Yet, once we apply this methodology
to human beings “in the wild,” identifying stimuli, behaviors, and their law-
like connections becomes much harder, arguably impossible, unless we posit
a variety of logical, linguistic and, generally, cognitive competencies.11
10 As Leahey (2018, p. 363) puts it, “Skinner assumed that physiology would ultimately be able to
detail the physical mechanisms controlling behavior, but that analysis of behavior in terms of func-
tional relationships among variables is completely independent of physiology. The functions will re-
main even when the underlying physiological mechanisms are understood.”
11 For related analyses, see Chomsky (1959); Dennett (1981); and Sober (2015).
Time to take stock. Behaviorists are often criticized for black-boxing

mental states. Said critique is in need of clarification. Skinner does introduce
various placeholders in the context of his causal explanations represented in
a model. The relation between environmental stimuli, mental states, and con-
duct is described functionally. Contrary to Watson, Skinner refuses to reduce
these psychological connections to physiology. These black boxes, frames
and difference-makers alike, are no more mysterious or problematic than the
ones constructed by Darwin, Mendel, and classical geneticists in biology. In
this sense, Skinner’s decision to black box is a success story, just like the ones
discussed earlier. At the same time, Skinner’s construction of black boxes
dismisses the relevance of mental states in the production of behavior. The
result—evident in all three phases of his construction: framing, difference-
making, and representation—entails that his radical behaviorism leaves out
an important aspect of psychology: mentality. This eventually caused his ge-
neral model to collapse. This, in a nutshell, was Skinner’s capital sin.
§6.5. Friedman’s Black Boxes
At a general level of description, neoclassical economics can be viewed as an

attempt to explain economic behavior by plugging choice data into the ap-
paratus of rational choice theory. At the core of the approach lie the notions
of choice, preference, and utility. To illustrate, suppose that an agent displays
a consistent preference for a over b and for b over c. Neoclassical economics
rationalizes and accounts for such behavior by positing that the agent
attaches more utility to a than b and to b than c. In short, choice shows prefer-
ence, preference reveals utility, and utility is what explains action.
As we saw in section 3.5 of Chapter 3, the revealed preference view of
utility played an important role in turning economics into a testable science,
conforming it to the strictures of logical positivism. Preference is displayed
in choices, and choices are observable and, thereby, testable events. At the
same time, this approach completely shields economics from the influence of
any other inquiry. This is because, from the perspective under present scru-
tiny, choice and preference are not constructs open to further empirical anal-
ysis. They are “black boxes.” This section analyzes the black-boxing of mental
states enacted by neoclassical economists, by applying our three-step recipe.
To narrow the scope of the discussion, I focus on some examples famously
developed by Friedman and Savage in their classic 1948 article, “The Utility
156 Black Boxes
Analysis of Choices Involving Risk.” Beginning with the framing stage, we

can ask: what exactly is the object of their explanation?
Friedman and Savage’s point of departure is a commonsensical observa-
tion. Human beings are constantly confronted with choices that differ, among
other things, in the degree of risk to which they are subject. Should I accept a
guaranteed or virtually guaranteed lower-salary job, or should I pursue a ca-
reer that could lead to a higher income, but may also not work out? Is it wiser
to invest my life savings in secure low-yield bonds or in more volatile stocks?
Note that the problem is not restricted to economic decisions. These kinds of
choices pervade virtually every aspect of our lives.
It is not uncommon for the same individual to respond very differently
when confronted with distinct decisions along these lines. For instance,
many of us enjoy gambling and also purchase insurance on our car or home.
This raises an interesting question. Are these behaviors mutually consistent?
At first blush, it might be hard for readers without an economic background
to see why they would not be. Yet, consider the situation along the following
lines. When I decide to insure my home, I am deliberately opting for the
certainty of a small loss, the monthly premium that I pay to the insurance
company, over the relatively remote possibility of a substantial loss, say, were
the house to burn down. I know perfectly well that the “bet” I am taking is
not a fair one. My monthly premium vastly exceeds the value of the house,
minus my deductible, modulo the chance of something happening. This is
precisely how insurance companies make profit. Still, I willingly eat this rel-
atively minor cost to purchase my peace of mind. The situation is reversed
when I buy a Powerball ticket, where the mirage of a huge gain is chosen over
a sure-fire loss, namely, the cost of the ticket. Once again, there is no mystery
that the bet is not fair. The lottery prize, no matter how sizable, is never com-
mensurate to the expected value of the gamble in question. Yet, I agree to pay
the premium for a tiny chance at great riches. Here is the rub. In the insur-
ance case, I willingly pay a premium to choose certainty over uncertainty. In
the gambling case, I pay a premium to choose uncertainty over certainty. In
the former scenario, I am risk-loving; in the latter, I am risk-averse. Is it not
inconsistent for the same individual to pay a premium to avoid risk, in one
situation, and to pay a premium to seek risk, in the other? Is there any uni-
formity across these decisions?
Well into the twentieth century, the standard economic solution to this
well-known puzzle involved drawing a distinction between two kinds of
choices, “risky” and “risk-less” ones. The decision between certain options
was explained in terms of maximization of utility. Economic agents were

assumed to choose as they would if they attributed some common quanti-
tative measure, “designated utility,” to various goods. A rational subject then
selects the bundle of goods that yields the largest total amount of utility.
Choices under risk, in contrast, were explained in different terms, typically
by appealing to ignorance of the odds or to personal dispositions of character.
Why was the same explanation—maximization of utility—not applied
to both risk-less and risky choices? The reason is a direct consequence of
the principle of diminishing marginal utility, which states that as an agent
consumes more of a good, the marginal benefit of each additional unit of that
same good decreases. It seems to follow from this basic economic tenet that
no utility-maximizing agent would ever participate in a fair game of chance.
This is because diminishing marginal utility entails that the gain from win-
ning +$x will be less than the corresponding loss −$x, rendering the utility
of a fair bet negative. In short, the principle of diminishing marginal utility
and the maximization of expected utility jointly imply that agents would
have to be paid to offset the cost of bearing risk in a fair game of chance.
The premium should increase to make an otherwise unfair bet appetizing.
But this is plainly absurd. Myriad people constantly choose to partake in un-
favorable gambles: lotteries, roulettes, risky investments, and the like. Thus,
economists like Marshall rejected utility maximization as an analysis of risky
choices. These widespread behaviors had to be explained in different terms.
Friedman and Savage’s goal was to show how neoclassical economics
could resolve this apparent tension. Contrary to a then widespread belief,
risky and risk-less behavior could be subsumed under a unified explana-
tion. The solution originates from Daniel Bernoulli’s celebrated analysis of
the St. Petersburg paradox, and it was molded into its modern form by von
Neumann and Morgenstern’s Theory of Games and Economic Behavior.12 For
the sake of brevity, in what follows, most technical details will be omitted
from my reconstruction of Friedman and Savage’s analysis. My goal is to
break down their strategy in terms of our three black-boxing steps.
12 The St. Petersburg paradox, also known as the “St. Petersburg lottery,” is a puzzle related to prob-
ability and decision theory. A fair coin is flipped until it comes up heads the first time, at which
point the player wins 2n where n is the number of times the coin was flipped. How much should
one be willing to pay for playing this game? The expected monetary value approaches infinity. And
yet, it would be rejected by virtually any gambler. While the paradox was first invented by Nicolaus
Bernoulli in 1713, it takes its name from its resolution by Daniel Bernoulli, Nicolaus’s cousin, who at
the time was a resident of the eponymous Russian city.
158 Black Boxes
It should now be clear that the explanandum—why agent s chooses op-

tion a over option b, or vice versa—is far from an event simpliciter. The
target in question is framed by rational choice theory, which lies at the heart
of modern neoclassical economics. Some presuppositions of the model are
stated clearly and explicitly. For instance, rational choice theory presupposes
that, in making a decision, regardless of whether the alternatives are risky or
certain, a consumer unit behaves as if the following were true. (i) The con-
sumer unit has a consistent set of preferences. (ii) These preferences can be
exhaustively described as a function attaching a numerical value (“utility”)
to alternatives. (iii) Finally, the objective of the consumer unit is to maxi-
mize expected utility, that is, to make its expected utility as large as possible.13
Other aspects of the model are less explicit. For instance, what does it mean
for the set of preferences attributed to a consumer unit to be “consistent”?
Simply put, the requirement is that these preferences must not be contradic-
tory and should not change over time. But how strictly should the temporal
proviso be understood? On its strongest reading, an agent will never change
its preferences. This, however, will make the model virtually inapplicable to
real agents, as no human being displays identical preferences from the cradle
to the grave. At the other extreme, if any whim can be accommodated as a
change in taste, then the model becomes irrefutable, as consistency is the
basic criterion for testing the hypothesis in question.14 The solution to these
issues cannot be stated explicitly. It must be determined pragmatically, on a
case-by-case basis.
In short, when Friedman and Savage set out to provide a utility analysis
of risky choices, terms such as “utility,” “choice,” and “preference” are used
as frames. They are placeholders standing in for specific ranges of behaviors
13 This, of course, is only one among many equivalent formulations of the axioms under-
lying the concept of rational choice presupposed in the neoclassical model. A broader, alternative
characterization—which is closer to the original formalization provided by von Neumann and
Morgenstern—states that an agent must choose in accordance with a system of preferences that sat-
isfies the following properties. (a) The system is complete and consistent. (b) Any object which is a
combination of other objects with stated probabilities is never preferred to every one of these other
objects, nor is every one of them ever preferred to the combination. (c) If object a is preferred to ob-
ject b, and b to object c, there will be some probability combination of a and c such that the individual
is indifferent between it and b. This formulation shows more clearly that there is little difference be-
tween the plausibility of this hypothesis and the typical indifference-curve explanation of risk-less
choices. Yet, it is more abstract than the one provided in the main text.
14 Here is an example of behavior that would contradict the hypothesis. Imagine an individual who
is willing to pay more for a gamble than the maximum amount she could win—for instance, a gam-
bler who is willing to pay $1 for a chance to win, at most, 99¢. Such an agent displays inconsistent
preferences. Therefore, her behavior is economically “irrational” and, as such, it cannot be captured
in terms of a monotonic utility function.
in need of explanatory coverage. Precisely which behaviors are subsumed

under the analysis in question is explicitly stated or implicitly presupposed
by the underlying theoretical model: the theory of neoclassical economics.
We can now focus on the causal explanation. The difference-maker for the
framed explanandum—risky vs. risk-less choices—is the agent’s assignment
of utility to alternatives. As we saw in section 3.5 of Chapter 3, utility is a place-
holder that plays a dual role. Qua frame, it stands in for patterns of behavior
in need of explanation. Qua difference-maker, it stands in for mechanisms
producing the behavior in question, thereby causally explaining it.
How should this difference-maker be represented? This leads us to the
third and final phase of the black-boxing strategy. The explanations offered
by neoclassical economics are explicitly advanced as “as if ” claims. They
do not presuppose that the agents themselves be able to specify the utility
assigned to choices, or the precise shape of the utility function. As long as an
agent behaves as if they are maximizing expected utility, the requirements of
the framework have been met. Whether or not the agent can further provide
an insightful rationale for their choices is interesting, but beside the point.
These framing assumptions have important implications for how utility
is understood. Mainstream neoclassical economics depicts utility as a the-
oretical construct, an ordinal ranking fully revealed by preference, with no
cardinal counterpart. Utility, choice, preference, and other constructs are the
black boxes of neoclassical economics. Friedman and his colleagues are ad-
amant that any specification of the nature and structure of these boxes—for
instance, how they are psychologically or neurally implemented by human
agents—is of no consequence for economic theory itself.
Contemporary psycho-neural economics, in contrast, views utility as a
neural value, instantiated in the brain and, therefore, quantitatively meas-
urable, at least in principle. In both economic theories—neoclassical and
psycho-neural economics—utility is a black box. Nevertheless, these boxes
are structured very differently. In both cases, they are placeholders. But, except
at very general levels of description (cf. §4.4, Chapter 4), these placeholders
stand in for explananda framed differently and difference-makers distinc-
tively represented. In short, they are quite different black boxes.
Does neoclassical economics provide an accurate and illuminating model
of economic behavior? Does it capture the salient features of real-life eco-
nomic decisions involving, for instance, investments, insurance, and gam-
bling? These questions remain open. Beautifully simple mathematical
precision, or drastically oversimplified mischaracterizations with little, if
160 Black Boxes
any, empirical import? Different readers will likely disagree on this judg-
ment, based on their background and general methodological assumptions.
§6.6. Black Boxes in Science
Time to tie up some loose ends. A black box is a placeholder—frame or

difference-maker—in a causal explanation represented in a model. This
chapter has applied this general definition, first introduced in Chapter 5, to
our case studies from the history of science. The tales of Darwin, Mendel,
Skinner, and Friedman all have in common the deliberate omission of de-
tail concerning central components of their explanations. These authors con-
struct black boxes, whose basic structure is summarized in Table 6.1. Yet,
significant differences persist.
Darwin successfully black- boxed the right difference makers: the
mechanisms of variation and inheritance. Mendel identified the laws that
govern inheritance and posited the existence of an entity—the gene or, in
his words, the “factor”—that underlies this process. But, for Mendel and his
followers early in the twentieth century, genes were still black boxes. These
boxes were eventually opened, with the advent of molecular genetics, only to
be subsequently replaced with other, more sophisticated black boxes.
Skinner purported to provide a comprehensive analysis of human con-
duct which dismisses the structure of the mind as causally irrelevant.
Table 6.1 Black Boxes in the History of Science
Darwin Mendel Skinner Friedman
Step 1: Evolution of Mendelian ratios Psychological Economic

Framing species behavior behavior
Step 2: Natural Mendel’s Laws Stimulus- Maximization
Difference- selection Response of utility
making
Step 3: Evolutionary Classical genetics Radical Neoclassical
Representation theory behaviorism economics
Black Boxes Competition Mendelian factors Mental and Utility
Variation Genes (Molecular psychological decision-
Fitness genetics) states making
Inheritance (Cognitive preference
(Descent with science) (Behavioral
modification) economics)
Unfortunately for him, mental properties do play a crucial role in the framing
and production of action. His capital sin was not black-boxing per se, which
is inevitable and important across the sciences. His mistakes pervade all
three phases: framing wrong explananda, providing incomplete causal
explanations, and representing them in inadequate models.
Are Friedman’s black boxes more like Darwin’s and Mendel’s, or more
like Skinner’s? Is neoclassical economics the crown jewel of the social sci-
ences or a regressive research project? This remains controversial, dividing
economists and social scientists into opposing camps. I shall not attempt an
answer here. I do, however, contend that the present framework provides
the resources to recast, in more perspicuous and fruitful terms, the ongoing
debate between neoclassical economists and psycho-neural economists. In
doing so, it provides the foundation for more productive exchanges.
We are now in a position to fully comprehend, to a greater degree, the cen-
tral claim advanced in Chapter 3. The history of science, I maintained, is a
history of black boxes, constantly unwrapping old ones and constructing new
ones. This process can be broken down to three constitutive steps: framing
the explanandum, constructing a causal explanation, and representing these
placeholders in a model. This chapter emphasized the prominent role of
the three phases of black-boxing in the history of science. With all of this in
mind, we can now move on to the philosophical payoff of all our hard work.
Chapter 7 begins this final portion of the book with a critical discussion of
the notion of mechanism, “black-boxing style.”
7
Diet Mechanistic Philosophy
Explanatory Knowledge opens up the black boxes of nature to reveal

their inner workings.
—Wesley Salmon, Four Decades of Scientific
Explanation, p. 182
§7.1. Introduction
Chapter 5 broke down the construction of a black box into three constitutive
steps. The framing stage sharpens the object of explanation. The difference-
making stage provides a causal analysis of this explanandum by identifying
the factors which influence significantly its production. The representation
stage optimizes the explanatory “bang for the buck” by embedding the causal
narrative into a suitable model. This led to the definition of black boxes as
placeholders in causal explanations represented in models. I optimistically
hope that readers find this intuitive enough. Still, this one-liner embeds var-
ious technical notions that required some elucidation.
Chapter 6 applied this three-step recipe to the case studies first introduced
in Chapter 3. We saw that there may be various, often conflicting reasons for
idealizing or abstracting away the structural details from a causal explana-
tion. Sometimes, the underlying mechanisms are unknown, partially identi-
fied, or incorrectly understood. This situation was exemplified by the stories
of Darwin and Mendel. The English evolutionist and the Bohemian geneticist
were perfectly aware that there must be some biological apparatus governing
the process of heredity. Still, whether or not they were personally invested in
finding it, neither was able to do so successfully. In other circumstances, de-
tail may be omitted because it is, rightly or wrongly, deemed irrelevant to the
target at hand. This was the case with radical behaviorism. Skinner noted ex-
plicitly that treating organisms as black boxes is a drastic oversimplification.
Nevertheless, he considered mental states and other “internal” psycholog-
ical variables as disposable intermediaries that make little to no self-standing
DOI: 10.1093/oso/9780190095482.003.0007
Diet Mechanistic Philosophy 163
contribution to human conduct. Finally, underlying entities and processes

may also be dismissed as pertaining to an altogether different field of inquiry.
This may also become an effective strategy for insulating a theory from “ex-
ternal” critique. Friedman and other neoclassical economists have long been
aware that brain states provide the foundation for decision-making. Yet, a
systematic investigation of their nature is bracketed as an endeavor for psy-
chology and, as such, not a direct concern for economists working within
this tradition. Setting these differences aside, our celebrated episodes have
a common denominator. They all revolve around placeholders in causal
explanations represented in models. This is to say, they all involve the con-
struction of black boxes.
It is high time to address and enjoy the philosophical payoff of our hard
work. To this end, the subsequent two chapters apply the black-boxing
strategy to the concepts of emergence (Chapter 8) and scientific progress
(Chapter 9). Chapter 10 aims to show that, by focusing on black boxes, one is
able to embrace all the positives of both reductionism and antireductionism,
while eschewing their more troublesome implications. Before doing so, the
present chapter compares and contrasts the approach presented here with a
movement that has gained much traction in the last couple of decades within
the philosophy of science: the “new wave of mechanistic philosophy.”
The new mechanistic philosophy, born as a reaction to the traditional re-
ductionism vs. antireductionism divide, shares many core tenets underlying
the strategy outlined here. This raises a concern. Is my treatment of black
boxes as novel and original as I claim it to be? Or is it just a rehashing of ideas
that have been widely discussed since the turn of the new millennium? As
we shall see, the black-boxing recipe fits in quite well with the idea of science
being in the business of discovering and modeling mechanisms. At the same
time, the construction of black boxes, as I present it here, mitigates many of
the ontological implications that characterize the contemporary landscape.
For this reason, I provocatively refer to black-boxing as a “diet mechanistic
philosophy” with all the epistemic flavor of traditional views, but hardly any
metaphysical “calories.”
Here is my plan for the pages to follow. Section 7.2 outlines the central
tenets of the new mechanism. Section 7.3 emphasizes how the steps under-
lying the construction of a black box have been stressed, in some form or
degree, in the contemporary literature on the discovery and representation
of mechanisms. Still, these analogies should not be overstated. Section 7.4
stresses some differences and presents my constructive alternative. Section
164 Black Boxes
7.5 draws a connection between mechanisms and black boxes. Section 7.6
discusses some implications of the “diet” approach, and section 7.7 wraps up
the discussion with concluding remarks.
§7.2. The New Mechanistic Philosophy
The dawn of the new millennium has witnessed a revamped attention

to the notion of mechanism across the philosophy of science. This claim
calls for some clarification. Mechanism is hardly a novel concept. Natural
philosophers began to characterize the world as a complex piece of ma-
chinery and, relatedly, to see science as fundamentally organized around
the search for mechanisms at the outset of the scientific revolution, around
the late sixteenth and early seventeenth centuries. The suggestive image of a
mechanistic universe has been influential ever since. While these are impor-
tant developments in the history of science, our focus in the present chapter
is on a more recent and circumscribed phenomenon: the so-called new
mechanistic philosophy.
The new mechanistic philosophy is a movement that digs its intellectual
roots into the 1970s, when Stuart Kauffman and Bill Wimsatt pioneered
an analysis of reduction in terms of breaking down wholes into parts. This
decompositional strategy was subsequently developed by many others, in-
cluding William Bechtel, Robert Richardson, Peter Machamer, Lindley
Darden, and Carl Craver. Over the last couple of decades, the product of
these conversations has spawned into a rich and lively research program.1
What is this movement all about? According to Glennan’s (2017, p. 1)
recent presentation, the new mechanistic philosophy “says of nature that
most or all the phenomena found in nature depend on mechanisms—
collections of entities whose activities and interactions, suitably organized,
are responsible for these phenomena. It says of science that its chief busi-
ness is the construction of models that describe, predict, and explain these
mechanism-dependent phenomena.” From this perspective, most—arguably
all—phenomena found in nature depend on mechanisms. What are these
1 Without any pretense of comprehensiveness, my succinct overview borrows from a selected

handful of representative works. In particular, the following discussion draws heavily from three re-
cent monographs: Bechtel and Richardson (2010); Craver and Darden (2013); and Glennan (2017).
Incidentally, to the best of my knowledge, the expression “new mechanistic philosophy” was coined
by Skipper and Millstein (2005).
mechanisms? Unfortunately, a clear, precise, and widely accepted definition

is still wanting. Nonetheless, as the preceding quote suggests, a mechanism
can be broadly characterized as an organized aggregation of entities and ac-
tivities that collectively produces a specific phenomenon.2
Thus presented, one could legitimately wonder, what makes this mech-
anistic movement novel, substantive, or even interestingly controversial?
Who would contest the preceding pronouncements? Would anyone seri-
ously disagree with the statement that the natural world is replete with or-
ganized collections of entities and activities that produce phenomena, some
of which are explanantia or explananda of scientific interest?
Clearly, there must be more to it. And, indeed, there is. The main phil-
osophical contribution associated with the new mechanism, its signature
move, is a change in focus, namely, a shift away from the primacy of laws and
theories toward a renewed attention to mechanisms and models. This claim
calls for further elucidation. From the perspective of most twentieth-century
philosophy, science is first and foremost a search for natural laws. In general,
concrete instances of events, together with the mechanisms that produce
them, were only interesting to the extent that they provide evidence—
confirmation or disconfirmation—of these generalizations. The particularist
ontology underlying the neo-mechanistic approach purports to turn this
relation on its head. Generalizations have heuristic value. But they hardly
mirror reality, true states of affairs. What exist are mechanisms, which can
be discovered, studied, and represented in scientific models. The implication,
of course, is not to deny the existence of laws, or law-like generalizations,
which can be quite useful in predicting and explaining what happens in
the world. The point is rather that laws are less fundamental than they were
taken to be. Such generalizations are produced, sustained, and explained
by mechanisms. It is the existence of a mechanism that grounds a law, not
the other way around. In this sense, laws, like much else that goes on in sci-
ence, are mechanism-dependent: descriptions of the behavior produced by
mechanisms. These theses are more substantial than the preceding truisms.
Equipped with these insights, the mechanistic movement aims to pro-
vide a new framework in which to tackle a number of classic problems in
2 While this is undoubtedly an oversimplification, consider: “A mechanism for a phenomenon

consists of entities (or parts) whose activities and interactions are organized so as to be responsible
for the phenomenon” (Glennan 2017, p. 17). “Mechanisms are entities and activities organized such
that they are productive of regular changes from start or set-up to finish or terminating conditions”
(Craver and Darden 2013, p. 15, italics omitted).
166 Black Boxes
the philosophy of science. Central among them is the nature of explanation,

where the emphasis on mechanisms and derivative notions constitutes an
effective antidote to the outmoded deductive-nomological approach. But
the promises of the new mechanism are hardly confined to the explanatory
enterprise. The philosophy of science in the new millennium is replete with
mechanism-based accounts of causation, reduction, models, reasoning in
discovery, underdetermination, laws, theories, and other classic concepts.
How successful are these attempts? This remains an open question. Sure,
the new mechanistic philosophy has gained a large—and still growing—
number of adepts. At the same time, in addition to praise and support, the
new mechanism has also triggered objections, ranging from unimpressed
skeptics to outright critics. I defer a discussion of naysayers to section 7.7,
which sketches a black- boxing- oriented response to counterarguments.
Before addressing the negatives, let us focus on the positives. Section 7.3
discusses some remarkable analogies between mechanistic approaches and
black-boxing.
§7.3. Three Steps: A Neo-Mechanistic Perspective
The term “mechanism” has been used extensively throughout this book.
Recall our distinction between two types of placeholders. Frames stand in
for behaviors in need of explanation. Difference-makers stand in for the
mechanisms that produce the behaviors in question. Such use of “mech-
anism” falls perfectly in line with the definition presented in section
7.2: a collection of entities and activities generating a specific phenomenon.
Indeed, as we shall now see, the analogies between the new wave of mecha-
nistic philosophy and black-boxing cut much deeper than this shared min-
imal definition.
In their recent monograph, Craver and Darden (2013) provide an influ-
ential analysis which breaks down the discovery of mechanisms into four
main phases. First, one must characterize the phenomenon under investi-
gation. This involves providing a precise-enough description of the behavior
in need of explanation. The second step involves representing the mech-
anism which produces the behavior in question. This is done by constructing
a schema which generates a space of possible mechanisms responsible for
the explanandum. Third comes the evaluation of this schema. In the fourth
revisitation phase, the initial representation can be amended and improved.
This process should have a familiar ring to it. Craver and Darden’s four
stages resemble the three steps of the strategy detailed in Chapter 5. These
similarities are hardly haphazard. They are the consequence of a shared
scientific outlook. Indeed, black-boxing echoes many important insights
stressed by the new wave of mechanistic philosophers, beginning with
its broadly naturalistic stance. This section shows how all three steps in-
herent to the construction of black boxes—framing, difference-making, and
representation—are present, in some form or degree, in the neo-mechanistic
literature and, more generally, across contemporary philosophy of science.
Does this mean that black-boxing is just the repackaging of old ideas and
well-known adages? No, it does not. Section 7.4 will stress some key points
where the present account departs, quite drastically, from traditional mecha-
nistic approaches. Before doing so, however, we should focus on similarities.
Begin with the first stage of the black-boxing strategy. Recall from sec-
tion 5.2 of Chapter 5 that framing involves specifying and contextualizing
the object of explanation. Making the explanandum fully explicit requires
the construction of a model, which is no trivial task, especially in the context
of a scientific hypothesis. Fortunately, in order to get the inquiry going, it is
sufficient to specify a frame, a preliminary scaffolding that provides a coarse-
grained depiction of a range of behaviors in need of explanation.
Framing is the preliminary step in model- construction. Craver and
Darden (2013, p. 62) posit an analogous process in the building of mecha-
nistic models:
The search for mechanisms must begin with at least a rough idea of the
phenomenon that the mechanism explains. A complete characterization
of a phenomenon details its precipitating, inhibiting, and modulating
conditions, as well as noting nonstandard conditions in which it can (be
made to) occur and any of its by-products. During discovery episodes, a
purported phenomenon might be recharacterized or discarded entirely as
one learns more about the underlying mechanisms. Lumping and splitting
are two common ways of revising one’s characterization of the phenom-
enon in the search of mechanisms.
The issue here is not merely the truism that, in order to discover the mech-
anism underlying a certain kind of behavior, one must first determine which
behavior is under investigation. The contention is stronger. Mechanists
are adamant in stressing that all mechanisms are mechanisms for some
168 Black Boxes
phenomenon. This entails that the very nature and boundaries of a mech-
anism depend, in large part, on the phenomenon being produced.3
In short, the inherently perspectival identity of mechanisms and the ex-
planatory relativity underlying the black-boxing strategy point to the same
core phenomenon. The significance of a framing process lies in sharp-
ening the explanandum and determining the nature and boundaries of the
explanans.
Moving on to the second stage of black- boxing, difference-making
provides a causal explanation of the framed explanandum by identifying the
factors that make a significant contribution to its occurrence. Once again,
I do not have a novel account of causal explanation to offer. Quite frankly,
I am skeptical that that any single, monolithic definition can be nuanced
and general enough to subsume the broad range of phenomena classified as
causes. Still, black-boxing, as presented here, need not commit to any specific
approach to causal explanation. Any theory that specifies difference-makers
of an event may be employed in the construction of a black box.
These observations raise questions. How should we understand the rela-
tion between mechanisms and causation? In what ways is the black-boxing
recipe similar to and different from the new wave of mechanistic philosophy?
The exact connection between mechanisms and causation is a controver-
sial matter. In general, three categories of approaches can be distinguished.
First, there are authors who argue that mechanisms are the key for under-
standing the nature of causation. For instance, Machamer (2004), Bogen
(2008), and Glennan (2017) have articulated theories of causation, all of
which take mechanisms to be truth-makers for causal claims. A second
family of views encompasses various responses which reject the need for a
distinctively “mechanistic” approach to causation. Craver’s (2007) anal-
ysis of explanation in neuroscience, for one, borrows from manipulability
theory, a non-reductive approach falling within the difference-making camp.
3 Glennan (2017, p. 44) is explicit on this point: “The fundamental point is that boundary
drawing—whether spatial boundaries between parts of mechanisms or between a mechanism and its
environment, or temporal boundaries between the start and endpoints of an activity or mechanical
process—has an ineliminable perspectival element. But the perspectives from which these bound-
aries are drawn are not arbitrary or unconstrained. The perspective is given by identifying some phe-
nomenon. This phenomenon is a real and mind-independent feature of the world, and there are real
and independent boundaries to be found in the entities and activities that constitute the mechanism
responsible for that phenomenon.” Similarly, Craver and Darden (2013, p. 52) stress that “character-
izing the phenomenon to be explained is a vital step in the discovery of mechanisms. Characterizing
the phenomenon prunes the space of possible mechanisms (because the mechanism must explain
the phenomenon) and loosely guides the construction of this hypothesis space (because certain phe-
nomena are suggestive of possible mechanisms).”
Third, and finally, there are philosophers, such as Bechtel and Richardson
(2010) and many of their collaborators, who view mechanisms primarily as
epistemic and explanatory constructs. Consequently, they try to avoid alto-
gether the thorny ontological issue of causation. In short, there is no offi-
cial consensus among mechanists regarding the relation between causes and
mechanisms. This remains an important open issue.
It should be evident that the black-boxing strategy fits quite naturally with
the second and third paths just delineated. Any account of difference-making
can be squared with the story sketched in section 5.4 of Chapter 5 and, if
mechanisms have nothing to say about causation, we can pick our favorite
alternative. This, however, is not to say that black-boxing is flatly at odds with
mechanistic or process-based accounts of causation. Causal relevance—the
heart and soul of difference-making—is such a central concept that all the-
ories of causation and causal explanation, mechanistic or otherwise, must
say something about it. After all, can we really do science without difference-
makers or some surrogate?
In sum, the relation between mechanisms and causation is complicated.
Some contend that mechanistic philosophy has something to say about the
nature of causation. Others rest content with supplementing mechanistic
insights with preexisting concepts. Either way, any theory of causation and
causal explanation worth its salt requires a notion of relevance. If my story of
difference-making is not quite it, it must be something rather similar.
Finally, let us consider the third and final phase of black-boxing. The rep-
resentation stage seeks to embed the causal narrative into a suitable model.
This is neither original nor controversial. The importance of models and
other vehicles of representation is hardly novel. As mentioned in section 5.4,
the idea that models function as mediators between theories and the world
has become rather commonplace, at least among philosophers of science.
The new wave of mechanistic philosophy has incorporated and devel-
oped these insights by focusing—unsurprisingly—on the representation of
mechanisms. Mechanistic models describe entities and activities respon-
sible for specific behaviors. A distinction is commonly drawn between two
components. The “phenomenon description” provides a model of the expla-
nandum; the “mechanism description” depicts the system that produces the
behavior in question. Once again, this dichotomy should have a familiar ring
to it. It roughly corresponds to the two kinds of placeholders introduced in
Chapter 4. Phenomena descriptions are what I called “frames.” Mechanism
descriptions pick out difference-makers that bring about the explanandum.
170 Black Boxes
What about the processes of abstraction and idealization required to op-

timize the explanatory power of mechanistic representations? This, too, has
been extensively debated in the neo-mechanistic literature, generating a va-
riety of nuanced positions, which can be placed on a spectrum.
Some authors have acknowledged how abstracted and idealized
depictions inevitably conflict with the ideal of completeness, without af-
fecting their overall efficacy. Bechtel (2011) and Glennan (2017), for in-
stance, stress that even the most complete mechanistic models require a
certain degree of abstraction, in order to isolate their targets. Thus, even
models purporting to represent the details of actual mechanisms will
typically be highly idealized and unrealistic, in significant respects. The
gradual transition toward more and more complete models is a regulative
ideal, which involves inevitable trade-offs reflecting, in part, technological
and cognitive limitations and, in part, the diversity of explanatory models.
From this standpoint, accuracy is an important goal, albeit one that is typ-
ically bought at the expense of other theoretical virtues, such as complete-
ness and generality.
Other mechanistic philosophers, who tend to view completeness as the
chief explanatory ideal of the mechanistic sciences, are less enthusiastic about
omission of detail. Initially, the prominence of completeness over other the-
oretical virtues was formulated quite crudely (Craver 2007). This led critics
to interpret it as a problematic “the more details a mechanistic representa-
tion contains, the better” stance. More recently, Craver and his collaborators
clarified their view, which should be understood as claiming that the more
relevant details, the better (Craver and Kaplan 2020). Evidently, describing
all the entities, activities, and organizational features of a mechanism would
yield a hefty depiction, unwieldy for purposes such as prediction and control.
Hence, some degree of abstraction is required to make descriptions perspic-
uous and parsimonious, relative to their explanatory target. But this is a nec-
essary evil, not a virtue of our models.
As we shall see in section 7.5, the mechanistic approach to abstraction and
idealization—and, especially, Craver’s “3M” approach—is different from my
black-boxing strategy in significant respects. Setting these nuances aside,
the important point, for the time being, is simply the role of abstraction and
idealization in making models congruent with the portion of the world they
represent and explain. This core aspect of my third representation stage has
also been recognized and stressed in the mechanistic literature.
Time to take stock. All three constitutive phases of black-boxing—

framing, difference-making, and representation—figure, in some form
or degree, within the new mechanistic philosophy. Once again, these
considerations raise a nagging concern. Is black-boxing no more than a
repackaged version of the new mechanism? The short answer is: No.
Analogies, unquestionably robust, should not be overstated. To this point,
the remainder of the chapter emphasizes some crucial discrepancies. These
differentiate the construction of black boxes from mainstream versions of
mechanism.
§7.4. Mechanisms, Models, and Metaphysics
In a nutshell, the distinctive feature of the notion of mechanism underlying

my analysis of black boxes is its relativization to models and, consequently,
its lack of ontological import. The present section presents this context-
relativity and discusses some of its implications.
Let us begin with a truism. There is a distinction between entities in the
world and their representations. No scientist or philosopher worth their salt
denies this, with the possible exception of idealists—advocates of a radical
position that I set to the side. Mechanists are no exception. It is common
within the movement to talk about mechanisms as things “out there,” on the
one hand, and the mechanistic models that depict them, on the other.
Nevertheless, potential confusion is just around the corner. The root
problem is simple. The term “mechanism” is ambiguous. It may refer to
things in the world or to their depiction. To wit, consider the following situa-
tion, admittedly oversimplified, but not altogether unrealistic. Suppose that a
team of researchers is investigating the process of gene expression in eukary-
otic cells. In the course of the project, they construct a model of the phenom-
enon under investigation, like the one depicted in Figure 7.1, very much akin
to the ones found in elementary cytology textbooks. This raises a concern.
When we talk about, say, the “mechanism” that unwinds DNA, are we talking
about actual cellular processes or their representations?
This question will undoubtedly raise more than a few eyebrows. Asking
whether we are talking about cells or representations is foolhardy. Obviously,
we are talking about both. Models, mechanistic or otherwise, represent enti-
ties and processes in the world. Therefore, claims about representations
are, ipso facto, claims about the world. To get the point across, consider the
172 Black Boxes
DNA RNA POLYMERASE
RNA
Figure 7.1. A simple model of gene expression.
following analogy. Imagine holding a picture of the Mont Blanc and uttering,
“That mountain is beautiful!” Am I talking about the picture or the moun-
tain itself? Obviously, since the photograph is supposed to faithfully mirror
reality, I am referring to both: the mountain depicted in the photo is beau-
tiful. The same applies to genes and cells. When I claim that gene transcrip-
tion requires the unwinding of the DNA double helix, this claim holds in the
model and in reality. In short, the question whether “mechanism” refers to
models or reality is wrongheaded. It refers to both.
Intuitive as it may seem, this response is overly simplistic. Let me be very clear
that I do not intend to deny that models purport to represent bits of the world.
Of course they do. Similarly, I know better than to question the existence of a
physical reality—idealism, as noted earlier, is out of the equation. My point is
rather that the model-world relation is much more nuanced, complex, and in-
direct than the photograph analogy would suggest. And, as we shall see, this has
robust philosophical implications.
To begin, take a closer look at the diagram in Figure 7.1. This representa-
tion of the cellular environment is inaccurate in several respects. First, DNA is
depicted as a self-standing double helix in which all base pairs are available and
accessible to protein. Second, all enzymes and transcription factors are either
absent or tacitly assumed to be present in just the right quantity. Third, there are
no repression mechanisms, such as DNA methylation, that could potentially in-
terfere with gene transcription. Finally, the cellular environment is represented
as a uniform, spacious system in which molecules are free to roam aimlessly
without impediments.
In short, this simplified model represents a set of ideal conditions that are
seldom or never instantiated in reality.4 Real-life cells, it goes without saying,
are way more complicated. DNA is tightly coiled around histones in ways
that prevent enzymes from accessing regulatory or structural regions of the
gene. Methyl-groups hide nucleotides from transcription factors. Proteins
required for the expression of the gene are often absent or inaccessible. And,
finally, molecular interactions do not occur in a void but, rather, in a crowded
ecosystem. These details are not minor omissions or irrelevant by-products.
They are what turn formless blobs of cells into finely tuned organisms ca-
pable of selectively coordinating the transcription and translation of nu-
merous genes in the right place and at the right time.
These considerations invite a natural question. Why would biology
textbooks deliberately provide simplified representations that do not ac-
curately represent real organisms? The obvious answer is that biological
organisms are extremely complex. Describing all the components of an actual
cell would complicate the representation exponentially. The oversimplifica-
tion, from this standpoint, is a small price to pay in exchange for tractability.
There is surely something true captured by this intuitive answer.
Pragmatic convenience is definitely a factor. But is this all there is to it? The
answer is negative. To see why, it is crucial to draw a distinction between
abstraction, the omission of detail, and idealization, the deliberate misrep-
resentation of detail. Tractability may explain the presence of abstraction
in the model. But how do we account for the deliberate distortion in the
exact places where the mechanistic descriptions gain force? Why present a
model that, as mentioned earlier, captures neither necessary nor sufficient
conditions for gene expression?5 It is one thing to leave things out. But
why the flat misrepresentation of reality? In short, perspicuousness is only
half of the story, and not the controversial part. So, the issue remains: why
would textbooks intentionally provide inaccurate representations?
4 More precisely, whether these ideal conditions are “seldom” or “never” instantiated depends on
how strictly we interpret the representation itself (Nathan 2015). On a looser reading, while real-
life cells sometimes instantiate the conditions portrayed in the diagram, they often do not. From a
stricter perspective, one could say that the conditions depicted here are never realized. However,
some real-life conditions resemble the situation more closely than others. For our purposes, we need
not choose between these two alternatives. The important point, from the present standpoint, is that
real cellular environments are substantially different from the diagrammed ones which, sensu stricto,
fail to provide necessary or sufficient conditions for the expression of genes.
5 For a detailed discussion of how mechanistic explanations often intentionally distort the central
difference-makers in causal explanations, see Love and Nathan (2015).
174 Black Boxes
A better answer becomes available as soon as we realize that the goal of

the model is not to accurately describe cells. What the diagram purports to
do—and does so quite well—is provide an explanation of gene expression,
by showing how the underlying processes are instantiated in abstract, ide-
alized settings. When the actual cellular environment suitably resembles
or approximates the conditions depicted in the diagram, genes are, in fact
reliably transcribed. In this respect, our model is analogous to the ideal
gas law (PV = nRT) or to the identification of water and H2O. These, too,
are idealized conditions that are seldom or never instantiated. No real
substance satisfies the ideal gas law, and pure H2O is extremely rare out-
side the lab. Distorted as they are, these depictions are nonetheless highly
explanatory.
The thesis that describing and explaining are goals that, generally, do not
go hand in hand has been articulated systematically by Nancy Cartwright.
In an influential article aptly entitled, “The Truth Does Not Explain Much”
(1980), she argues that many philosophers are guilty of conflating two
functions of science that should be kept distinct: truth and explanation.
More specifically, explanation is commonly perceived as a by-product of
truth. Scientific theories are taken to explain by dint of the description
they give of reality. It follows that descriptive accuracy is all that science is
really after. Explanatory power is a welcome free lunch.
This misconception, Cartwright goes on to argue, was fostered by the
deductive-nomological model and its presupposition that all explana-
tion requires is knowledge of laws of nature, in addition to some basic
logic and probability theory. This is a mistake, and a pernicious one too.
There are perfectly good explanations that are not covered by any law. In
addition, and more provocatively, read literally, these laws are not only
false; they are known to be false. Tacitly prefixed with ceteris paribus
hedges, they only cover those sporadic cases where the conditions are
just right. Either way, this is bad news for covering-law approaches to
explanation.
Forty years after the publication of Cartwright’s essay, the deductive-
nomological model is all but dead and gone. Few philosophers still be-
lieve that explanation can be fully reduced to the logical derivation of
an explanandum from a complex explanans that contains laws of nature
together with a specification of initial conditions. Yet, interestingly, the
identification of description and explanation into a single indivisible
goal lives on in the new wave of mechanistic philosophy.6 Many con-

temporary philosophers suggest that scientific explanation is ultimately
a matter of describing mechanisms. As noted in section 7.3, whether
such explanations should be as accurate as possible remains an open and
controversial matter. Still, there is a widely shared presupposition that
mechanistic descriptions take care of both endeavors at once.7 A single
model, sufficiently rich in detail, can simultaneously provide accurate
descriptions and illuminating explanations.
What does any of this imply concerning the representation of mechanisms?
The relativization of theories, models, and inquiries points to the importance
of a different and less intuitive kind of idealization, namely, multiple-models
idealization. This is the practice of constructing multiple, related, but incom-
patible models that capture distinct and independent aspects of a complex
system.8 The idea of multiple, independent models bears significant meta-
physical implications. First, explanatory relativity threatens the ontolog-
ical import of explanation. Over the last few decades, it has become fairly
popular among philosophers to adopt an “ontic” conception of explanation.
From this perspective, “explanations are objective features of the world”
(Craver 2007, p. 27). The necessary construction of various incompatible
frameworks, inherent to multiple-models idealization, is hard to reconcile
with this kind of objectivity. Explanation is not a matter of ontology. It is ul-
timately an epistemic act, a matter of representation.9 Some mechanists have
conceded this point and have responded accordingly. For instance, Glennan
(2017, Ch. 3) endorses the importance of abstraction and idealization and
6 This has a paradoxical flavor, given the instrumental role of mechanists in replacing deductive-
nomological approaches in favor of more naturalistic analyses of explanation.
7 For instance, Glennan (2017, Ch. 3) criticizes attempts to distinguish between “phenomenal” and
“mechanistic” models. He does so on the grounds that separating these functions requires an appeal
to the intentions of the modeler.
8 The concept of multiple-model idealization is borrowed from Weisberg’s (2013, Ch. 6) three
kinds of idealization. In addition to the multiple-models approach, Weisberg adds two more variants.
The first is Galileian idealization, the practice of introducing distortions for the sake of simplifying
theories. The second, minimalist idealization, corresponds to the practice of introducing only causal
factors that make a difference to the occurrence of an event or phenomenon. All three kinds of ide-
alization play a prominent role within the black-boxing strategy. Multiple-models follow from the
framing stage. If models are relativized to the framing of an explanandum, then reframing the object
of the explanation along alternative lines will require a different model, with varying abstractions and
idealizations. Minimalist idealization occurs at the difference-making stage, when the causes that
significantly influence the (framed) explanandum are distilled. Galileian idealization chiefly pertains
to the representation stage, when the causal explanation is embedded in an appropriate model, that
should be presented as perspicuously as possible.
9 This point was originally developed in Love and Nathan (2015). For a more overarching critique
of the ontic conception of explanation, see Wright and Van Eck (2018).
176 Black Boxes
explicitly rejects completeness as a mechanistic ideal. Similarly, Bechtel and

Richardson (2010) eschew ontic approaches to explanation in favor of repre-
sentational, epistemic views.
But the importance of multiple incompatible models triggers even more
radical implications when coupled with the explanatory relativity discussed
in section 5.2 of Chapter 5. Specifically, it suggests that a single notion of
mechanism cannot simultaneously play an ontological and an epistemo-
logical role. Allow me to elaborate. At the beginning of this section, I noted
that the distinction between mechanisms and mechanistic models is both
commonplace and uncontroversial. Still, it is typical to freely employ the
term “mechanism” to refer to both entities and their representations. This
might appear like a natural move: after all, like photographs, models pur-
port to represent real-world systems. Yet, this is problematic, as the idealized
entities postulated in models are quite different from the complex universe
we inhabit. In short, real mechanisms are typically nothing like represented
mechanisms. Consequently, extrapolating properties from one to the other
requires more than a bit of caution. Employing the same term—“mech-
anism”—to refer to real and idealized entities can only foster confusion.
For this reason, I propose the following terminological convention. I shall
use the terms “process” and “system” to refer to actual entities in the world.
“Mechanism” is used exclusively to pick out model-theoretic representations
of these systems.
In conclusion, it is of paramount importance to distinguish two aspects of
the practice of introducing and modeling mechanisms: ontology and episte-
mology. This point is stressed clearly by Glennan (2016, p, 796):
Philosophers have contemplated mechanisms in connection with two dif-

ferent kinds of projects. The first is broadly metaphysical, investigating
the nature of mechanisms as an ontological category and exploring how
mechanisms may be related to or explain other ontological categories like
laws, causes, objects, processes, properties, and levels of organization. The
second is methodological and epistemological, exploring how mechanistic
concepts and representations are used to explore, describe, and explain
natural phenomena. Opinions are divided as to how these two projects
are related. Some philosophers see mechanisms as part of the deep struc-
ture of the world and ground the methodological importance of mech-
anistic thinking in this fact. Others have a more instrumentalist view in
which mechanistic thinking is an important part of science but is one that is

deeply pragmatically tinged.
On this realism vs. instrumentalism spectrum, I fall decisively in the latter

camp. “Mechanism” here becomes a pragmatic vehicle of representation, de-
void of virtually any ontological import. With these considerations in mind,
section 7.5 brings black boxes back into the equation. Section 7.6 presents
the advantages of this “diet” approach.
§7.5. Mechanisms and Black Boxes
The previous section (§7.4) attempted to draw a wedge between the meta-
physics and the epistemology of mechanisms. Specifically, I argued that
mechanisms should not be characterized as ontological posits. Mechanisms
are not things “out there” in the world. Rather, mechanisms are best under-
stood as epistemic constructs, model-theoretic vehicles of representation.
Section 7.6 will argue that, by adopting this deflationary stance, one may ef-
fectively respond to some concerns that have been raised against traditional
mechanistic theory. Before doing so, the present section discusses the con-
nection between mechanisms and black boxes. After clarifying the nature of
the relation in question, I compare and contrast my conception of black boxes
with two influential accounts taken from the two most systematic accounts
of black boxes in the philosophical literature, to the best of my knowledge.
The upshot of our discussion is that mechanisms are best understood
as vehicles of representation. Given that models may depict real systems
at different scales, it seems reasonable to insist that mechanisms are a kind
of placeholder. After all, a depiction, verbal or pictorial, “stands in” for the
system it purports to represent. All of this is, hopefully, rather intuitive.
Attentive readers, however, will surely note that this claim is in tension with
some tenets advanced in the first half of this book. Let me explain.
Back in Chapter 5, black boxes were defined as placeholders in causal
explanations represented in models. Previously, Chapter 4 distinguished
two kinds of placeholders: frames, which stand in for patterns of behaviors
in need of explanation, and difference-makers, which stand in for the
mechanisms that produce the behavior in question. It follows that one kind
of black box—namely, difference-makers—are, for all intents and purposes,
placeholders standing in for mechanisms, framed and represented in a
178 Black Boxes
model. The problem should now be evident. It looks like a mechanism is a

black box, and a black box stands in for a mechanism. But this seemingly
makes no sense: a mechanism cannot stand in for itself! What is going on?
My proposed solution is simple and, hopefully, intuitive. There is no
paradox here: mechanisms are black boxes and black boxes stand in for
mechanisms. The key lies in relativizing these statements to different levels
of description and explanation. To illustrate, consider the scenario depicted
in Figure 7.2. We have a mechanism (Mn) that is described at a specific level
(Ln). Now imagine providing a coarser description (Mn+1) of mechanism Mn
at higher-level Ln+1. Note that Mn+1 contains a black box. This black box is
a placeholder that stands in for mechanism Mn. Thus, at this coarser scale,
the mechanism described in Mn is a black box. The process also works in
the opposite direction. Mn itself will contain a number of black boxes. These
black boxes stand in for the mechanisms that are left unpacked in Mn. We
could open and, thereby, eliminate these black boxes by providing a finer-
grained description of mechanism Mn–1 at a lower level Ln–1. Doing so would
replace them with a series of mechanistic descriptions which, in turn, would
themselves contain a series of black boxes. This process may continue in-
definitely, until we reach an ontological bedrock of entities, if such ground
level even exists. Otherwise, it’s black boxes—like the famous turtles—all the
way down!
In sum, a mechanism can be defined as a model-theoretic representation
of an organized aggregation of entities and activities that collectively pro-
duce a specific phenomenon. A black box—more precisely, the type of black
Mn + 1
Mn
Mn – 1
Ln + 1
Ln
Ln – 1
Figure 7.2. Mechanisms and black boxes are relative to levels of explanation.
box that I have dubbed a “difference-maker”—is also a representation of a

causal relation relative to a certain scale. Both mechanisms and black boxes
are placeholders: they stand in for more comprehensive descriptions that
produce an effect. But there is no tension between the claims that a mech-
anism is a black box and that a black box stands in for a mechanism. The
reason is that the same model-theoretic depiction counts as a mechanism
at one level of description, and as a black box, relative to a different, coarser
grain. These considerations raise important questions concerning the rela-
tion between descriptions at different levels of the hierarchy:
. . . Ln-1, . . . Ln, . . . Ln+1, . . .
These issues will be addressed in Chapter 10. In the meantime, in the re-
mainder of this section, I further clarify my perspective on the relation be-
tween mechanisms and black boxes by comparing and contrasting it with
two influential depictions due, respectively, to Craver and Strevens.
In a classic discussion, Hanson (1963, Ch. 2) draws a distinction between
three different kinds of “boxes” in science: black, gray, and glass. Simply put,
a black box is the representation of a phenomenon where the component
functions are unknown. A gray box is a model where all components are
specified in some degree of detail. Finally, a glass box is a depiction of the
target system that is complete or, more modestly, complete for the purposes
at hand. All relevant components are exhaustively depicted. Elaborating on
this distinction, Craver characterizes the process of scientific discovery as the
gradual transformation of a black box into a gray box and, eventually, into a
glass box.10 This progressive opening of boxes is part of the general process of
turning a “mechanistic sketch”—an incomplete depiction that characterizes
parts, activities, and features of a mechanism while leaving gaps—into a
“complete mechanistic model,” that is, a comprehensive description of all
the components and activities of a mechanism. From this standpoint, a black
box is part of a mechanistic sketch. It is something that marks a “hole” in
our knowledge of how things work. This, Craver notes, does not necessarily
invalidate the description in question. Ignorance is not a crime: just ask
Darwin! Yet, these gaps are often masked by filler terms that create a mis-
leading impression that the explanation in question is complete.
10 See Craver (2007) and Craver and Darden (2013).

180 Black Boxes
Black boxes, question marks, and acknowledged filler terms are innocuous
when they stand as place-holders for future work or when it is possible to
replace the filler term with some stock-in-trade property, entity, activity, or
mechanism (as is the case for “coding” in DNA). In contrast, filler terms are
barriers to progress when they veil failures of understanding. If the term
“encode” is used to stand for “some-process-we-know-not-what,” and if the
provisional status of that term is forgotten, then one has only an illusion of
understanding. For this reason, neuroscientists often denigrate the authors
of black-box models as “diagram makers” or “boxologists.” (Craver 2007,
pp. 113–114)
As I understand him, Craver draws attention to two important points. First,

the construction of black boxes is a necessary phase in science. When dealing
with complex systems, such as the human brain, many constituents, enti-
ties, and activities are bound to be unknown. Nevertheless, explanation can
proceed steadily in the face of ignorance and error by black-boxing the un-
known processes, or parts thereof. These coarse-grained functional analyses
containing black boxes can be viewed as “mechanism sketches”.11 A good
illustration of this is provided by Darwin’s and Mendel’s treatment of the
mechanisms of inheritance. Second, there is nothing wrong with black-
boxing when the black box explicitly signals our lack of knowledge. Yet, over
time, these filler terms become increasingly crystallized and ingrained in our
practice. Consequently, their tentative, provisional status tends to be for-
gotten. When this happens—when a black box is misleadingly presented as
a gray box or as a glass box—we only have an illusion of understanding, not
true comprehension.
I agree with both remarks, which also apply to my own account. At the
same time, it is important to note that the notion of black boxes developed
in the previous chapters is significantly broader. First, Craver focuses on the
role of black boxes in covering ignorance. In contrast, I stressed that there are
other, equally important reasons for introducing a black box, from drawing
methodological boundaries to isolating a field from “external” critique.
Second, and more important, my conception of black box qua placeholder
in a causal explanation represented in a model also encompasses Hanson’s
“gray” and “glass” boxes. Allow me to elaborate.
11 The notion of mechanism sketches is developed in Piccinini and Craver (2011).

Begin by comparing the black-boxing strategy with the “more detail is al-
ways better” approach. From this standpoint, adding any kind of specificity
to a representation invariably increases its explanatory power. The outcome
of this process of including micro-constituents, taken to the limit, is a truly
complete mechanistic model that contains no black boxes or gray boxes. All
complex components have been broken down to their fundamental building
blocks, whatever that may mean. Thus understood, the mechanistic ap-
proach is fundamentally at odds with the black-boxing strategy advanced in
this book. It roughly corresponds to my crude characterization of radical re-
ductionism, from the first two chapters.
Nevertheless, it should be stressed that this “the more detail the better”
approach is not the perspective articulated and defended by Craver and his
collaborators. Indeed, Craver and Kaplan (2020) explicitly reject the “all
details are necessary” thesis as “ridiculously stringent” (p. 310). They clarify
that their position should rather be understood along the lines of a “more
relevant details are better,” according to which “if model M contains more
explanatory relevant details than M* about the [Salmon-Completeness]
mechanism for P versus P′, then M has more explanatory force than M* for P
versus P′, all things equal” (2020, p. 303).
We need not delve into the specifics of the “3M” approach.12 The impor-
tant point, for present purposes, is that, if not all details matter, then there
will be some aspects of the mechanism that are inevitably left out of the
representation. Determining which features are relevant and which are not
requires the three phases characterized in detail in Chapter 5—framing,
difference-making, and representation—or something along those lines.
What this goes to show is that my black-boxing recipe does not merely en-
compass what is commonly described as a “black box,” that is, a represen-
tation where no underlying feature has been uncovered. In addition, it also
captures placeholders where the underlying mechanisms or pattern of be-
havior is only partially understood, or whose representation is more or less
deliberately left incomplete. These correspond to the Hanson-Craver gray
12 Simply put, “3M” captures the “model-to-mechanism mapping” requirement that makes the
commitment to a mechanistic view of explanation explicit: “In successful explanatory models in
cognitive and systems neuroscience (a) the variables in the model correspond to components, ac-
tivities, properties, and organizational features of the target mechanism that produces, maintains,
or underlies the phenomenon, and (b) the (perhaps mathematical) dependencies posited among
these variables in the model correspond to the (perhaps quantifiable) causal relations among the
components of the target mechanism” (Kaplan and Craver 2011, p. 611). For a more nuanced, con-
trastive formulation, see Craver and Kaplan (2020, p. 297).
182 Black Boxes
boxes and, indeed, their “glass boxes,” once these are rightly understood in a
contextual fashion.
Another influential discussion of black boxes has been provided by
Strevens in his book Depth, where he articulates the “kairetic” strategy, the
account of causal explanation outlined in Chapter 5. Strevens claims that
explanations containing black boxes do not stand on their own. To make
his point, he begins by distinguishing two ways in which a scientific expla-
nation can be treated as “standalone.” First, an explanation stands alone if
it specifies all the difference-makers for its explanandum. An explanation
containing black boxes may technically be complete in this former sense, as
long as the black boxes stand in for the mechanisms that produce the expla-
nandum event. A second, stronger kind of standalone explanation provides
a full understanding of the explanandum in an unqualified, absolute fashion.
A “deep” standalone explanation of this kind provides an exhaustive under-
standing of the explanandum.13 Deep standalone explanations contain no
black boxes. All filler terms have been unraveled; all mysteries unveiled. Thus,
on Strevens’s picture, explanations containing black boxes are technically
complete, but only in the former, derivative sense. Black-boxed explanations,
at best, confer partial, qualified, understanding of their explanandum. This
is because they require the support of a mechanism-containing framework
in order to stand alone. These technical remarks call for some clarification.
What exactly does he have in mind?
Strevens clarifies the sense in which black boxes are only complete in a de-
rivative sense by distinguishing two situations, depending on whether or not
the black box in question is embedded in an explanatory framework. Focus,
initially, on the latter scenario, where the black box is not framed. Strevens
argues that a causal model containing a black box outside of the context of
a framework is deficient, for two reasons. First, since—on Strevens’s view,
not mine—black boxes are causally empty, they have no causal powers. Black
boxes cannot entail their target because they lack the capacity to produce
anything; they have no mechanistic components. Second, since black boxes
are multiply realizable, they are typically incohesive. The range of systems
that black boxes stand in for is so broad that the box itself does not consti-
tute a homogeneous entity. For these reasons, an explanation containing a
13 The limiting case of these “deep” standalone explanations is what Strevens calls “exhaus-
tive” standalone explanations, that is, maximally elongated, possibly infinite, indefinitely large
characterizations of the causal history of the explanandum. There seem to be some analogies here
with Craver’s notion of a “complete mechanistic model.”
black box does not stand alone. Only a “deep standalone explanation,” which
contains no black boxes and exhaustively describes all of its fundamental
constituents, can survive outside of a framework.
Now, shift to the second, alternative scenario. When a black box is un-
derstood as a placeholder for a mechanism within the context of a frame-
work, the two problems discussed in the previous paragraph do not subsist.
This is because the box stands in for a specific mechanism, specified by the
framework. This makes the black box cohesive, and it bestows upon it some
causal capacities. Thus, Strevens claims, assuming that everything else is in
order, an explanation containing a black box may stand alone. Thanks to the
tacit support of the framework, and in virtue of the causal powers of the un-
derlying homogeneous mechanism, black-boxed explanations can be causal,
cohesive, and complete. There is, however, a catch. In these circumstances,
what is being explained is not an explanandum simpliciter, an absolute event.
Rather, the object of explanation is the occurrence of something given that a
certain mechanism is in place. The framework-relativity of the explanation
thus limits the explanatory power of the model. To explain a phenomenon,
given that so-and-so is in place, Strevens claims, is a lesser achievement than
explaining a phenomenon in an absolute sense. The ultimate goal of science
is unqualified understanding, which can only be achieved via framework-
independent explanatory models. Once the interior of all black boxes is illu-
minated, their darkness will finally be dispelled.
Time for a brief summary. Strevens’s and Craver’s positions on scien-
tific explanation diverge in several important respects. For instance, while
Strevens advances an overtly reductionist stance, Craver stresses the antire-
ductionist flavor of his mechanistic approach as an integrative perspective.
Correspondingly, their analyses of black boxes go in different directions.
Nevertheless, they share some important features. For one, both philosophers
treat black boxes as placeholders standing in for mechanisms. Second, on
both views, black boxes seem to have no “productive” role to play in scientific
practice. Sure, when used correctly, they allow researchers to proceed in the
face of ignorance, postponing further inquiry until more auspicious times.
When things go south, they become an impediment to true comprehension.
Either way, we are always better off when a black box is replaced by a gray box
and, eventually, with a glass one. In other words, on both accounts, black-
boxing is something that we must come to terms with. But it is unclear that
willingly introducing a black box brings any added value to scientific know-
ledge and understanding. In Chapter 10, I shall argue that black boxes play
184 Black Boxes
a productive role in science. In the meantime, let me stress three differences

that set my analysis aside from these illustrious predecessors. As I will try to
argue, these disanalogies stem from my characterization of mechanisms as
epistemic throughout this chapter.
First, let me address a relatively minor disagreement. Strevens claims that
“to black-box an unknown mechanism is to offer an explanatory hostage to
fortune” (2008, p. 154). The reason is that, according to Strevens, an explan-
atory model containing a black box is successful only if there turns out to be
a single, cohesive, causal mechanism standing behind the black box itself. 14
In contrast, on the view defended here, the success of a black-boxed explana-
tion does not depend on the presence of a single mechanism. Sure, cohesive-
ness is always a welcome virtue, which makes explanations more elegantly
unified. But the explanatory success of a black box does not hinge on it. This
is because the number and kind of mechanisms depends inherently on the
level of description. In addition, Strevens claims that a black box without a
framework is causally empty and incohesive. I take it a step further: a black
box sans framework isn’t a black box at all!
Second, Strevens maintains, a placeholder, in and of itself, has no causal
relevance. This, he argues, resolves the long-standing puzzle of how it is
possible for a property, defined partly in terms of its ability to cause a par-
ticular event, to nevertheless usefully appear in a causal explanation of that
same event. The solution suggested by Strevens is that the property in ques-
tion does not partake in any causal explanation by means of its own dubious
causal powers. Rather, it does so in virtue of being a surrogate for something
with real causal “oomph.” This position can be reconciled with my own ac-
count. One could certainly claim that fitness has no causal power of its own.
As such, fitness only drives evolutionary forces in virtue of the underlying
properties it instantiates. At the same time, I should also stress that my view
does not presuppose or invite any causal fundamentalism of this sort, ac-
cording to which true causal relations only occur at the lowest physical level
and all that is left at higher levels is a sort of explanatory relevance. In black-
boxing a mechanism, we bestow upon it a genuine causal role, relative to a
particular model. To be sure, compared to a different level of description, the
causal “oomph” may lie elsewhere. But, from an evolutionary perspective, fit-
ness can be treated as having bona fide causal powers.
14 To be precise, Strevens notes, this becomes more of a problem when one is explaining a regu-
larity or a generalization, as opposed to an individual event, where the presence of a cohesive causal
mechanism can be usually taken for granted.
This brings us to the third and most significant difference. Strevens and
Craver both maintain that explanations containing black boxes are not ex-
haustive. But exhaustive with respect to what? Presumably, relative to the un-
derlying layer of mechanisms that, if described in detail, would render the
explanation complete. It should now be evident why this idea makes little
sense from my perspective. If the difference between mechanisms and black
boxes is only a difference in levels of description, replacing the former with
the latter will make the explanation no more exhaustive or complete. If all
scientific explanations are perspectival, they will inevitably contain boxes.
And the presence of a black box in a model does not necessarily signal the
incompleteness of the model in question. Now, to be sure, we can always
open the box and provide more detail. Yet, as we shall see in Chapter 10, this
presupposes a shift in explanandum. We are not invariably explaining more
and better. We are explaining something altogether different.
One final remark. What about mechanistic philosophers, who explicitly
advocate a less-ontological mechanistic approach? Is my “diet” mechanistic
philosophy just a version of the metaphysically sober, epistemic conception
of mechanisms advanced by Bechtel and his collaborators? It seems to me
that the answer is negative. While traditional epistemic mechanism explicitly
dampens its ontological implications, it remains committed to controversial
mechanistic tenets, such as decomposition and localization.15 In contrast,
the diet approach, which treats mechanisms as placeholders, also relaxes
these methodological constraints. While my black-boxing recipe recognizes
mechanistic explanation as one form of explanation in science, it poses
hardly any strictures on how it works. That is what ultimately distinguishes it
from extant forms of mechanistic philosophy.
In conclusion, we can all agree that black boxes are placeholders. These
placeholders may stand in for mechanisms (difference-makers), but they can
also stand in place of patterns of behavior in need of explanation (frames).
They may be introduced because of ignorance, but also to implement and
promote a specific research program. As I have defined them—placeholders
in causal explanations represented in models— black boxes are broad,
encompassing what are traditionally conceived as black, gray, and glass
boxes. Black boxes are mechanisms, and may also stand in for mechanisms,
as long as we relativize black boxes and mechanisms to different explanatory
levels. In this respect, I disagree with both Strevens’s idea of a “standalone”
15 For a discussion of these tenets, see Silberstein and Chemero (2013).

186 Black Boxes
explanation and Craver’s “ontic” conception of explanation. The key is to rec-

ognize that all scientific inquiry is framed and qualified. Unqualified under-
standing and completeness are never to be found in actual scientific practice.
At best, they are regulative ideals.
In the words of a prominent twentieth-century statistician, “Since all
models are wrong the scientist cannot obtain a ‘correct’ one by excessive
elaboration. On the contrary following William of Occam he should seek an
economical description of natural phenomena. Just as the ability to devise
simple but evocative models is the signature of the great scientist so overe-
laboration and overparameterization is often the mark of mediocrity” (Box
1976, p. 72)—nomen omen! We shall explore the implications of this claim
in Chapter 10. Before doing so, our next task is to wrap up our discussion of
mechanisms by exploring some implications of my proposal.
§7.6. What’s in a Mechanism?
Section 7.4 argued that the concept of mechanism is better understood as

a model-theoretic construct, not to be confused or conflated with physical
entities and processes interacting in the universe. What are the advantages
of this approach? This section shows that my “diet” mechanistic philosophy
successfully addresses issues with traditional mechanistic approaches.
One complaint about the new wave of mechanistic philosophy is that its
key terms—entity, activity, complex system, and the like—are slippery, rend-
ering the definiens, the concept of mechanism, hopelessly vague and am-
biguous.16 After all, virtually anything can be characterized as “organized
aggregation of entities and activities collectively producing a specific phe-
nomenon.” A concept that applies to almost everything seems trivial.
If, as some philosophers suggest, mechanism is supposed to denote a true
ontological category—some would say, a natural kind—then the worry
becomes serious. Unless one provides a clear-cut definition of what exactly
counts as a “mechanism,” its extension becomes a hodgepodge of systems,
entities, and processes with little in common except, perhaps, some generic
claim that they make something happen. What use is such a wildly heteroge-
neous ontological category for scientific and other empirical practices?
16 For a general overview and discussion of this difficulty, see Boniolo (2013).
From our diet perspective, mechanisms do not constitute a natural kind

at all. They are a pragmatic classification. We call “mechanism” the model-
theoretic representation of entities or processes that produce an effect that
we are interested in studying. This preliminary description is a placeholder
that represents a set of behaviors or the activities that bring them about. The
structure of non-fundamental mechanisms, the mechanisms described in
the special sciences, can always be further unpacked, revealing its status as
placeholders, as black boxes. In short, mechanisms play a very important role
within theory and practice. But not as an ontological category.
This brings us to a second concern. Nicholson (2012) provides a histor-
ical overview concluding that, over time, the term “mechanism” has acquired
various meanings and uses in biology. Specifically, Nicholson argues that
“mechanism” is used in the biological sciences in three senses. First, it may
refer to what he calls mechanicism, the philosophical conception of living
organisms as machines that can be exhaustively explained in terms of the
structure and interaction of their component parts. Second, “mechanism”
may refer to machine mechanism, that is, the collective internal workings of a
machine-like structure. Third, and finally, “mechanism” sometimes refers to
causal mechanism: a step-by-step explanation of the mode of operation of a
causal process that gives rise to a phenomenon of interest.
The heterogeneity of “mechanism”—a concept with a rich and nuanced
history—seems both plausible and unsurprising. But Nicholson’s point, as
I see it, is not merely the vagueness and ambiguity acknowledged in the pre-
vious paragraphs. His critique and ensuing suggestions are based on a con-
sequence of semantic conflation. The problem, in short, is that mechanisms
often end up wrongly classified as an ontological category. This has unpalat-
able consequences. As he puts it, “Mechanismic philosophers tend to con-
ceive mechanisms as real things in the world existing independently from
our conceptualization of them” (2012, p. 158). In contrast, he argues, “causal
mechanisms are better understood as heuristic explanatory devices than as
real things in nature, and [ . . . ] the reason why most mechanistic philosophers
think otherwise is because they inadvertently transpose the ontic status of
machine mechanisms onto their analyses of causal mechanisms” (p. 152).
I concur with both the diagnosis of the problem and the follow-up. As
noted in the previous pages, it is unfortunate that the same term, “mech-
anism,” is used to refer to both stuff in the world and to the model-theoretic
representation of this stuff. The standard distinction between mechanisms
and mechanistic models goes some way toward alleviating the confusion.
188 Black Boxes
But it hardly goes far enough. Furthermore, I agree with Nicholson that the
scientifically important notion is the concept of causal mechanism, which
includes heuristic devices that facilitate the explanation of phenomena. The
abstract and idealized entities postulated in models are often quite different
from the entities that are represented, and the two must be kept separate.
My solution is simple. Draw an explicit distinction between real enti-
ties and processes in the world, on the one hand, and mechanisms, under-
stood as theoretical representations, on the other. By treating mechanisms
as placeholders in causal explanations framed in models, we can eschew the
confusion noted by Nicholson. Characterized as placeholders, mechanisms
shed their ontological implications. They become boxes: black, gray, or glass,
depending on the amount of detail represented or idealized away.
Finally, let us address a third concern. In a recent article, Franklin-Hall
(2016) provides a general assessment of the philosophical impact of the
mechanistic wave. Judging by both the language of the new mechanists and
the influence of their work, one would legitimately expect that the new mech-
anism has served up a bevy of solutions. In contrast, she argues, with respect
to the central task of elucidating the nature of explanation, the movement
as a whole has hardly delivered on its promises. Interestingly, her critique is
not centered on the falsity or incorrectness of the central mechanistic tenets.
Rather, the main concern is that “mechanistic explanatory accounts offered
to date—even in their strongest formulations—have failed to move beyond
the simple and uncontroversial slogan ‘some explanations show how things
work.’ In particular, I argue that proposed constraints on mechanistic expla-
nation are not up to the task required of them: namely, that of distinguishing
acceptable explanations from those that, though minimally mechanistic, are
incontrovertibly inadequate” (2016, p. 42).
Franklin-Hall concludes that, although the new mechanistic attempt to take
scientific practice seriously—and, more generally, its naturalistic stance—is
important and admirable, much remains to be done. We need to keep looking
under the hood of explanatory practice and detail its workings. The reason is
that “rather than opening the black boxes of the scientific enterprise—with re-
spect to causation, part individuation, and explanatory level—philosophers
have largely taken those practices for granted” (2016, p. 71).
I wholeheartedly agree. Describing the mechanistic structure of a system
is only one step in a long and tortuous stride. Prediction, explanation, abduc-
tion, and other inferences require framing, identifying difference-makers,
abstraction, idealization, modeling, and many other forms of representation.
The focus should be on the nature of these epistemic practices, not on the no-
tion of mechanism, characterized as an ontological category.
There is one point of detail where I depart from Franklin-Hall’s assess-
ment. In her final footnote, she writes: “There is one mildly ironic excep-
tion to my general diagnosis. The only putative black box that mechanists
have opened is the scientists’ concept of ‘mechanism.’ On reflection, this
focus was imprudent. Not every concept used by scientists is meaty, and not
every term reflects a genuine black box; ‘mechanism’ is not a theoretical term
within the science, but a mere pointer, a placeholder—similar perhaps to the
philosopher’s term ‘conception’ ” (2016, p. 71, fn. 17). I would rephrase the
point in slightly different terms. A mechanism is, indeed, a pointer, or place-
holder. But, precisely for this reason, it is also a black box: a placeholder that
stands in for a range of behaviors or whatever produces it. Once again, the
power of black boxes should not be underestimated.
§7.7. The Diet Recipe
In conclusion, it is time to tie up some loose ends. The new mechanistic phi-
losophy is a welcome contribution to the philosophy of science. First, it has
promoted a healthy naturalistic attitude, which calls philosophers to draw
attention to the details of actual scientific practice. In addition, it offers a fur-
ther, lucid analysis of why physics should not be heralded as the paradigm of
science. Mechanistic disciplines, such as biology and neuropsychology, are
inherently different from mathematized and law-based physics. My black-
boxing approach owes a substantial intellectual debt to the new mechanists.
Still, framing, difference-making, and representation draw a wedge between
ontology and epistemology, which do not go hand in hand, as it is too fre-
quently assumed. Hence, we should not use the same term—“mechanism”—
to refer to both. In short, black-boxing tempers the ontological implications
of mechanistic approaches to science.
At the root of the trouble lies the notion of mechanism itself. Mechanisms
have a two-sided nature. On the one hand, philosophers tend to adopt an
ontology according to which mechanisms are a fundamental category or nat-
ural kind: they are real stuff in the real world. On the other hand, mechanisms
constitute a model-theoretic construction; they are a vehicle of scientific rep-
resentation. From this latter standpoint, neo-mechanists emphasize the sig-
nificance of discovering, modeling, and studying these entities. Mechanistic
190 Black Boxes
ontology and mechanistic epistemology are commonly treated as two sides

of the same coin. This seems intuitive enough. If mechanisms are the cement
of the universe, then it is hardly surprising that science seeks to discover and
represent their structure.
In contrast, I argued that a single, well-defined notion of mechanism
cannot play both roles at once. Metaphysical and epistemic considerations
require distinct conceptions of mechanism. The general problem is that
explanations and descriptions do not always proceed hand in hand—a point
famously raised by Cartwright, but often overlooked. Explanation is a par-
amount scientific goal. So is description. But one cannot generally fulfill
both endeavors at once. The black-boxing strategy is an attempt to capture
why this is the case. Truth and explanation typically presuppose different
abstractions and idealizations in the representation of frames and difference-
makers. A single model, no matter how general or overarching, is not going
to cut it.
My proposal involves a radical separation of the two dimensions of
mechanisms— metaphysics and epistemology. Mechanisms, suitably
represented and embedded in models, play a central role in the construction
of black boxes and, more generally, within scientific practice. But they do not
constitute a natural kind. For all we know, there may well be an ontological
category of “mechanism,” or something akin to it, in the world. Still, even
assuming that such a homogeneous group exists, it does not correspond to
the model-theoretic construct at the heart of scientific practice. Mechanisms
are placeholders. They stand in for entities that produce a phenomenon. This
role is epistemic. For all these reasons, I playfully refer to black-boxing as a
form of “diet” mechanistic philosophy, with all the epistemic flavor of the tra-
ditional views, but no ontological calories.
8
Emergence Reframed
In general, nature does not prepare situations to fit the kinds of

mathematical theories we hanker for. We construct both the theo-
ries and the objects to which they apply, then match them piecemeal
onto real situations, deriving—sometimes with great precision—a
bit of what happens, but generally not getting all the facts straight
at once.
—Nancy Cartwright, How the Laws of Physics Lie, p. 162
§8.1. Introduction
Let us retrace our steps back to the early stages of our journey. Reductionism
contends that science invariably advances by descending to lower
levels. Antireductionism flatly rejects this tenet. Some explanations,
antireductionists counter, cannot be enhanced by breaking them down any
further. But why should this be so? What makes explanations “autonomous”?
A popular way of cashing out the autonomy thesis, the core of antireduc-
tionism, involves the concept of emergence.
The main intuition underlying emergence is simple. As systems become
increasingly complex, they begin to display properties which, in some sense,
transcend the properties of their parts. As such, they exhibit behavior that
cannot be predicted by, explained with, or reduced to laws that govern sim-
pler entities. Common wisdom shows that there is more to a team than just
a bunch of players, and a musical symphony is not merely an arrangement
of notes. The main task of a philosophical analysis of emergence is to spell
out the “in some sense” qualifier. In what ways, if any, do emergents tran-
scend aggregative properties of their constituents? How should one under-
stand the alleged unpredictability, non-explainability, or irreducibility of the
resulting behaviors of teamwork, musical compositions, and other sophisti-
cated ensembles? Answering these questions might look simple. But it has
challenged scientists and philosophers alike for quite some time.
DOI: 10.1093/oso/9780190095482.003.0008
192 Black Boxes
The general idea of wholes transcending the sums of their parts is hardly
novel. It can already be found in Aristotle’s Metaphysics. More modern
discussions of emergence trace their roots back to J. S. Mill’s distinction be-
tween “homopathic” and “heteropathic” laws and have been shaped into
their contemporary form in the early 1900s by the work of philosophers such
as Broad, Alexander, and Lovejoy, and scientists like Morgan and Sperry. Set
aside during the heyday of logical positivism, when it was regularly dismissed
as confused and teetering on incoherence, over the last few decades, emer-
gence has worked its way back into the mainstream.
The comeback of emergence onto the main stage has had a polarizing ef-
fect, dividing the camp into enthusiastic supporters and skeptical naysayers.
Authors with antireductionist tendencies, scientists and philosophers, appeal
to emergence to capture a form of non-reductive physicalism. Reductionists,
on the other hand, treat emergence as an obscure, muddled notion that
should be definitively confined to oblivion. Once again, we are stuck between
Scylla and Charybdis. But is this the only path? Is emergence truly incompat-
ible with reduction? I suggest a negative answer.
This chapter presents, motivates, and defends a recharacterization of
emergence and its role in scientific research, grounded in our analysis
of black boxes. Here is the plan. Section 8.2 sets the stage by providing an
overview of extant accounts. Section 8.3 examines some ways in which
emergence is employed in complexity science and concludes that current
definitions have trouble accommodating such uses. Readers previously fa-
miliar with the debate should consider skipping directly to section 8.4, which
offers a constructive proposal. Emergents, I maintain, are best conceived as
black boxes: placeholders in causal explanations represented in models. My
suggestion has the welcome implications of bringing together various usages
of emergence across domains and reconciling emergence with reduction.
This, however, does come at a cost. It requires abandoning a rigid perspective
according to which emergence is an intrinsic or absolute feature of systems,
in favor of a more contextual approach that relativizes the emergent status of
a property or behavior to a specific explanatory frame of reference. The final
sections discuss refinements and implications (§8.5), as well as a few ade-
quacy conditions and advantages (§8.6) of the proposal.
Before moving on, two disclaimers are in order. First, my proposal does not
presuppose that all accounts of emergence should be judged from a scientific
perspective or, more generally, that science provides the ultimate standards
for philosophy. I simply maintain that making sense of scientific concepts is a
Emergence Reframed 193
worthwhile philosophical endeavor, and emergence is no exception. Second,

my suggestion is not to discard and replace all existing definitions of emer-
gence. My contention is that many extant analyses have locked themselves
into conceptual straitjackets by unnecessarily presupposing that emergence
is either an absolute feature of systems or is completely dependent on the
current state of our knowledge. By dropping these assumptions and rejecting
the emergence vs. reduction contrast, we make room for a definition that
reconciles seemingly incompatible aspects of emergence, providing a more
flexible, unified, and ecumenical framework.
§8.2. Emergence: Metaphysical or Epistemic?
To cash out the general thesis underlying emergence, it is helpful to return

to the rough, context-dependent partition of the natural world into dis-
tinct levels, first introduced in section 1.2 of Chapter 1. Our “wedding cake
model,” illustrated in Figure 1.1, depicted physics, biology, neuropsychology,
and economics, arranged in terms of constitution, with smaller constituents
at the bottom and larger-scale entities arranged toward the top. With this
image in mind, emergents can be characterized as higher-level properties
that, in some sense, are both dependent and autonomous from their under-
lying lower-level basis. Intuitive as it may seem, this preliminary definition
is inadequate, for two reasons. First, the combination of dependence and au-
tonomy generates an obvious tension: are these notions not contradictory?
Second, as noted at the outset (§8.1), the “in some sense” qualifier renders the
analysis hopelessly vague, bordering on vacuity. An account of emergence
worth its salt must clarify our first-pass analysis on both scores.
Our “wedding cake” layering of reality into distinct levels can be
interpreted either metaphysically or epistemically. This suggests two general
strategies for refining our preliminary sketch. On the one hand, “metaphys-
ical” approaches maintain that emergents are not exhausted, ontologically,
by their constituents. Wholes are temporally dependent (synchronically or
diachronically1) and causally, modally, or nomologically autonomous from
parts. “Epistemic” readings, on the other hand, typically agree with their
metaphysical counterparts on the temporal dependence of emergents, but
1 While I focus on synchronic dependence between emergents and their basis, these relations can
also be characterized diachronically or dynamically (Humphreys 2016).
194 Black Boxes
differ in treating wholes as epistemically, as opposed to ontologically, auton-

omous from parts. This section outlines both families of views, in turn.
A popular strategy for cashing out metaphysical emergence, which goes
all the way back to British emergentism of the early 1900s, is to analyze
“ontological autonomy” in terms of non-reducibility.2 On this view, un-
like properties that are merely resultant, emergents are not explainable or
predictable from the lower-level constituents from which they emerge, so-
called basal conditions. This is because emergent properties are, by defini-
tion, irreducible.
Compelling as it may sound, this “supervenience emergentism” faces
powerful objections.3 In work spanning a few decades, Jaegwon Kim
has argued that irreducibility runs the risk of trivializing emergence
by turning it into an empty doctrine. According to Kim, all proper-
ties are either physical properties or reducible to physical properties via
functionalization.4 From this perspective, there seems to be no room for
genuine emergents, with the possible exception of subjective states of con-
sciousness (“qualia”) and other properties that fall outside the scope of
current scientific inquiry.
In short, the main question underlying supervenience emergentism is:
can emergent properties be reduced to their bases? Emergentists argue
in the negative. Critics, such as Kim, respond that such reduction is fea-
sible, at least in principle. Kim’s influential argument has raised a host of
reactions. Is specifying the physical realization of a functional descrip-
tion sufficient for reduction? Is there a unique and complete lower-level
characterization of all higher-level phenomena? These questions should
have a familiar ring to them, as they mirror the stalemate between friends
and foes of reductionism depicted in section 2.6 of Chapter 2. And the
prospects for a positive resolution seem no better here. The concept of re-
duction, alas, is no clearer than emergence. Consequently, purporting to
explicate the latter in terms of the former will get us out of the frying pan
and right back into the fire.
One solution is to drop the problematic appeal to reduction and to de-
fine emergence in terms of other—allegedly less slippery—ontological
2 An excellent overview of early emergentism can be found in McLaughlin (1992).

3 The view under present scrutiny is sometimes dubbed “supervenience emergentism” because it
depicts emergence as a brute supervenience relation holding in virtue of fundamental emergent laws.
4 As we saw in section 2.6, “functionalizing” a property H involves describing H in terms of inputs
and outputs, and then showing how H’s causal-role description is realized at a lower level. This influ-
ential strategy is developed in detail in Kim (1999, 2005).
concepts, such as causal powers. One may thus distinguish two forms of
metaphysical emergence. “Strong emergence” occurs when higher-level
properties synchronically depend upon their basal conditions and, yet,
these macro-properties display distinctive causal powers that are absent
from lower levels. A less demanding conception, which fits in better with
the physicalist outlook by eschewing controversial forms of “downward
causation,” retains the synchronic clause. Yet, it replaces the assumption
that emergents have novel causal powers with the requirement that the
powers of emergents must be a proper subset of the set of powers found at
the basal level. “Weak emergence,” thus defined, avoids the controversial
implication that new causes may arise at higher levels. Still, it purports to
vindicate a form of non-reductive physicalism, as higher and lower levels
can be distinguished via an application of the indiscernibility of identicals
principle.5
Intuitive as they may seem, definitions of emergence appealing to causal
powers are hardly unproblematic. For one, the precise individuation of
powers constitutes a thorny challenge. The issue is that virtually any de-
scription of any phenomenon involves properties, causal or otherwise,
that are lost by ascending or descending levels. How do we determine
“novelty,” and on what basis? Are the liquidity of water and the elasticity
of rubber “distinctive” powers, or do they restate the capacities of their
constituent molecules? The bottom line is that, in spite of the struggle to
avoid the vagueness of reductionism, appeals to causal powers are plagued
by at-root identical problems: properties and their linguistic descriptions
do not wear their level and status on their sleeve. Consequently, providing
a set of clear-cut individuation criteria and illuminating analyses is no
trivial endeavor.
We shall return to causal powers and metaphysical strategies in sections
8.3 and 8.4. Before doing so, let us introduce the alternative route.
The most influential epistemic analysis of emergence was offered by
Hempel and Oppenheim, who reject the idea of emergence as an ontolog-
ical property inherent to specific systems. They characterize emergence as an
epistemological stance, indicative of the scope of our knowledge at a given
5 Simply put, the indiscernibility of identicals states that identical entities must share all properties.
Thus, if some object a lacks some property of object b, the two objects a and b can be deemed onto-
logically distinct.
196 Black Boxes
time, thus making it extrinsic or relational, as opposed to intrinsic or abso-

lute in character.
The occurrence of a characteristic W in an object w is emergent relative

to a theory T, a part relation [i.e., a part-hood relation] Pt, and a class G of
attributes if that occurrence cannot be deduced by means of T from a char-
acterization of the Pt-parts of w with respect to all attributes in G. (Hempel
and Oppenheim 1965, p. 64)
Epistemic accounts of emergence remain popular. Bedau (1997, 2002) spells

out a form of weak emergence characterized by “non-derivability sans sim-
ulation.” An effective illustration is Conway’s game of life, where the beha-
vior of the higher-level patterns is completely determined by the micro-state
and yet, because of computational complexity, one cannot predict how the
system will evolve over time without running a computer simulation.
Others refine the original insight by rendering emergence sensitive to the
kind of explanation offered. Hempel and Oppenheim treat emergence as lack
of explanation tout court. Yet, one may allow the same event to be emergent
with respect to one kind of explanation, say, mechanistic explanation, but not
with respect to a different kind, say, functional explanation.6
Time to take stock. Modern accounts of emergence come in two fami-
lies: metaphysical and epistemic.7 Traditional metaphysical approaches pur-
port to cash out ontological autonomy in terms of reducibility. More recent
endeavors replace the focus on reducibility with causal powers, fundamen-
tality, or necessitation.8 Epistemic accounts define emergence in terms of how
complex systems may (not) be explained. Differences should not be exag-
gerated. Many metaphysical accounts take epistemic features as indicative of
emergence. Conversely, explanation, or lack thereof, may often depend on
the nature of the system. Still, metaphysical and epistemic approaches disa-
gree on the constitutive features of emergence and, hence, they are best kept
distinct.
Our overview of philosophical discussions of emergence barely scratches
the surfaces of a rich, long-standing, and nuanced debate. My sketchy outline
6 For a proposal along these lines, see Taylor (2015).

7 Humphreys (2016) prefers a trichotomy between ontological, inferential, and conceptual accounts
of emergence. An even finer-grained categorization can be found in Gillett (2016). I opt for a slightly
simpler taxonomy.
8 See, respectively, Wilson (2021); Barnes (2012); and Chalmers (2006).
admittedly fails to do justice to the extant and growing literature on the topic.
Still, I hope to have done enough to motivate the significance of two general
methodological questions. First, do extant accounts capture the multifarious
ways in which emergence is employed across disciplines, effectively pro-
viding a unifying definition? Second, if they do not, should we keep using
a laden concept like “emergence,” or would we be better off replacing it with
allegedly clearer and less controversial expressions, such as “unpredictable”
and “unexplainable”? The next two sections address both issues, in turn.
§8.3. Emergence in Action: Systems Neuroscience
The previous section suggested that providing a clear, compelling, and un-
controversial definition of emergence is no trivial endeavor. There are sev-
eral promising accounts currently on the market, which I have categorized
as metaphysical vs. epistemic. While all of them pinpoint important aspects
of emergence, the question is whether any one is general and well-rounded
enough to fully capture this historically elusive concept.
Despite the lack of an uncontroversial, widely accepted analysis, the term
“emergence” continues to be employed extensively both in philosophy and
across the natural and social sciences. Fascinating examples come from com-
plexity theory, a growing discipline which encompasses a diverse array of
interdisciplinary approaches aimed at understanding and explaining sophis-
ticated biological and psychological systems. This section introduces and
discusses some illustrations. Our goal is to examine whether, and to what ex-
tent, the theoretical accounts of emergence surveyed in the preceding section
map onto empirical practice.
To get the ball rolling, let us focus our attention on a specific subfield of
complexity science: systems neuroscience. In a contribution to a recent
volume discussing prospects and challenges of contemporary neurosci-
ence, Olaf Sporns (2015) argues that, to fully understand the workings of the
brain, neuroscience must “shift perspective, towards embracing a view that
squarely acknowledges the brain as a complex networked system, with many
levels of organization, from cells to cognition that are individually irreduc-
ible and mutually interconnected” (p. 90). What motivates this shift in per-
spective, especially given the historically successful record of brain studies?
Sporns acknowledges that the so-called neuron doctrine—the hallowed tenet
that neurons are the fundamental computational units in the brain—has
198 Black Boxes
provided much insight. Yet, it is now evident that the power of neurons
derives from their collective action as parts of networks, bound together by
connections which facilitate their interaction, competition, and cooperation.
Comprehending the brain requires grasping these interconnections.
Sporns describes this breakthrough, his envisioned shift in perspective, as
the replacement of an old-fashioned account of neural circuits with a more
sophisticated alternative. Traditionally, the actions of a neural circuit were
assumed to be fully determined by the additive sum of highly specific point-
to-point exchanges of information among its elements, typically, individual
neurons. Consequently, the structure of the entire circuit was taken to be
decomposable—in principle, given complete data—into neat sequences of
individual causes and effects. Paraphrasing this into philosophical jargon,
the overall function of a circuit was assumed to be the resultant of the col-
lective interactions of individual components. In contrast, Sporns maintains,
cutting-edge approaches to complexity theory and network neuroscience
emphasize how global outcomes cannot be broken down into aggregates of
localized causes and how the functioning of the system as a whole transcends
the functioning of each individual element. In short, the lesson that neuro-
science imported from systems biology is that the behavior of cells depends
on gene regulatory networks, signal networks, and metabolic pathways,
which shape and govern interactions among individual molecules.
How is any of this relevant to our discussion? The answer is that, at this
point in Sporns’s presentation, our old friend emergence enters the pic-
ture. The proposed shift in perspective concerning neural circuits purports
to capture configurations of complex networks, which do not arise at lower
levels of organization. These are typically global states of brain dynamics in
which huge collections of neurons engage in coherent and collective beha-
vior through local interactions, individually quite weak, but aggregatively
powerful enough to generate large-scale patterns. Sporns mentions “neural
synchronization”—the coordinated, synchronized firing of large numbers of
neurons—and the “critical state,” a dynamic regime where systems engage
in a wide range of flexible and variable tasks. These, he claims, are emergent
phenomena.
Unsurprisingly, these neural processes become quite complex very fast.
Technical details need not concern us here.9 The question that I want to ad-
dress is a philosophical one. Can we adequately conceptualize the emergence
9 For an extensive but readable overview of these findings, see Sporns (2011, 2012).
of phenomena, such as neural synchronization and the so-called critical

state, in terms of irreducibility, causal powers, unpredictability, or lack of ex-
planation, that is, along the lines drawn by extant accounts of emergence?
Prima facie, there seem to be reasons for cautious optimism. Statements
along the lines of Sporns’s remarks that the functioning of complex neural
networks as a whole “transcends the functioning of each of its individual
elements” (2015, p. 92) and “give rise to new properties that do not exist or
arise at new levels of organization” (2015, p. 93) seem to fit in well with both
metaphysical and epistemic accounts of emergence. Yet, upon further scru-
tiny, complications arise, and the picture becomes less promising.
Focus, first, on classical epistemic analyses, such as Hempel and
Oppenheim’s, which characterize emergence as an overarching lack of expla-
nation. Despite consistent progress, many aspects of the human brain remain
deeply hidden mysteries. At the same time, synchronization, critical states,
and other modes of connectivity do not seem to emerge from lack of expla-
nation tout court. After all, these systems have been extensively studied and,
challenges notwithstanding, we do have detailed, if only partial, accounts.
Indeed, the mapping of a complete “connectome”—a wiring diagram of the
brain at single neuron resolution—may soon become a reality.10
To further complicate things, the classic epistemic approach does not gen-
eralize nearly enough. Even if neural systems did emerge in virtue of the cur-
rent unavailability of adequate explanation, it would be simply preposterous
to attribute the emergence of, say, liquidity, isomers, protein folding, and
many other paradigmatic examples to lack of analysis. Elementary chem-
istry textbooks provide clear and exhaustive accounts of isomers and how
hydrogen and oxygen combine to produce water. These structural proper-
ties explain the transparency and liquidity of water, its thirst-quenching ca-
pacity, and other types of higher-level behavior.11 In short, classic epistemic
accounts à la Hempel and Oppenheim fail to justify the emergent status of
neural circuits and various other scientific posits.
10 See, for example, Sporns (2012) and Zador (2015).

11 To circumvent the difficulty, one could insists that what makes these processes emergent is the
absence of complete explanation. This move runs the risk of trivializing the entire account. The reason
is that we lack a complete explanation of virtually any scientifically interesting phenomenon. The
notion of “complete explanation” is rather murky. Sure, neural systems and protein structure exceed
the grasp of a comprehensive description. But could one not say the same about gene expression,
RNA synthesis, and other cytological processes? Should one label “emergent” anything that cannot
be depicted exhaustively? When exactly, if ever, are we entitled to dub an explanation “complete”?
200 Black Boxes
What about more sophisticated epistemic routes? For instance, fol-

lowing Bedau, one could maintain that complex brain states are only deriv-
able via simulation. This, on his definition, makes them weakly emergent.
Or, perhaps, could the missing component be an appropriate explanation
of these neural states—a truly mechanistic description? I am sympathetic
to both proposals, which capture important aspects of neural complexity.
Nevertheless, explanation, or lack thereof, falls short of constituting the en-
tire story. Emergence, as it is standardly used in science, has a dimension that
transcends our current state of knowledge. Allow me to elaborate.
At the most fundamental level, what all epistemic accounts of emergents
have in common—indeed, what makes them “epistemic” in the first place—
is the presupposition that, were our mental states, cognitive capacities, or
general state of knowledge different, then some emergent properties would
no longer be emergent. To wit, if we could predict the effects of neural syn-
chronization at a glance, without the aid of simulation, then, by Bedau’s own
lights, this phenomenon would not count as emergent. Similarly, if we could
provide the appropriate kind of reconstruction, then, on all explanation-
based accounts, emergent properties would ipso facto cease to be so. I do not
doubt that many emergent properties cannot currently be explained, and
the required kind of understanding may well exceed our grasp. But is this
the decisive factor that makes them emergent? This seems disputable. Note
how, despite Sporns’s emphasis on explaining and understanding the brain,
nothing he says makes the emergence of neural systems dependent on the
state of scientific knowledge. The emergence of complex patterns is a fea-
ture of the brain itself. We can already explain a good deal about these states,
and further discoveries would not change that such complexities only arise
at higher levels of organization, in a non-additive fashion. In this respect, the
emergent systems described by Sporns fit better with traditional ontological
analyses than with their epistemic counterparts.
With this in mind, consider the prospects of reconstructing the emergence
of neural and other scientific systems in metaphysical terms. To begin, if “ir-
reducibility” is understood as “unpredictability” or “inexplicability,” then we
immediately run into all the difficulties just encountered. Given that neural
synchronization, critical states, and protein folding can already be predicted
and explained rather effectively, a fortiori, they are also predictable and ex-
plainable. Hence, let us focus on different ways of making sense of new prop-
erties arising at higher levels of organization.
Causal powers are a promising candidate, which provides precious re-

sources to analyze the logic and commitments behind scientific arguments.
In what follows, I set aside problems specific to this or other proposals, none
of which is uncontroversial. Rather, I focus on a broader, more general issue.
All attempts to analyze emergence in metaphysical terms share one short-
coming. They leave out the other side of the story: epistemology.
To clarify, let us return to Sporns’s new frontier of neuroscience. Our
understanding of the brain, he claims, is moving away from additive views,
toward more systematic approaches that reject the identification of com-
plexity with localized interactions between constituent links. Do these
higher levels exhibit genuine novel powers, the signature of strong emer-
gence, or a proper subset of more fundamental capacities, as required by
weak emergence? As noted in section 8.2, trying to pinpoint powers and
causal descriptions drags us back into the quagmire of the reductionism
vs. antireductionism rhetoric which, I argued in Chapter 2, ends up being
more verbal than substantial.
The key insight is better conceptualized along different lines. Networks are
rapidly becoming the basic units of organization in the study of the brain. The
point is not that we cannot break these systems down to their constituents
and interactions—of course we can. But doing so is futile: it has little to no
explanatory value. This interpretation of emergence eschews a pure ontolog-
ical reading, based entirely on what neural systems are, toward an epistemic
focus on how complex neural systems ought to be conceptualized and ex-
plained. This is where black boxes enter the stage.
Time to take stock. Current analyses fail to fully capture the use of emer-
gence in complexity science. Of course, the philosophical moral of our
examples is quite general. As prophetically foreseen by Russell (1927),
metaphysical stances have trouble explaining why the emergent status of
a scientific entity often involves theoretical shifts in perspective, explana-
tory refinements, and other forms of epistemic progress. At the same time,
the focus on lack of explanation tout court or of a specific kind, which
characterizes epistemic views, is not the full story either. The emergent status
of an entity or property cannot be boiled down to the state of knowledge of
specific agents or communities. In short, what is required is a conception that
combines the flexibility of epistemic accounts with the mind-independence
of ontological approaches. The next section suggests how both perspectives
can be molded into a general analysis, based on our black-boxing recipe.
202 Black Boxes
§8.4. Emergents and Black Boxes
Section 8.3 maintained that no extant philosophical analysis, either epistemic

or metaphysical in character, fully captures the role of emergence in science.
What ought one conclude from this? And how should we move forward?
One response is to relax the monistic presupposition and give up alto-
gether on the project of finding a univocal concept of emergence that spans
across the board. Developing a single definition that fixes the meaning of
emergence across, say, physics, biology, cognitive science, and economics
and connects it to all its uses in metaphysics, epistemology, ethics, and other
areas of philosophy may well be an overly ambitious endeavor. But what ex-
actly is the alternative? A more flexible, pluralistic approach that rejects a
monolithic definition of emergence in favor of local differences still raises the
question: in virtue of what are these variations of emergence?
This section sketches a proposal for systematizing various uses of emer-
gence. As stressed at the outset (§8.1), it is not my intention to discard all pre-
vious insights and replace them with a brand new analysis. The aim is rather
to provide a broader conceptual framework, a definition schema that leaves
room for different ways in which phenomena may be emergent, while cap-
turing the core features that all the overlapping uses have in common.
Here is the punchline. Contrary to many metaphysical accounts, what
makes a property, system, or process “emergent” is independent of on-
tological or causal novelty. Lack of explanation, a core tenet of epistemic
approaches, is not the decisive factor either. At the most general level, the
characteristic of an emergent is an extrinsic relation to its basal condition.
Property P is emergent, relative to a model, theory, or hypothesis H, if P is
used in a causal explanation, in the context of H, without specifying further
its nature, structure, or lower-level implementation. This, as we shall see, has
both metaphysical and epistemic components. An emergent, in this sense,
is a black box: a placeholder in a causal explanation represented in a model.
I now clarify the proposal by illustrating it with examples.
Consider the folding of protein. Is it an emergent process? Extant accounts
deliver conflicting verdicts. Is the behavior of amino acids irreducible, unpre-
dictable, or unexplainable? Is it causally novel? The answer depends on how
the explanandum is framed, explained, and represented. What degree of pre-
cision is expected from the depictions? How much structure are we allowed
to assume? Can we idealize or abstract away some of the detail? It should be
evident that, unless these preliminary questions are addressed, it becomes
meaningless to ask whether the folding of protein is “emergent” along these

lines or any other. The problem has nothing to do with the folding of protein
per se. The core issue is the vagueness of notions like reduction, causal power,
and novelty. Thus, the question remains: is the behavior of protein emergent?
And, if it is, what makes it so?
Let us explore an altogether different strategy. Set aside the phenomenon
of protein folding, in and of itself, and focus instead on how protein behavior
figures in biochemical explanations. The action of amino acids is invoked
in various molecular- biological explanatory contexts involving, for in-
stance, gene regulation and embryonic development. Regardless of whether
protein folding is an explanans or an explanandum, these depictions typi-
cally omit the details of why and how protein assumes its tertiary or quat-
ernary structure. Is it because the relevant information is unavailable? No.
Contemporary biochemistry does not treat amino acids as irreducible or un-
explainable. Is it because the underlying micro-processes make no difference
to the macro-process? Again, the answer is negative. The composition of the
solvent, the concentration of salts, pH, temperature, cofactors, and molec-
ular chaperones, to name a few cytological factors, are all crucial to the final
structure of the protein. This is the outcome of a complex, stable process
which requires a number of background conditions.
Then why are all these features often omitted? The reason is that, fasci-
nating as they are, once the folding of protein is appropriately contextualized,
these structural details are not required for describing the relevant frames
and difference-makers. Sure, all these “gory details” contribute decisively to
the folding of protein. Yet, they are neither relevant nor useful at a higher
level of description. Protein folding is emergent because it is a black box.
None of this hinges on the specifics of protein folding. Analogous con-
siderations can be applied, mutatis mutandis, to isomers, liquidity, language
processing, and several other prime examples of emergence. All these phe-
nomena play a substantive predictive and explanatory role across various
contexts. The psychological capacity to process natural language, for in-
stance, is a complex disposition that most humans acquire spontaneously
during a specific developmental window or, with much greater effort, later
in life. What makes the mastery of language “emergent” is neither the lack of
physiological, evolutionary, or neuropsychological explanations of its nature,
nor its alleged irreducibility. The core factor is rather that the—frustratingly
elusive—details of the process are irrelevant for linguistic capacities to figure
in macro-explanations, from folk-psychological scenarios (“Sam could not
204 Black Boxes
follow the movie because she does not speak German”) to serious path-
ological conditions such as aphasia, autism, or dyslexia. In short, what all
emergents have in common is that their micro-structure can be idealized and
abstracted away in the context of macro-explanations. They are black boxes
and this, I contend, makes all of them emergent.
What about neural systems? A similar analysis provides the conceptual
resources to capture the theoretical shift in neuroscience described and
prescribed by Sporns. Neural synchronization, the critical state, and other
networks can already be explained, at least partially, as witnessed by Sporns’s
own research. Furthermore, brain networks and their powers are neither
novel nor “irreducible” to individual interacting parts. What else constitutes
a circuit, in addition to a large aggregation of neurons, their additive and
non- additive interactions, and background conditions? The instructive
lesson imparted by contemporary neuroscience is that a complete descrip-
tion of brain function in terms of individual neurons and their relations is
not required for brain systems to do the explanatory heavy lifting. The reason
is that the core elements and processes that explain the workings of the brain
are difficult to describe at lower levels of organization, and thus require a shift
to a broader perspective. These macro-building blocks are, indeed, consti-
tuted by these neurons. But such configurations are stable and autonomous
enough to be identified and represented as wholes, without breaking them
down further to more fundamental units. In sum, the emergence of neural
states resides in the role of complex circuits in producing and maintaining
neuro-psychological states, which does not require a detailed account of how
these processes are physically realized at the unitary level. Thus framed, these
circuits can be effectively black-boxed. It is their status as black boxes that
makes them emergent.
I conclude this array of illustrations by briefly noting that my proposal
captures the essence of many other traditional examples of emergence.
Hempel and Oppenheim were correct, of course, that life and consciousness
are phenomena for which we could not fully account. Unfortunately, despite
scientific strides accomplished during the fifty years or so since the publica-
tion of their essay, we still cannot claim to have completely uncovered the
material underpinnings of these properties. At the same time, our incapacity
to explain qualia has not frustrated attempts by psychologists and cognitive
scientists to study conscious mental states, any more than the elusive mys-
tery of life has prevented biologists from studying, systematizing, and cat-
egorizing living organisms. Life and consciousness are emergent because
fully explaining these concepts is not a prerequisite for fruitfully appealing

to them across the sciences and the humanities. Now, surely, unraveling the
nature of life and mind would be a momentous theoretical and empirical
achievement. Nevertheless, in the meantime, these concepts can continue to
be assumed and employed as black boxes within biological, psychological,
philosophical, and other kinds of explanations.
Analogous treatments, I contend, shed light on the alleged emergence of
many other classic candidates, like the transparency and liquidity of water,
the behavior of markets and other macro-economic phenomena, and var-
ious normative properties, such as ethical and aesthetic ones. If these should
count as “emergent” at all, we can explain their emergence in terms of black
boxes: they are placeholders in causal explanations represented in models.
§8.5. Refinements and Implications
Section 8.4 outlined my emergents qua black boxes account and illustrated
it with various traditional and more cutting-edge examples. This section
refines the view by addressing three implications and a potential objection.
Beginning with the first implication, it follows from our discussion that
there are various distinct families of emergent properties. More specifically,
emergents can be arranged along two orthogonal dimensions. On the one
hand, emergents can be distinguished by the rationale for omitting detail: ne-
cessity vs. pragmatic convenience. On the other hand, emergents can be cat-
egorized depending on whether they function as explananda or explanantia,
that is, whether they play the role of frames or difference-makers relative to
an explanatory context. Let me consider both cases, in turn.
First, the micro-structure of an emergent property may be left out because
it is currently unavailable or lacking tout court. The situation is exemplified
by Hempel and Oppenheim’s treatment of qualia. Subjective mental phe-
nomena cannot (yet) be accounted for at a physical or biochemical level and,
for all we know, a satisfactory account may well be beyond the grasp of sci-
ence. Here, including the required detail is not an option. At the same time,
there are also situations in which, while the microstructure is known, it is
irrelevant for the explanation and may thus be omitted for the sake of con-
venience. This is the case with many aspects of isomers, protein folding, and
liquidity, where the amount of included detail is a matter of choice. These
two motivations for abstraction and idealization—necessity vs. pragmatic
206 Black Boxes
convenience—display substantial differences and thus require independent

treatments. Nevertheless, both share a common root: the black-boxing of
structure that is irrelevant in the context at hand. This is what makes them
instances of the same core feature: emergence.
The second distinction between kinds of emergents mirrors a funda-
mental difference between black boxes. As we saw in Chapter 4, there are two
basic types of placeholders. Frames stand in for patterns of behavior in need
of explanation. As such, they are explananda. Difference-makers, in contrast,
are explanantia standing in for the mechanisms that produce the patterns of
behavior in question. If, as I suggested, emergents are black boxes, the same
distinction will apply to these systems and properties. Indeed, examples of
both kinds are easy to find. Qualia are often treated as explananda, as they
are commonly employed in the philosophy of mind: what underlies my sen-
sation of peacefulness when I gaze at the ocean? Or they can play the role of
explanantia, as when I attribute my unwillingness to come to work to a head-
ache. In both cases, the quale is a black box, as witnessed by the lack of struc-
tural detail that does not affect the autonomy of the higher-level explanation,
when represented in the appropriate model.
In sum, we can classify emergents along two orthogonal dimensions: ne-
cessity vs. convenience, and explanantia vs. explananda (Table 8.1). What
makes all of them “emergent” is their status as black boxes.
With all of this in mind, let us move on to a second implication of my
proposal. Emergence is traditionally presented by contrasting it to reduc-
tion (§8.1). Reductionists argue that scientific explanations are invariably
deepened by breaking them down to micro-constituents and their relations.
Antireductionists retort by endorsing a form of epistemic autonomy, which
denies that the reductionist tenet generalizes across the board. Emergence is
typically viewed as a way to cash out the antireductionist thesis: higher-level
events cannot reduce to micro-levels because they are emergent. This iden-
tification of emergence with irreducibility, I have argued, is ineffective. The
notion of (anti)reduction is no clearer than the concept that it purports to
Table 8.1 Four Types of Emergents
Necessity Convenience
Frame Type 1 Type 2

Difference-maker Type 3 Type 4
elucidate, namely, emergence. Thus, explicating the former in terms of the

latter is unlikely to yield any groundbreaking insight.12
Our discussion provides an alternative. The analysis of emergents as
black boxes is consistent with both reductionist and antireductionist
outlooks. As such, it should not be contrasted with either. Allow me to elab-
orate. Modest reductionists acknowledge that higher- level explanations
are bona fide explanations. Their contention is that opening the black box
and unpacking the details from the abstract and idealized depiction will in-
crease their depth. Sophisticated antireductionists, in contrast, maintain that
the “gory details” do not necessarily enhance the explanatory power. I shall
postpone a discussion of how the debate can be recast in a more productive
fashion until Chapter 10. For the time being, the important point is that both
parties, reductionists and antireductionists, can converge in recognizing
that emergents are black boxes. The disagreement presumably lies in the fate
and role of these constructs. Reductionists will take them as provisional ep-
istemic scaffoldings awaiting to be opened and discarded. Antireductionists
will likely favor a more robust interpretation of these boxes, according to
which these will prove useful even once they have been opened. But both
sides can agree on what makes them emergent.
In sum, the existence of emergent properties and emergent systems is in-
dependent of the reductionism vs. antireductionism rhetoric. As we shall
see in the chapters to follow, there are some important differences between
thinkers belonging to the two traditions. Nevertheless, it would be a mistake
to insist that emergence be viewed in opposition to reduction.
A third noteworthy implication of my proposal is that emergence is never
an absolute state of a system. Because of the initial step of the construction of
a black box, the framing stage, the emergent status of a property or system is
always relative to a theory, hypothesis, or explanatory context. Consequently,
asking whether or not property P is “emergent” tout court is an ill-posed
question. This is because it is elliptical. P could be emergent in, say, contexts
C1 . . . Cn and not emergent in contexts C*1 . . . C*n.
A few examples should elucidate the main insight. Consider protein
folding. The structural details of how and why a protein folds are often ir-
relevant for the macro-explanation of gene expression, as witnessed by the
observation that this information is typically idealized or abstracted away
in introductory textbooks in genetics and molecular biology. At the same
12 For a related perspective, see Gillet (2016).

208 Black Boxes
time, if the explanandum is reframed as a more specific inquiry, for instance,

if the target is a depiction of the precise mechanisms governing a genetic
switch, then many cytological components will become relevant to the expla-
nandum. Protein folding is thus emergent relative to the former frame, but
not the latter. For this reason, I contend, it is meaningless to ask if this process
is emergent in a context-independent fashion.
As a different illustration, take choice in economics. Is it an emergent pro-
perty? Once again, an adequate answer requires more specificity. Consider
the situation from the perspective of neoclassical economics. All that matters
here is what an agent chooses and that the pattern of choices satisfies some
consistency constraints. The details underlying the choice in question, why
the agents choose the way they do, are completely irrelevant to the expla-
nation provided by expected utility theory. Here utility is a black box and,
therefore, it is an emergent feature. Things look very different, however,
from a psycho-neuro-economic standpoint. In this case, the psychological
mechanisms underlying choice are crucial to the explanation of economic
behavior, and cannot simply be omitted, idealized, or explained away. In this
context, choice is not a black box and thus not an emergent property at all.
So, is the notion of choice in economics emergent or not? The answer is: it
depends. Relative to the neoclassical framework, it is emergent, because it is
treated as a black box. Relative to the psycho-neural framework, in contrast,
it is not emergent because the details of the underlying mechanisms cannot
be omitted. Analogous explications, I contend, may be given for virtually any
other higher-level system and property, from Mendel’s laws to isomers, from
liquidity to qualia, from life to fitness.
I should make it very clear that the context-dependency of emergents
does not entail a rampant relativism according to which the emergent
status of an entity varies according to the particular epistemic state of indi-
vidual speakers, or even the same speaker at different times. The relativity of
emergents is coarser: it depends on the specifics of the explanation and thus
remains constant across epistemic communities. At the same time, opening
a black box does not change its emergent status in all the contexts where
such details are irrelevant. It would just specify new explanatory frameworks
where the phenomenon in question does not count as emergent.
In sum, black-boxing confers to emergence a pragmatic character that
allows one to capture the meaning and significance of this important con-
cept across philosophy and the sciences, mirroring the flexibility of epistemic
definitions. Yet, following several metaphysical approaches, it also protects
the objectivity and independence of emergence from aleatoric changes in

knowledge.
Before moving on, let me address a potential objection. Some readers
might be concerned with the breadth of the present proposal, which treats
a vast array of properties, systems, and processes as emergent. As long as
the phenomenon in question is a placeholder that stands in for behaviors
or mechanisms represented in a model—that is to say, as long as it is a black
box—the present framework will count the system in question as emergent,
even if it is fully explainable at lower, more fundamental levels. To illustrate,
in the context of an explanation of why Alex believes it is about to rain, well-
known physical facts about the atmosphere can be safely dismissed, effec-
tively rendering lightning emergent, even on the assumption that lightning
can be fully explained, and thereby reduced, to physical states. Any analysis
willing to countenance fully reducible phenomena such as lightning among
emergents, the objection runs, borders on triviality.
This worry, I maintain, is a legacy of the old reductionism vs. antireduc-
tionism rhetoric, which treats emergence and reduction as incompatible,
mutually exclusive notions. This is an assumption that, I’ve been arguing,
should be rejected. From the perspective of black boxes, reduction and emer-
gence are no longer at odds. Thus, while the framework is indeed applicable
to virtually any higher-level entity, the relativization to specific explanatory
contexts may help mitigate trivialization worries.13 Some readers may still
cringe at this reconceptualization of a traditional philosophical concept,
considering it a mischaracterization. Nevertheless, revisiting some deeply
entrenched presuppositions about emergence, I retort, is a relatively small
price to pay for a framework that promises to be broad and flexible enough to
capture and unify appeals to emergence across several domains.
§8.6. Adequacy Conditions and Advantages
This section wraps up our discussion by, first, presenting some adequacy
conditions and, subsequently, some advantages and of my proposal.
Bedau (2012) puts forward three useful criteria to assess any account of
emergence. First, he contends, the analysis should be clear and explicit, in
13 The virtual ubiquity of emergence has been noted and defended from trivialization concerns by
other authors, such as Humphreys (2016).
210 Black Boxes
particular with regard to the interpretation of dependence and autonomy,

which lie at the conceptual core of the concept (§8.1). Second, any character-
ization of emergence worth its salt must be coherent and consistent with eve-
rything we know or commonly accept in both philosophy and science. Third,
it should be useful and relevant to empirical work. I defend the promise of
my proposal by arguing that it fits the bill on all three scores.
First, it should now be clear in what sense emergents are both dependent
upon and autonomous from their basal conditions. To be sure, the iden-
tification of emergents with black boxes, in and of itself, poses hardly any
metaphysical constraint. Indeed, the black-boxing framework described ear-
lier could also be reconciled with forms of Cartesian dualism, according to
which entities belonging to different explanatory levels turn out to be dif-
ferent kinds of substances. Nevertheless, my presentation has assumed—and
strongly recommends—a naturalist interpretation according to which, in
line with modest physicalist interpretations and current understandings of
weak emergence, higher-level systems are fully constituted and exhausted
by physical entities. In this uncontroversial sense, emergents ontologically
depend on their bases. At the same time, regardless of how this token phys-
icalism is cashed out—supervenience, mereology, classical reduction, or
something else—emergents are autonomous from an explanatory perspec-
tive. As black boxes, in many contexts, their structural details may be ideal-
ized or abstracted away.
Some readers may feel that I am merely sweeping the dust under the rug.
Emergence is puzzling precisely because it is hard to see how a property or
system that is ontologically reducible from its base can also be methodolog-
ically autonomous. Simply stating that they are leaves the account as mys-
terious as previous suggestions. I agree. I have signed an “I owe you” note.
I still have not shown in what sense emergents are both reducible and auton-
omous. This chapter simply suggests that emergents can be understood as
black boxes, which entails a conditional claim: if black boxes are both auton-
omous and dependent, then so are emergents. Turning this conditional into
an indicative statement will be the task of Chapter 10.
Bedau’s second condition demands that the analysis of emergence
under scrutiny must not violate any major philosophical or scientific tenet.
Hopefully, the previous chapters have done enough to show that the black-
boxing strategy is not only compatible with mainstream scientific practice,
but actually sheds light on prominent historical episodes. In a nutshell,
readers should be convinced by now that nothing spooky is going on here.
The third and final adequacy condition requires definitions of emergence

to be useful for and relevant to empirical work. I trust that the flexibility of
the emergents qua black boxes framework is evidenced by how smoothly it
captures the various examples presented in the previous sections. We have
discussed classic paradigms of emergence, such as life and qualia. We have
explored other well-known cases: liquidity of water and protein folding. And
we have examined less familiar, cutting-edge scientific hypotheses, such as
the critical state, neural synchronization, and economic utility.
Alas, it is always hard to derive and support a general thesis from a handful
of illustrations. Still, I am optimistic that, at the very least, our survey shows
how the proposal at hand holds some promise and deserves to be explored in
greater detail. To further sweeten the deal, let me conclude our discussion by
presenting three advantages of my reframing of emergence.
First, analyzing emergents as black boxes allows one to retain the intui-
tive analysis of emergence in terms of causal powers, if one wishes to do so,
while avoiding well-known pitfalls. As we saw in section 8.2, a well-known
problem with powers is that, without a clear-cut principle for identifying
and distinguishing these capacities, it becomes susceptible to trivializa-
tion worries. A billion snowflakes have the ability to make my roof collapse,
whereas no individual snowflake has that capacity. Is this an aggregative pro-
perty of the individual snowflakes or a collective property of the snow mass?
Is this an emergent capacity at all? Does the snow mass have “novel” causal
powers that the collection of individual snowflakes lack? Do we lose or gain
abilities when shifting from additive wholes to holistic properties?
The black-boxing framework allows us to introduce talk about the causal
power of emergents without getting bogged down into these troubling issues.
From our perspective, there is no need to determine whether the powers of
the snow mass are novel, a proper subset of the collective capacities of indi-
vidual snowflakes, or something else. This terminological choice ultimately
depends on how the explanandum is framed, which is ultimately a pragmatic
matter. As long as the snow makes a difference to the collapsing of the roof,
and if this difference-maker can be black-boxed by idealizing and abstracting
away the structural details of crystallized water molecules, that is sufficient to
dub the property or behavior in question “emergent.”
A second advantage of the present analysis is that, despite its pragmatic
flavor, it can capture more robust metaphysical intuitions. Consider, for in-
stance, a proposal advanced by Barnes (2012), according to which an en-
tity x is emergent, from an ontological perspective, if and only if x is both
212 Black Boxes
fundamental and dependent. Simply put, emergents are characterized as “fun-

damental” in virtue of being part of the “real, core stuff ” in the world. At the
same time, the metaphysical dependence of emergents purports to capture
their novelty, that is, their existence over and above their basal conditions.
Barnes’s list of emergents includes minds, living beings, “gunk,” and tropes,
as well as examples from physics, such as quantum entanglement.
Prima facie, this kind of robust metaphysical account seems flatly in-
compatible with the diet mechanistic philosophy presented in Chapter 7,
which explicitly rejects ontological claims of this ilk. However, the idea of
fundamental dependency has a “diet” counterpart that can be used to pre-
serve the main spirit of the proposal. Fundamentals, after all, are prime
candidates for entities that can be assumed and employed in the con-
text of an explanation without specifying their nature and structure—in
a word, they are “primitives.” In short, the framework is consistent with
treating fundamentality or other metaphysical concepts as the conceptual
heart of many examples of emergence, if one wishes to do so. Sure, this is
an epistemic reading of fundamentality, but it is a type of fundamentality
nonetheless.
As a different illustration of how the present account can reformulate
metaphysical intuitions, consider a recent debate over the notion of emer-
gence in physics. Following Batterman (2002), Morrison (2012) argues that
a higher-level state is emergent when it is not predicted and explained from
lower-level micro-processes and is independent of them, in the sense that
changes to the micro-physical base would not affect the macro-level emer-
gent process. This proposal seems to capture rather well physical examples
of emergence, such as superconductivity. Yet, it is not broad enough to cover
applications of the concept in fields such as biology, neuroscience, and psy-
chology, where micro-processes do seem to affect the higher level but can still
be omitted from their explanations because of ignorance or pragmatic con-
venience. One response, of course, would be to dismiss many of the examples
discussed earlier as bona fide cases of emergence, reserving the label for pro-
cesses occurring at the fundamental physical levels.14 Yet, as I stressed in sec-
tion 8.4, I would rather opt for a more flexible and ecumenical approach.
Ontological physical dependence is a kind of emergence, and an important
14 This form of fundamentalism may well be what Morrison has in mind when she claims that
“epistemic independence—the fact that we need not appeal to micro-phenomena to explain macro
processes—is not sufficient for emergence since it is also a common feature of physical explanations
across many systems and levels” (Morrison 2012, p. 161).
one for sure. But there are several other reasons why structural details might
not be included in macro-explanations. All these are different, but equally le-
gitimate, cases of emergence.
A third and final advantage of my framework is that it preserves the cen-
trality of prediction and explanation in emergence, while avoiding Kim’s
trivialization worries and related concerns about emergence being spooky,
mysterious, or otherwise problematic. Emergents play an important predic-
tive and explanatory role across scientific and philosophical domains. Still,
contrary to both classic and contemporary accounts, the emergence of a
property or process is not inextricably tied to obscurity, lack of explanation,
non-derivability sans simulation, etc. To be clear, emergence can be associ-
ated to any of the preceding, as well as to causal powers, a mix of fundamen-
tality and dependence, or various other metaphysical features. Yet, none of
these conditions is necessary or sufficient for emergence. Despite substantial
ontological and epistemic differences, the common core feature of all these
properties, which makes all of them “emergents,” is their role as black boxes
in explanation.
§8.7. Concluding Remarks
In conclusion, this chapter proposed a reconceptualization of the long-

standing debate on emergence by identifying emergents with black boxes.
I suggested moving away from the idea that emergence should be char-
acterized in opposition to reduction, which is a legacy of the old reduc-
tionism vs. antireductionism debate. I tried to shift the focus toward the
explanatory autonomy of black boxes, understood independently of re-
duction. Some skeptics may view this as little more than a semantic tweak.
However, it turns out to be a surprisingly powerful move. In particular, it
allows one to hold a middle ground between a rampant monism, which
posits a single, once-size-fits-all definition of emergence, and an uncom-
promising pluralism, which treats emergence as a hodgepodge of loosely
related concepts.
As noted earlier, defining emergents as black boxes is hardly the end of
the story. I still need to show how black boxes can be both reducible and au-
tonomous, how they escape the old reductionism vs. antireductionism rhet-
oric. This will be the objective of the concluding Chapter 10. Before doing
so, in the next chapter, I want to turn my attention to another central topic
214 Black Boxes
in the philosophy of science: the question of scientific progress. For the time
being, I hope to have motivated and inspired an overarching and coherent
reframing of emergence that, nonetheless, leaves room for—and, indeed,
justifies—various metaphysical and epistemic assumptions, which have a
long and hallowed place in the history of philosophy.
9
The Fuel of Scientific Progress
The central long run philosophical problem facing people generally

is how to maintain a belief in progress without a belief in Utopia.
And I want to argue that that requires change in our conception of
progress.
—Hilary Putnam, “A Note on Progress,” p. 1
§9.1. Introduction
The previous two chapters explored some philosophical implications of the

black-boxing strategy laid out in Chapter 5 and illustrated in Chapter 6.
Specifically, Chapter 7 spelled out a “diet” approach to mechanisms that shares
a general epistemic outlook with the new wave of mechanistic philosophy
while mitigating its ontological presuppositions. Chapter 8 proposed the
reframing of an old debate by identifying emergent properties with black
boxes. The present chapter continues our philosophical exploration by
pointing our discussion toward another hallowed topic: scientific progress.
Providing the relevant background requires us, once again, to retrace our
steps back to the roots of our discipline: the dawn of contemporary philos-
ophy of science.
As we saw in Chapter 1, logical positivism built upon a popular, intuitive,
and prima facie compelling account of the scientific enterprise. From this
familiar perspective, the main goal of science is to provide a reasonably com-
plete and generally accurate description of the universe. For the most part,
positivists maintained, scientists have been successful at attaining this noble
endeavor. Successive generations of researchers have filled in more and more
details on this grand canvas depicting the universe. Of course, everyone was
perfectly aware that this grandiose wall or mosaic-like construction was
not nearly complete. Obviously, there is much that science does not know.
And parts of the story, as currently narrated, are bound to be mistaken. Still,
DOI: 10.1093/oso/9780190095482.003.0009
216 Black Boxes
positivism perceived an overall upward trend toward the attainment of truth

or, at the very least, increasingly precise approximations thereof.
Optimistically naïve as it may appear to a contemporary audience, this
view of science provided a simple but effective answer to long-standing phil-
osophical quandaries. In particular, it offered a relatively straightforward ac-
count of scientific progress, according to which science advances through
a slow, constant, and painstaking accumulation of truth. It also ascribed a
precise role to philosophy, namely, to make explicit the methodological
standards, the tacit canon of rules that underlie this quest for knowledge.
Few readers will be surprised to hear that the good old days of posi-
tivism are gone. Once-popular attempts to uncover the structure of the-
ories, the logic of confirmation, and the form of explanation are now
scornfully dismissed as ahistorical, shallow oversimplifications. In the wake
of Kuhn’s groundbreaking work, the positivist ideal was replaced by a his-
torically informed and realistic depiction of scientific theory and practice.
Importantly, Kuhn and his allies did not try to replace the formalized ac-
counts of “good” scientific reasoning advanced over the previous decades by
authors such as Popper, Carnap, Hempel, Reichenbach, and Nagel with more
apt characterizations.1 The “purely observational” vocabulary on which pos-
itivism rested the foundation and ultimate epistemic warrant of the scien-
tific edifice could not be amended and restored via a process of enrichment.
The Kuhnian blade cut deeper. It dismissed as misguided the entire project
of spelling out an objective, logical, mathematized account of scientific rea-
soning. The question then became: what should we put in its place?
Over half a century has passed since the first publication of Structure, in
1962. Decades of discussion, clarification, and refinement did not yield a
fully developed, viable replacement for the model of science presupposed
by positivism. Some of Kuhn’s most radical colleagues—most notably Paul
Feyerabend, Norbert Russell Hanson, and, later, Bruno Latour—were willing
to bite the bullet and throw in the towel. They interpreted Kuhn’s message as
flatly rejecting the existence of a single set of epistemic standards that could
objectively capture the rationality of all substantial decisions undertaken
1 As Hacking (1983) aptly notes, Kuhn did not single-handedly engineer this transformation in the
history and philosophy of science. When Structure was first published in 1962, similar themes were
being expressed by a number of voices and the history of science was forming itself as a self-standing
discipline. The fundamental transformation in philosophical perspective was that science was finally
becoming a historical phenomenon.
The Fuel of Scientific Progress 217
across and within fields. In Feyerabend’s (1975) provocative slogan, when it

comes to scientific method and rationality, “anything goes.”
Kuhn himself, however, was not quite willing to take it that far. In subse-
quent publications, he firmly resisted the complete relinquishment of sys-
tematic discussions of rationality, in favor of sociological explanations of
scientific behavior. In particular, adamantly rejecting charges of “mob psy-
chology,” Kuhn insisted on the existence of bona fide epistemic methodo-
logical standards which demarcate “good” from “bad” instances of scientific
reasoning. And yet, contrary to a core tenet of positivist orthodoxy, these
rules are neither algorithmic nor perfectly objective. Consequently, they
cannot be articulated generally or made fully explicit. Nevertheless, Kuhn
stressed, science is a rational endeavor and must be portrayed as such.
Over the years, there have been remarkable attempts to embed Kuhn’s
insights into a rigorous framework accounting for scientific rationality and
success.2 However, around the dawn of the new millennium, mainstream
philosophy of science underwent a paradigm shift of its own. The field even-
tually abandoned the project of developing a grand overarching account of
the scientific enterprise. New approaches traded in the search for generality
for a focus on particular sub-disciplines: physics, biology, neuropsychology,
and the like. As a result, Kuhn’s expensive tab is still largely unsettled.
The systematic development of a post-Kuhnian replacement for the out-
moded positivist image of science remains a prominent item on philosophy’s
unfinished agenda. This is a substantial endeavor that transcends my pre-
sent concerns. Still, my account of black boxes has something to offer to the
overall project. My aim, in the pages to follow, is to show that the three-step
recipe for the construction of black boxes, developed in the first half of the
book, sheds light on an issue that lies at the core of any general account of sci-
ence, namely, progress.
Here is an outline of the current chapter. To kick things off, section 9.2
reconstructs the genealogy of the concept that, more than any other, has
sparked controversy by undermining traditional accounts of the advance-
ment of science: meaning incommensurability. Section 9.3 surveys the strategy
adopted by many contemporary philosophers to deal with the puzzle, namely,
proposing and refining referential models of meaning. The remaining sections
bring black-boxing into the equation. Readers well-versed in these philosoph-
ical debates should consider beginning here. Specifically, section 9.4 shows how
2 See, for instance, Lakatos (1970); Shapere (1974); Laudan (1977); and Kitcher (1993).
218 Black Boxes
black boxes can be used to enrich the conception of progress underlying refer-
ential models, while mitigating the perilous consequences of meaning holism.
Section 9.5 borrows some insights from referential accounts of theory-change
to sharpen and develop the three-part recipe presented in Chapter 5. Finally,
section 9.6 wraps up our discussion of scientific progress with a few concluding
remarks.
Before getting down to business, allow me to preempt a potential ob-
jection.3 Some readers may take my call to revisit traditional questions of
progress, incommensurability, sense, and reference as a step in the wrong
direction. These are topics reminiscent of 1970s philosophy of science which,
like most of philosophy back then, took its cue from the philosophy of lan-
guage. Many have come to view the divorce from the issue of language, and
its replacement with an increased attention to empirical practice, as one
of the great achievements of contemporary philosophy of science. This
being so, should we really resurrect these tired old issues of meaning? Even
philosophers of language have finally set them aside, and for good reason.
To be clear, I have no intention of dragging philosophy of science back to
its positivistic roots. Nor do I want to disavow the current naturalistic stance,
as my attention to various case studies will hopefully testify. Nevertheless, the
question of progress is a central constituent of any account of science worth
its salt. And to the best of my knowledge, it has not been solved. It has been
quietly swept under the rug, where it has lain dormant ever since. My goal
is hardly to rehash the good old linguistic turn. It is to bring important phil-
osophical issues back onto the main stage. And if an honest, “naturalized”
discussion of progress requires reviving long lost concepts such as meaning,
reference, and incommensurability—is there a better alternative currently
on the table? —then so be it. Let’s not throw out the baby with the bath water.
If you just can’t stomach it, flip to the following chapter. No hurt feelings!
§9.2. The Roots of Incommensurability
Let me begin by setting things straight. Section 9.1 casually referred to “the”
logical positivist account of science. Strictly speaking, this is an oversimpli-
fication. There admittedly is no uniform, monolithic depiction of science
collectively adopted by all positivists. Schlick, Carnap, Hempel, Neurath,
3 I am grateful to John Bickle for lucidly bringing this reaction to my attention.

and Reichenbach—just to mention a few key figures—had subtly different

conceptions of the scientific enterprise. Furthermore, Kuhn’s proposal was
equally at odds with the views of Karl Popper, who would have vehemently
rejected any association with positivism.4 Nevertheless, setting aside substan-
tial disagreement among the aforementioned authors, the main protagonists
of philosophy of science during the first half of the twentieth century did
share a general outlook. As Hacking (1983) observes, this common model
presupposes the following assumptions. First, a sharp distinction is drawn
between observation and theory and between the context of discovery and
the content of justification. Second, science is largely a cumulative endeavor.
Third, there is a general methodology that unifies all of science. Fourth and
finally, this underlying methodology can be characterized precisely in a
formal and essentially ahistorical fashion. For lack of a better term, I call this
background conception the “positivist” image of science.
This cumulative approach to scientific knowledge is epitomized in Nagel’s
The Structure of Science which, as discussed in Chapter 2, has become a classic
of late positivist philosophy. Nagel tells an optimistic tale of stable structure
and continuity across the epochs. Still, he also realized that, from time to
time, theories are in dire need of a makeover or even a replacement.
This commonplace observation raises a crucial question. When is it ra-
tional to switch from an older theory T1 to a newer theory T2? Nagel, who
absorbed the positivist lesson well, had a simple and effective answer. For
such a switch to be warranted, the new theory T2 must be able to explain
all the phenomena already accounted for by T1 and, similarly, preserve all
the successful predictions of its predecessor. In addition, T2 should exclude
parts of T1 which turn out to be faulty, erroneous, or otherwise misguided.
Alternatively, the new theory must cover a wider range of explanations and
predictions compared to the old one. Ideally, the new theory accomplishes
both goals, in which case, Nagel claims, we can say that T2 “subsumes” T1.
This rough sketch corresponds to the broad picture rejected by Kuhn
and his allies. Specifically, Kuhn famously replaced this standard mono-
lithic model with a two-factor approach, which distinguishes two stages of
science: normal science and revolutionary science. In a nutshell, normal sci-
ence is a puzzle-solving activity largely engaged in relatively minor tinkering
4 This is chiefly because Popper rejected “bottom-up” theories of confirmation, a hallmark of

positivism, which purport to build knowledge from individual observation statements. Popper
espoused a “top-down” approach, which eschews all forms of induction and confirmation in favor of
a hypothetico-deductive falsificationist scientific methodology.
220 Black Boxes
with established theory. Against the assumptions of positivists and their tra-
ditional adversaries, confirmation, verification, and falsification play a rel-
atively marginal role in everyday “normal” scientific practice. For instance,
contrary to Popper’s dictum, mismatches between theory and observation
should not be considered falsifications. Rather, they are routinely treated as
“anomalies,” that is, nagging counterexamples to be captured and explained
away by the reigning paradigm. Nevertheless, despite these substantial
differences, Kuhn’s normal science can be easily reconciled with Nagel’s
model. In particular, normal science may be viewed as a cumulative collec-
tion of statements and concepts within a specific domain. Progress is tanta-
mount to any contribution to this growing body of knowledge. Yet, this is
only one half of Kuhn’s story and not the most exciting one.
Kuhn observed that sometimes anomalies stubbornly resist resolution.
Rather than washing away, they pile up, and a few may come to be viewed as
especially pressing. As researchers try to fix the problem, counterexamples
accumulate. Consequently, the field enters a state of crisis. The typical way
out of this quagmire is the development of a fresh start, equipped with a host
of novel tools and concepts. As the new framework rapidly establishes it-
self and makes progress, superseded questions and ideas are set aside and,
eventually, they are forgotten. When this happens, when an older paradigm
is replaced by a newer one, we have what Kuhn dubs a scientific revolution.5
The new paradigm eventually crystallizes into normal science, adopting a
consensus-forging role, producing fresh anomalies, which eventually trigger
a crisis, followed by another revolution, and so on, in a continuous cycle.
The outcome of the revolution—the new paradigm—typically exhibits dif-
ferent goals and interests compared to its predecessor. The revamped normal
science may ask innovative questions, postulate novel concepts, posit dis-
tinctive laws, and so forth. Still, the occurrence of a revolution, by itself, does
not jeopardize the overall rational or progressive trajectory of science. The
threat to progress and rationality stems from the very nature of paradigm
shifts, which Kuhn famously compares to religious conversions and gestalt
switches in psychology. Members of the new paradigm, he claims, “live in
a different world” from their predecessors. They speak different languages
5 Providing a precise, univocal characterization of “scientific paradigms” is a notoriously thorny en-

deavor. Following Hacking (1983), I focus on two prominent meanings. First, paradigms can be con-
sidered achievements, in the sense of being the ultimate goal of normal science. Second, paradigms
are sets of shared values: collections of common methods, standards, and basic assumptions that
jointly underlie specific lines of inquiry.
that cannot strictly be compared or translated into each other. The only way
to transition from one to the other is via a eureka-style intuition, as opposed
to deliberative reasoning. This is where trouble begins. “Living in a different
world” has substantial implications for progress.
On Kuhn’s picture, sensu stricto, where novel paradigms carry new lan-
guages with them, one might not even be able to convey the ideas of the
replaced theory in the language of the replacing one. Indeed, Kuhn initially
suggested that there is literally no way of specifying a theory-neutral lan-
guage in which to express and compare the two frameworks. This is a striking
departure from the old cumulative model of scientific knowledge. Recall that
Nagel did recognize the need for theory change. Still, despite the occasional
setback or difference in focus or perspective, the new theory always takes the
success of its predecessor under its wings, while eschewing some problems,
failures, and misconceptions. The basic idea is that the two theories can be
rationally compared, and the “better” one is selected. After all, substantial
change in theory is warranted if and only if the new paradigm explains the
known data and predicts new observations more accurately than its prede-
cessor. This, Kuhn claims, is what happens in the context of normal science.
Yet, he continues, when we zoom out and focus on larger time scales, science
does not work that way. After a revolution has swept in, a substantial compo-
nent of the old paradigm is dismissed and forgotten. It eventually becomes
accessible only to historians who, through slow and painstaking work, are
able to reconstruct the discarded Weltanschauung.
This, simply put, is Kuhn’s notorious concept of incommensurability, which
leads to conceptual relativism, the controversial doctrine that the language
used in a field of science changes so radically during a revolution that the
old and the new theories are not mutually inter-translatable.6 Etymologically,
“incommensurability” derives from ancient Greek mathematics, where the
term had a precise meaning, denoting two lengths which have no common
measure.7 Kuhn borrows this idea and applies it, figuratively, to the compar-
ison of scientific paradigms. The notion of incommensurability may be used
to describe different phenomena, with various controversial implications.
6 As Kitcher (1978) remarks, it is interesting that Kuhn and Feyerabend, vigorous contenders of
the relevance of history for the philosophy of science, have also advanced theses which imply that the
task of the historian of science cannot be successfully completed.
7 Two lengths p and q have a common measure if it is possible to lay x of one against exactly y of the
other, thereby measuring p in terms of q, or vice versa. Not all lengths are commensurable, as shock-
ingly discovered by the Pythagoreans when they realized that the diagonal of a square is “incommen-
surable” to the length of the sides.
222 Black Boxes
Following Hacking, I distinguish three distinct concepts of incommensura-

bility, as they pertain to the history and philosophy of science.
First, there is what can be called “topic incommensurability.” As noted,
Nagel followed his fellow positivists in claiming that when a theory is replaced
by a successor, the new theory always does the same job “better” than the old
one. Kuhn and Feyerabend saw that this view is unduly narrow. The para-
digm that is established in the wake of a revolution is frequently quite dif-
ferent from its predecessor. For instance, it may ignore a lot of past successes
and it might re-describe phenomena in different ways. When this happens,
later generations may find previous theories unintelligible, until they play the
role of historians, effectively reinterpreting the old theory. Revolutionary as
it was in the 1960s, this topic-incommensurability has now become widely
accepted. Few contemporary scholars would deny that, while subsumption
captures many prominent episodes in the history of science, it does not cover
all cases.
A second, stronger kind of incommensurability, which is central to
much of Feyerabend’s provocative discussion in Against Method, is what
Hacking dubs “dissociation.” This occurs when the gap between a super-
seded theory T1 and its successor T2 is so significant that, from the per-
spective of T2, one cannot even attach a truth value to many statements of
T1. Hacking illustrates this phenomenon with the example of Paracelsus,
a sixteenth-century alchemist whose theories of matter were so far-
fetched from the perspective of contemporary chemistry that they cannot
be matched with anything we want to state today. Sure, one may express
Paracelsus’s claims in modern English. Yet, we often cannot meaningfully
assert or deny what is being said.
Hacking’s third kind of incommensurability stems from a puzzle con-
cerning the meaning of theoretical terms, that is, terms that refer to un-
observable states (cf. §6.4, Chapter 6). The problem is simple but deep.
Children and adults alike often learn the meaning of mundane concepts
by ostension. To illustrate, one can convey the meaning of “apple” by
displaying a paradigmatic apple. Similarly, one can acquire the word
“bitter” by tasting a substance that falls under said category. But how do we
assign meaning to concepts referring to entities that cannot be observed?
How do we learn about electrons, black holes, molecular chaperones,
protons, and so forth? One obvious suggestion is that we acquire these
concepts by memorizing definitions. Upon further scrutiny, however,
this strategy turns out to be a non-starter. For one, precise definitions are
hard to come by. Furthermore, and even more problematic, the definitions
of many theoretical terms appeal to other, even more technical notions.
For instance, learning the standard definition of “electron” as a negatively
charged subatomic particle presupposes that the subject already grasps
the concept of charge, which is arguably more complex and theory-laden
than “electron” itself. These and similar difficulties suggest a different,
more plausible alternative. We explain theoretical terms by specifying a
theory. The meaning of words like “electron” and “black hole” is provided
by their position within the structure of the entire corresponding theory.
From this holistic standpoint—pioneered by Duhem, endorsed by many
logical positivists, and developed to its fullest extent by Quine8—it follows
that, say, “mass” does not mean the same thing in classical and relativistic
physics, and “planet” is a different concept for Ptolemy and Copernicus.
The reason should now be evident. If the meaning of a theoretical term
is determined by an entire corresponding theory, changing the theory
thereby changes the meaning of all its relevant concepts.
At first blush, these conclusions will strike some readers as inconse-
quential and, perhaps, quite plausible. Did Einstein’s groundbreaking
intuitions not change the very meaning of “mass” by relativizing it to a
frame of reference? Meanings are constantly in flux. All this is fine. Yet,
problems arise as soon as we try to compare theories. If terms like “elec-
tron” and “mass” get their meaning from their place within a network
of statements and laws, then, as noted, when the theory is modified,
the meaning of these terms changes as well. But if the meaning of core
concepts varies from theory to theory, how can we compare the theories
themselves? If, in principle, theories never talk about the same thing, then
we have no common measure to assess them. Hence arise the notorious
problems of incommensurability and theory change, which turn Nagel’s
plausible doctrine of subsumption, as well as the very possibility of “cru-
cial” tie-breaking scientific experiments, into a logical impossibility. In
what follows, it is this third and strongest meaning-incommensurability
that will be the focus of our attention.
The thesis of meaning-incommensurability was met with cries of out-
rage. Some dubbed it untenable because its rests on the fundamentally
incoherent idea of incompatible conceptual schemes. Others rejected it
on the grounds that there is clearly enough similarity of concepts across
8 See Duhem (1954); Carnap (1956a); Hempel (1965, 1966); and Quine (1953).
224 Black Boxes
successive theories to allow for meaningful comparisons.9 Kuhn him-

self was taken aback by the way his work, and the contributions of his
colleagues, produced a crisis of the concept of rationality in the philos-
ophy of science.10
In the wake of the heated debates ensuing the publication of Structure,
Kuhn (1977) clarified his position by stating that he never intended to
deny the customary “virtues” of scientific theories. Theories should be ac-
curate, logically consistent (both “internally,” in their own structure, and
“externally,” when conjoined with other theories), broad in scope and
rich in consequences, simple in structure, and fruitful. Following these
observations, Kuhn’s own “revolution” in the philosophy of science may thus
be conceived along two orthogonal axes. First, these and similar values are
never sufficient to make a decisive choice among competing theories. Other
considerations come into play, factors for which there could, in principle, be
no formal decision-algorithm. Second, Kuhn maintains that proponents of
rival theories are like native speakers of different languages. There are sub-
stantial limits to what can be effectively conveyed. Still, despite these com-
munication breakdowns, scientists immersed in different theories can, albeit
with difficulty, present to each other concrete technical results, available to
those who practice within each theory, and their implications.
In conclusion, pace authors like Feyerabend and Shapere, who would
like to expunge semantic issues from the philosophy of science, the no-
tion of incommensurability raises, at the most basic level, a question
concerning how theoretical concepts acquire their significance. And
addressing this quandary, unfortunately, presupposes at least a rough con-
ception of meaning—a notoriously thorny endeavor. There is, however,
some good news. The past few decades have witnessed the development
of referential models of meaning, which promise to resolve the problem of
incommensurability by allowing radically different theories to talk about
the same entities. Section 9.3 turns to some popular variants of this influ-
ential approach.
9 The first critique was famously explored by Davidson (1974). The second route was developed
primarily by Shapere (1966); Achinstein (1968); and Kordig (1971).
10 Hacking (1983) suggests that Kuhn did not originally intend to address the issue of rationality at
all. Things are different, however, in the case of Feyerabend, a philosopher whose radical ideas often
overlap substantially with Kuhn. Feyerabend is a longtime foe of dogmatic rationality. For him, there
is no canon of rationality, no privileged class of “good reasons,” and no absolute paradigm of science.
There are many rationalities, as opposed to a single one, and the choice among them can never be
fully objective.
§9.3. Referential Models and Incommensurability
Traditional philosophical discussions of language are heavily influenced

by the seminal work of Frege, who distinguished between two components
of meaning: the sense of an expression, which must be grasped by all
communicators, and the reference, the entity denoted by the expression in
question. Now, supplement Frege’s insight with the holistic presupposition—
which came to be accepted by logical positivists but, to be clear, Frege him-
self would have rejected—that the sense of a term is determined by its place
within a network of theoretical statements. It follows that the meaning of the
term always shifts whenever the underlying theory undergoes change. Ecce
meaning incommensurability, as presented and discussed in section 9.2.
Putnam pioneered a response to the problem by enriching Frege’s pic-
ture.11 Meaning, he claims, has not two, but four components. First, each
expression has syntactic markers, which characterize its syntactic structure
(e.g., noun vs. adjective vs. verb). Second, words have semantic markers
that assign the term in question to the appropriate category. Thus “rabbit”
denotes organisms and “iron” picks out a metal. Putnam’s most original con-
tribution is the third constituent of meaning, the stereotype, which can be
characterized roughly as the conventional image or description associated
with a word. For instance, rabbits are typically depicted as furry, timid, fuzzy
mammals. Similarly, water is associated with an image of a tasteless, trans-
parent, thirst-quenching liquid. Stereotypes can be inadequate. Some, even
all, elements of the description may be mistaken, as revealed by the observa-
tion that not all rabbits have long ears and, alas, not all water comes in clear,
tasteless form. Furthermore, conventional depictions are subject to change.
The once-prominent stereotype of wolves qua vicious, blood-thirsty beasts is
fortunately turning into a thing of the past.
If syntax and semantic markers are too general, and stereotypes are subject
to change, is there anything specifically tied to an expression that remains
constant in meaning across contexts? The answer lies in the fourth compo-
nent of meaning: reference. While stereotypes may evolve as we find out more
about a certain thing or kind, the denoted entity—or the extension, the set
of entities falling under the predicate, in the case of properties and natural
11 Frege’s classic distinction between Sinn and Bedeutung comes from his 1892 article. Putnam’s
well-known proposal is detailed in “The Meaning of Meaning” (1975b).
226 Black Boxes
kinds—does not vary. Thus, the fundamental identity for expressions is de-
termined neither by senses nor by stereotypes. It comes from reference.
With this in mind, it becomes clear how it is possible for concepts
embedded in radically different theories to nevertheless talk about the same
things. This happens when a term maintains the same stable reference across
Kuhnian paradigm shifts. Thus, Democritus, Newton, Laplace, Thomson,
Lorentz, Bohr, and Millikan, all of whom have very different theoretical
presuppositions concerning physical particles, can be talking about the same
entities when they use the term “electron.” In short, this referential approach
provides the resources to address the meaning of theoretical terms without
being lured into problems—or pseudo-problems—of incommensurability
and relativism.
Things, unfortunately, are not that simple. Putnam’s sketch works rea-
sonably well for “success stories,” where authors with very different
presuppositions may nonetheless use a term to refer to the same natural
kind: electrons, viruses, gravity, and the like. But how do we apply this model
to concepts, such as acid, which have competing definitions and are very
likely not to be homogeneous natural kinds? Even more problematically,
proponents of theories positing nonexistent entities, such as aether and ca-
loric, seemingly communicate their ideas just as well as researchers with dif-
ferent views about “real” theoretical entities, like viruses and electrons. How
do we account for this observation? How do we explain the substantial con-
tribution of these theories to contemporary science, despite the lack of stable
referential relations across the board? Some dismiss these examples on the
grounds that “the notion of meaning is ill-adapted to philosophy of science.
We should worry about kinds of acids, not kinds of meaning” (Hacking 1983,
p. 85). Others have picked up the tab by revising Putnam’s insights.
An influential refinement has been offered by Kitcher.12 Situating himself
within a philosophical tradition which feels uncomfortable with intensional
entities like Fregean senses, Kitcher urges the benefits of a “referential” ap-
proach to the semantics of scientific terms. As noted, the development of an
extensionalist (referential) account of conceptual change was already well on
its way, at the time Kitcher was writing.13 However, Kitcher argues, referen-
tial change, in and of itself, is neither necessary nor sufficient for conceptual
12 Kitcher develops his views over a series of publications. The following reconstruction brings to-
gether aspects from Kitcher (1978, 1982, and 1993).
13 The idea of cashing out incommensurability in terms of reference was an old adage, proposed
over a decade earlier by Scheffler (1967), and refined by Putnam, among others.
relativism.14 What would truly inhibit inter-paradigm understanding is a

particular kind of referential change, culminating in a mutual inability to
specify the referents of terms used in elucidating the rival position. Thus, at
the heart of relativism is the thesis that scientists working in the same fields
across revolutions are unable to compare their theories. This could happen
when, for some expression-types, we are unable to specify a set of entities
whose referent belongs to that set. In his own words, “the idea of the incom-
parability (‘incommensurability’) of two scientific theories presupposes that
there is no adequate translation of the language used in presenting one of
those theories in the language of the other” (Kitcher 1978, p. 529). What ex-
actly is going on here?
In the years following the publication of Structure, philosophers of science
typically responded to the challenge advanced by Kuhn and Feyerabend—
the issue of incommensurability presented in section 9.2—by presupposing,
more or less explicitly, that the only alternative to conceptual relativism is the
claim that all scientific language can be translated via a context-insensitive
theory of reference. This, Kitcher argues, is a false dichotomy. Neither rela-
tivism nor traditional invariant approaches offer an adequate account of how
scientific terms evolve across paradigms.
Kitcher unfolds his argument by focusing on a detailed case study: the lan-
guage developed by Joseph Priestley, Henry Cavendish, and other advocates
of phlogiston theory. Such theory, first introduced in section 4.4 of Chapter 4,
purported to provide an explanation of various chemical reactions and, in
particular, the process of combustion. The problem, simply stated, is that the
concept of phlogiston was intended to pick out a substance—a “principle”—
which is emitted in all instances of combustion. Yet, since there is no sub-
stance which is emitted in all cases of combustion, the term “phlogiston” fails
to refer. This being the case, given that “phlogiston” is a non-referential con-
cept, then how could Priestley and his colleagues have made the significant
discoveries which legitimize their rightful spot in the Pantheon of scientists?
Let me clarify the issue by restating it in the form of a dilemma. On the one
hand, in rejecting the existence of phlogiston, we are forced to concede that
not only “phlogiston,” but also all other derivative notions that contain “phlo-
giston” as a constituent—such as “phlogisticated air,” and “dephlogisticated
14 A trivial kind of conceptual relativism sans reference change occurs when languages contain
non-overlapping expressions. More interestingly, reference-change may occur without incommen-
surability, if shifts in reference can be specified in the new language.
228 Black Boxes
air”—are non-referential. This entails that statements involving these enti-

ties are false, as the required truth-makers do not exist. On the other hand,
it is hard to deny that when Priestley talks about, say, the positive sensation
and beneficial effects of breathing dephlogisticated air, he is not speaking
falsely at all. Indeed, he seems to be making true statements about a sub-
stance still unknown to him, namely, oxygen. But how is this possible? How
can a statement be true when one of its core constituents fails to refer? How
can terms like “phlogiston” and derivative notions be both non-referential
and explicative?
Note that we cannot resolve the conundrum by appealing directly to
Putnam’s insights. This is because reference here clearly does not remain con-
stant across theories. As far as we know, “oxygen” picks out a natural kind, a
chemical element with atomic number 8. “Phlogiston,” in contrast, has no
real-world counterpart. The solution, Kitcher suggests, is not to settle for an
inferior translation that renders many of Priestley’s statements inexplicable
from our contemporary perspective. Rather, he claims, we should abandon
the search for an overly demanding context-independent theory of reference
in favor of a context-sensitive one. From this perspective, different tokens of,
say, “dephlogisticated air” pick out different entities. Some occurrences refer
to oxygen, some to hydrogen, some fail to refer, etc.
How can tokens of the same expression-type denote different things? The
key to answering this question comes from the philosophy of language. In the
1970s, it became clear that, contrary to the then-received view, it is possible
to use a term referentially without being able to offer any informative de-
scription of the term itself.15 The reason is that usage of words is parasitic on
the presence of experts within a linguistic community. These are specialists
who are able to make fine-grained distinctions, real or apparent, between the
extensions of similar concepts, without necessarily having precise definitions
in mind. This notion of direct reference is fixed via a causal story beginning
with some primal act of baptism.16 Important as it is, this social dimension of
language, Kitcher maintains, should not lead us to disregard the crucial role
of speakers’ intentions in determining the reference of expressions.
15 Major contributions are Kripke (1972); Donnellan (1970, 1974); and Putnam (1973).
16 More precisely, following Kripke and Donnellan, Kitcher assumes that this process of fixing ref-
erence takes the form of a “historical explanation.” The referent of a token is that entity which figures
in the appropriate way in a correct explanation of the production of the token in question. Any such
explanation will consist in a direct or indirect description of a sequence of events that begins with a
primal act of baptism and whose terminal member is the production of the token.
Within scientific communities, Kitcher views linguistic practice as

governed by three general principles. First, a maxim of conformity: refer to
that to which others refer. Second, a maxim of naturalism: refer to natural
kinds. Third, a maxim of clarity: refer to that which you can specify. Under
many circumstances, these principles may come into conflict. To illustrate,
consider a physicist who has discovered what appears to be a new kind of
particle. Should this researcher try to refer to the same entities that have been
theorized by others before? Or should they posit the existence of an alto-
gether new natural kind? What if they are unable to specify the nature of
the newly identified entity with clarity? As a result, scientists have to choose,
more or less consciously or deliberately, among them. And since on dif-
ferent occasions even the same author may opt for different choices, different
tokens of the same expression-type may refer differently.
How do we determine the reference of each token of an expression, given
this context-dependency? The correct hypothesis about the reference of
a speaker’s token, Kitcher suggests, is the hypothesis which best explains
why the speaker said what they did. Positing a principle of humanity, which
enjoins us to impute to the author a pattern of relations among psychological
states and the world that is as similar to ours as possible, the best explanation
will tell the correct story about the speaker’s intentions in uttering a term.17
This involves relating those intentions to the external circumstances of the
speech act and the speaker’s verbal behavior, with the understanding that, on
different occasions, even the same author can be guided by different maxims.
Good. But how does all of this work in practice?
Scientific expressions are associated with a complex conceptual appa-
ratus that Kitcher calls their reference potential. The reference potential of
an expression-type is a compendium of the ways in which the denotation of
term-tokens is fixed for members of a community. More precisely, the ref-
erence potential of a term, relative to a speaker, is constituted by the set of
events which the speaker is disposed to admit as initiating events of tokens of
that term. In turn, community members are defined as individuals who are
disposed to accept the same initiating events for tokens of a term in virtue of
sharing a set of dispositions to refer in various ways.
Why do scientific terms frequently have heterogeneous reference
potentials, that is, reference potentials which include a broad and diverse
17 This “principle of humanity,” which Kitcher borrows from Richard Grandy (1973), is also known
in philosophy as the “principle of charity.”
230 Black Boxes
set of initiating events? The reason is that scientists who engage in different
projects frequently find it useful to initiate their tokens by different events.
For instance, these tokens may correspond to the different ways in which
a chemical substance or reaction can be produced. When this occurs, as it
frequently does, we may reasonably conclude that the term in question is
“theory laden.”
What exactly is Kitcher alluding to? Setting technicalities aside, a simple
example should help drive the point home. Consider the concept of gene, in
the context of Mendelian genetics. When classical geneticists talked about
“genes,” what exactly were they referring to? The obvious answer is that genes
corresponded to units of function. The effects of the genotype of an organism
on their phenotype, from this standpoint, could be traced back to the effect
of a gene or set of genes. This is true. However, it can hardly be the entire
story. Within the theory of classical Mendelian genetics, genes also played at
least two other theoretical roles.18 Genes were also taken to be units of recom-
bination, in the sense that changes in linkage relationships either separated
genes that were previously linked or linked genes which previously segre-
gated independently. Finally, genes were characterized as the units of muta-
tion: changes in genes give rise to new alleles, variants of that same gene. In
short, as we shall discuss in greater detail in the following section, scientists
within and across eras and fields can use the same term, “gene,” to pick out
very different entities. This can be captured effectively by claiming that “gene”
is a theory-laden term with a multifaceted reference potential. What makes
classical geneticists a “linguistic community” is their shared disposition to
use the term “gene” to refer to the same entities or events.
The thesis that theory-laden scientific expressions have a heterogeneous
reference potential—that is, these terms may refer to different entities in dif-
ferent contexts—is the conceptual core of Kitcher’s extensionalist solution
to both incommensurability and scientific progress. I take up both issues,
in turn.
Begin by focusing on the former issue: incommensurability. The heter-
ogeneous reference potential of an expression depends, at least in part, on
the theoretical presuppositions which pertain to the particular paradigm
to which the expression belongs. Hence, it will often occur that the refer-
ence potential of a term belonging to the language of a theory cannot be
matched by any expression in its post-revolutionary successors, when these
18 For an insightful discussion of these theoretical roles, see Griffiths and Stotz (2013).
subsequent theories do not share these same presuppositions. If by “incom-

mensurability” Kuhn and Feyerabend simply mean a mismatch of reference
potentials across theories, then there are indeed cases of conceptual rela-
tivism in the history of science. Priestley’s phlogiston is a prime candidate for
an expression whose reference potential lacks an analog in Lavoisier’s chem-
istry. However, this type of relativism does not involve any inability to specify
the referents of tokens across paradigm shifts. In short, reference potential
allows for radical conceptual revision without conceptual discontinuity.
Theory-ladenness does not thwart the possibility of successful communica-
tion. In fact, it is an integral component of it.
Moving on to the second point— progress— reference potential also
lies at the heart of Kitcher’s account of the advancement of science, which
revolves around two types of cognitive progress: conceptual progress and
explanatory progress.19 In conceiving of science as “progressive,” we envi-
sion it as a sequence of consensus practices that get better and better with
time. Improvement need not be constant, but there should be a general up-
ward, positive trend. The challenge, of course, is to clarify what “better and
better” and “upward” mean in this context, given that the mere accumula-
tion of facts is off the table. Conceptual progress can now be understood as
progressively adjusting the boundaries of our categories and providing more
adequate specifications of referents, in a way that makes them conform to
natural kinds. These kinds of shifts can be captured as changes in the ref-
erence potential of key theoretical terms. Explanatory progress, in contrast,
consists in improving our understanding of the structure of nature, an ac-
count embodied in the schemata of our practices. Improvement involves
either matching our schemata to the mind-independent ordering of phe-
nomena or producing schemata that are better able to meet some criterion
of organization, such as greater unification. Progress, in short, in not merely
an increment in knowledge. It also encompasses better, more accurate, more
perspicuous concepts, better equipped to describe and explain nature.
In sum, Kitcher’s discussion of reference potential provides a powerful
tool to analyze, address, and explain away meaning incommensurability, the
core of Kuhn’s and Feyerabend’s critique of traditional philosophy of science,
19 “Cognitive” progress is understood in contrast to “practical” progress, which involves techno-

logical advancements and other means of manipulating nature. While cognitive progress may ini-
tially appear to be more ethereal and elusive, Kitcher (1993) argues that it can be approached far more
easily because of the relative narrowness of the set of human impersonal epistemic goals compared to
the constantly in-flux practical goals.
232 Black Boxes
and to capture the notion of progress. Incommensurability boils down to

mismatch in reference potential. This form of conceptual relativism is wide-
spread in the history of science. But it does not undermine the possibility of
viewing the relation between theories as “progressive.”
With this in mind, it is time to bring black boxes back on the main stage. In
the remainder of the chapter, I articulate and defend two related claims. First,
the black-boxing strategy, detailed in Chapter 5, brings to the table a signifi-
cant array of resources that can be used to supplement and refine referential
accounts of conceptual change. In particular, it suggests a richer and more
plausible model of progress—a long-standing and significant philosophical
issue that has been unduly neglected over the last quarter of a century. This
thesis will developed in section 9.4. Second, reference potential, explanatory
extension, and other notions discussed in the first part of this chapter can
be used to further clarify the nature and structure of black boxes. This latter
tenet will be the subject of section 9.5.
§9.4. Black Boxes and Reference Potential, Part I
The previous section briefly covered some influential approaches to the time-
worn issue of scientific progress. Referential models have two substantial
strengths. First, they allow talk about theoretical change without positing
senses or other intensional entities that many authors find ontologically sus-
picious. Second, they pave a way for heeding Kuhn’s and Feyerabend’s call to
take seriously the history of science, while eschewing the more radical and
problematic implication of their view: meaning incommensurability.
The strategy outlined in section 9.3 is not devoid of controversial
implications. Kuhn himself was sympathetic to Kitcher’s proposal that the
language of modern chemistry can successfully identify the referents of the
key expressions of phlogiston theory. Yet, Kuhn did not fully accept Kitcher’s
characterization of reference-determination as a bona fide “translation” and
the related suggestion to bring talk about incommensurability to a close.
We may paraphrase Kuhn’s qualms along the following lines. Referential
agreement may well be a necessary condition for comparing theories and
ideas across paradigms. But is it sufficient? Kuhn answers in the negative.
Communication requires more than a shared interpretation based on exten-
sional semantics. It presupposes a true “translation,” that is, agreement on
what is said about the referents. Thus, formulations and resolutions of in-
commensurability must go beyond mismatches in reference potential.20
In fairness, Kitcher does not flatly identify translation with referential
agreement. As we shall shortly see, he acknowledges the difficulty and offers
a possible way out. Still, Kuhn’s general point deserves to be emphasized. His
remarks can be turned into an adequacy condition, posing a dilemma-style
argument for any general account of conceptual change. Purely referential
models dodge incommensurability by showing how speakers with radically
different theories can nevertheless talk about the same stuff. But reducing
translation to referential agreement, Kuhn notes, is problematic. Successful
communication requires convergence on what is said about the referents.
This requires positing something akin to Fregean senses. But, first, senses are
ontologically spooky. Second, and more important, senses are what trigger
the problem of incommensurability in the first place, when coupled with the
modest holistic assumption that the meaning of theoretical terms is deter-
mined by their position within the structure of the entire theory.
In short, here is the dilemma. On the one hand, reference alone is insuf-
ficient to characterize inter-paradigm communication and, hence, the ad-
vancement of science. On the other hand, richer, intensional accounts solve
the problem of communication but lead us back to forms of incommensura-
bility. Can we find a conception of progress that is substantial enough to do
the trick, but sufficiently slender to avoid unpalatable consequences?
Section 9.6 will take a look at Kuhn’s own attempt at taking a stab at the
problem—an admittedly metaphorical characterization of the intensional
component of translation in terms of culture and the structure of language.
Meanwhile, let us focus on Kitcher’s proposal that, while not entirely explicit
on this score, offers a promising starting point for a solution.
At various points, Kitcher suggests that his notion of reference potential
captures something of Frege’s non-referential dimension of meaning and
the quasi-holistic dictum that concept formation and theory formation go
hand in hand.21 Allow me to elaborate. The notion of sense posited by Frege
is supposed to play (at least) two fundamental roles. First, the sense is what
20 My reading of Kuhn’s (1982) response to Kitcher is inspired by Carey (1991, 2009). “The problem
with Kitcher’s argument is that it identifies communication with agreement on the referents of terms.
But communication requires more than agreement on referents; it requires agreement on what is said
about the referents. The problem of incommensurability goes beyond mismatch of referential poten-
tial” (Carey 1991, p. 462).
21 Contrary to common wisdom, modest holism—typically associated with Quine and Kuhn—
was accepted and endorsed by mature logical empiricists such as Hempel (1966).
234 Black Boxes
is “grasped” by anyone who understands an expression; it is the meaning of

the term(s). Second, sense encompasses the way in which the reference is
expressed; it is the “mode of presentation” of the term(s). Reference poten-
tial, Kitcher argues, is akin to sense in this second connotation: it constitutes
the mode of presentation of a theoretical term. In addition, since reference
potential partly depends on the presuppositions of a paradigm, it becomes
clear why scientific terms must holistically absorb theoretical hypotheses. In
short, reference potential is the key to understanding conceptual change.
But what allows reference potential to fulfill this role? How does reference
potential specify the mode of presentation of a term? Kitcher’s answer ap-
peals to the intentions of speakers and their dispositions to behave:
Conceptual change in science is to be understood as change in reference

potential. The reference potential of a term for a speaker is the set of events
which a speaker is disposed to admit as initiating events for tokens of that
term. A linguistic community, with respect to a term, is a set of individ-
uals disposed to admit the same initiating events for tokens of the term. An
event is the initiating event for a token if the hypothesis that the speaker
referred to the entity singled out in that event provides the best explanation
for her saying what she did. Explanations are judged by their ability to pro-
vide a picture of the speaker’s intentions which fits with her environment
and history and with the general constraint of the Principle of Humanity.
Three kinds of intentions are prominent: the intention to conform to the
usage of others, the intention to conform to natural kinds, and the intention
to refer to what can be specified. (Kitcher 1982, p. 347)
Plausible as this sounds, intentions, dispositions, and other psychological

states are elusive notions. Identifying, characterizing, and attributing these
features across agents is a notoriously thorny task. Can we make this any
more precise? Sure we can! The remainder of this section shows how black
boxes capture reference potentials, explaining how they can play the role of
sense qua mode of presentation in conceptual change and progress. For now,
I focus on phlogiston. Different examples are developed in section 9.5.
Phlogiston has long been a poster child for conceptual change.22 As we saw
in section 4.4 (Chapter 4) and section 9.3, phlogiston theorists purported to
account for a number of chemical reactions and, in particular, the process of
22 My condensed presentation is largely based on Kitcher’s (1978) insightful overview.

combustion. The core assumption was that all flammable substances are rich
in a “principle”—a substance, namely, phlogiston—which is imparted to the
air during combustion.
The challenge confronting historians and philosophers of science is to ex-
plain how it is that many of Priestley’s insights constitute accurate and sig-
nificant scientific discoveries, given the non-referential nature of its core
concepts: phlogiston and derivative notions. This calls for elucidation.
How do we explain the transformations involving, say, the burning of
fuel and the heating of metal? Priestley and his colleagues had an interesting
story to tell. When we burn a log en plein air, the phlogiston contained in the
cellulose is released in the atmosphere, leaving ash behind. Similarly, when
an iron bar is heated, the combination of heat, metal, and air causes iron to
release phlogiston into the surroundings, leaving the “calx” of the metal as a
residue. Even more impressively, the theory also provided an account for the
reverse of these reactions. Heating the red calx of mercury in an air-rich con-
tainer results in a combination of mercury and a different kind of air, which
Priestley called “dephlogisticated air.” How does this work? Phlogiston theory
provided an ingenious explanation. Heating the red calx of mercury causes
the calx to take up the phlogiston contained in the atmosphere, turning
“normal” air into “dephlogisticated air.” Priestley’s explanations were backed
up by successful predictions and retrodictions. For instance, because phlo-
giston is released during the process of combustion, the residue (ash, calx, or
the like) should weigh less than the original substance: wood, metal, etc. For
similar reasons, the surrounding air should be altered by the reaction. Both
predictions are actually borne out and were experimentally verified.
Regardless of these partial— albeit remarkable— successes, phlogiston
theory did not withstand scrutiny. Eventually, the new quantitative chemistry
of Lavoisier provided better explanations of these and related phenomena.
We now have conclusive evidence against the existence of any principle or
substance being emitted in all reactions of combustion. Oversimplifying a
bit, from the modern perspective, heating an iron bar produces metal oxide
and releases into the atmosphere air that is poor in oxygen. Similarly, heating
mercury oxide, what Priestley calls the “red calx of mercury,” produces mer-
cury and releases pure oxygen into the surrounding air. In short, phlogiston
theory was eventually superseded and replaced by atomic chemistry. But the
question remains: how can we explain the partial successes of phlogiston
theory, given the non-referential nature of its core concepts?
236 Black Boxes
Kitcher wisely abandons the search for an overly demanding context-

independent theory of reference spelling out the one reference of “phlo-
giston” and related notions. He favors a context-sensitive approach, which
allows the same expression to vary in reference on various occasions. But
when does “phlogiston” refer to oxygen, or to a different natural kind? When
does it refer to a nonexistent principle? Sure, intentions have an important
role to play. But how do intentions change the functional role of a term within
a causal explanation? Let me sketch an answer to this question.
The first key observation is that “phlogiston” acts as a placeholder in a
causal explanation represented in a model. To see this, consider the situa-
tion from the standpoint of our three-step strategy. First, the inquiries sub-
sumed under phlogiston theory are far from unqualified. The experimental
work of Priestley and Cavendish is embedded in a rich thicket of concepts,
presuppositions, expectations, and goals. It is only relative to this framework
that we can determine and assess the proper explananda, explanantia, and
standards for explanation of phlogiston theory. In a word, these requests for
explanation are heavily framed. Second, relative to this framework, we have
a set of well-formed causal hypotheses that identify the crucial difference-
making factors. Third, these causal explanations are represented in a model,
which guides the appropriate amount of abstraction and idealization for the
inquiry at hand. It also determines which background assumptions may be
left implicit and which must be stated explicitly. In short, for all intents and
purposes, phlogiston can be constructed as a black box.
It follows from the previous considerations that the term “phlogiston”
may figure in two different ways in the context of an explanation. On the one
hand, it may designate a mechanism that produces a pattern of behaviors.
As such, it is a difference-maker: it is a placeholder that stands in for what-
ever entity, substance, or process plays the required causal role in sustaining
the outcome. Thus, when Priestley maintains that heating the red calx of
mercury produces “dephlogisticated air,” there must be something that ac-
counts for the observable transformation at hand. Here, “phlogiston” refers
to the underlying reaction and, consequently, “dephlogisticated air” corres-
ponds to the substance produced, namely, oxygen. On the other hand, “phlo-
giston” can also be used as an explanandum, as a description of process to
be accounted for, a chemical reaction in need of explanation. For instance,
one may ask: how is phlogiston extracted from the log and released into the
air? Here, again, “phlogiston” is a placeholder. Yet, it does not stand in for the
mechanism producing the reaction. It takes the place of a pattern of behaviors

that calls for an explanation. In this latter case, phlogiston is a frame.
Distinguishing between frames and difference-makers helps reveal the
complex reference potential of theory-laden terms. When phlogiston is
used as a difference maker, it typically picks out whatever produces the be-
havior identified by the corresponding frame. It may stand in for oxygen,
hydrogen, or other substances unknown to Priestley and other phlogiston
theorists. It may also stand in for nothing at all, when the frame specifies
an explanandum that is not produced by anything. For instance, one may
ask: in what ways does phlogiston interact with the cellulose in the wooden
log, such that it gets released into the surrounding air? Here phlogiston is a
non-referential term. It picks out nothing at all because there is nothing at all
to be explained. In contrast, when used as a frame, “phlogiston” may stand in
for the combustion-events or for other chemical reactions in need of expla-
nation. Some frames provide well-formed explananda, from a contemporary
perspective. What principle or substance is emitted in all cases of combus-
tion? Answer: none. Other explananda can be dismissed as ill-posed, from
a modern standpoint. What is the atomic number of phlogiston? How does
phlogiston interact with hydrogen? Answer: these questions have no clear
place in our contemporary chemical theories.
In sum, Kitcher is correct that conceptual change should be understood
as change in reference potential, that reference potential captures Frege’s
notion of sense qua mode of presentation, and that speakers’ intentions, as
well as other psychological notions, play a significant part in the determina-
tion of reference. But reference potential itself can be further broken down
in terms of its functional role in the context of an explanation. Identifying
theoretical terms with black boxes distinguishes their act as frame and
difference-makers. The three steps of the black-boxing strategy—framing,
difference-making, and representation—help us see when reference re-
mains constant across paradigms (“planet”), when reference potential needs
to be reassessed in the light of future evidence (“atom”), and when the term
is better dropped altogether (“phlogiston”). In short, “Kuhnian” concep-
tual change depends on how a black box is packed or unpacked within its
framework.
This section showed that the black-boxing strategy provides the resources
to elaborate the notion of reference potential, refining our understanding
of non-cumulative scientific progress. Intentions are elusive notions. Black
boxes help make them more perspicuous. At the same time, the notion of
238 Black Boxes
reference potential sheds light on black boxes. The following section applies
the conceptual resources inherited from our discussion of conceptual change
to re-examine the role of black boxes in the success stories of Darwin and
Mendel, the mistakes of behaviorism, and tales like neoclassical economics,
where the final verdict is yet to be uttered.
§9.5. Black Boxes and Reference Potential, Part II
Chapter 3 presented four episodes in the history of science: Darwin and

Mendel’s momentous contributions to biology, Skinner’s behavioristic psy-
chology, and Friedman’s “as if ” economic models. Chapter 6 revised these
case studies in the light of our three-step black-boxing recipe. I concluded
that the success and limitations of black boxes can be explained in terms of
framing, difference-making, and representation. Our excursus into theory
change, conceptual refinement, and scientific progress provides the resources
to develop our analysis by addressing an important question: how should we
understand the relation between black boxes across paradigm shifts?
§9.5.1. Gemmules, Variation, and All of

That Darwinian Jazz
Let us begin with evolutionary biology. The relation between Darwin’s own
approach and contemporary evolutionary theory is not ordinarily viewed as
one of “incommensurability.” Indeed, whether Kuhn’s account of scientific
revolutions applies to biology at all can be, and has been, questioned.23 Still,
as we shall presently see, the unfolding of the theory of evolution since its in-
ception raises puzzles similar to the ones discussed in previous sections.
Consider Darwin’s own views of inheritance, as presented in Chapter 27
of Variation of Plants and Animals under Domestication (cf. §3.2). According
to his “provisional hypothesis of pangenesis,” both the transmission of
heritable qualities and the unfolding of ontogeny are caused by invisible
particles. These microscopic entities, called “gemmules,” are thrown off by
cells and, when supplied with sufficient nutrients, they can multiply by self-
division. Just a few years after the publication of this conjecture, Weismann
23 For a clear and insightful discussion of this issue, see Godfrey-Smith (2003).
conclusively showed that gemmules do not exist. In this respect, they re-
semble phlogiston. Just as phlogiston is the heart of Priestley’s theory of com-
bustion, gemmules play a central role in Darwin’s views on inheritance. As we
saw, Priestley’s theory raises a puzzle: how can a non-referential concept—
namely, phlogiston—contribute to scientific discoveries? The same question
can be asked with respect to Darwin’s pangenesis, with the additional com-
plication that, contrary to phlogiston theory, Darwinian evolution has not
been superseded. What is going on here?
One option is to flatly ignore Darwin’s claims in Variation and focus on
the earlier, more influential framework in Origin, where the English sci-
entist remained agnostic on the nature of the mechanisms of inheritance.
Rather than solving the conundrum, this whiggish historical reconstruction
merely sweeps the dust under the rug. After all, Darwin did make this mis-
taken claim about the nature of inheritance. And his errors did not affect
the explanatory success of his theory. The central question is: why did these
mistakes leave the overall success of the theory unscathed?
A much more promising strategy borrows Kitcher’s context-sensitive no-
tion of reference potential. From this standpoint, we can recognize how the
reference of various tokens of “gemmules” varies with the circumstances.
Thus, for example, when Darwin speculates that “the gemmules in their dor-
mant state have a mutual affinity for each other, leading to their aggregation
either into buds or into the sexual elements” (Variation, p. 374), it seems rea-
sonable to interpret him as asserting a mistaken claim about nonexistent
entities, namely, gemmules. Now, contrast this with Darwin’s assertion that
gemmules “are supposed to be transmitted from the parents to the offspring,
and are generally developed in the generation which immediately succeeds,
but are often transmitted in a dormant state during many generations and are
then developed” (Variation, p. 374). By applying the principle of humanity,
we may surmise that, here, “gemmule” refers to (what are now known as)
genes and other cytological gears, making Darwin’s statement true, or ap-
proximately so. Following Kitcher, we conclude that this referential hetero-
geneity makes the concept of gemmule “theory laden.”
The black- boxing strategy extends and develops this point, fur-
ther explaining why and how reference potential works the way it does.
Specifically, when does “gemmule” refer to genes? When does it fail to refer?
The answer to this question should now be evident. Darwin’s notion of gem-
mule is part of a much broader theoretical framework. His explanandum is
not a generic: how are traits inherited? As we saw in section 6.2 of Chapter 6,
240 Black Boxes
his objective is a particular account of inheritance, given certain known facts

about physiology, given that some of these traits must make a difference to fit-
ness, given observed distributions within and across species, etc. Against this
backdrop, it becomes clear how, in some cases, the entities responsible for
these patterns are explananda. What is the structure of the mechanisms re-
sponsible for the transmission of trait t across generations? In Origin, Darwin
candidly admits that he has no answer to offer. In Variation, he sketches a
speculative proposal, which turned out to miss the mark. In neither case does
it seem plausible to claim that Darwin has genes in mind. In other cases, how-
ever, gemmules are used as explanantia, as when Darwin appeals to these in-
visible particles to explain the intra-generational transmission of traits. Here,
by applying the principle of humanity, we can take “gemmule” to stand in for
whatever mechanism(s) fulfill the appropriate causal role: genes and other
cytological gears. In the first situation, gemmule is a frame. In the second, it
is a difference-maker. In both cases, it is a placeholder in a causal explanation
represented in a model. Either way, it is a black box.
These considerations are perfectly general and can be applied beyond
gemmules. Evolution by natural selection is a balanced blend of various
ingredients: competition, variation, fitness, heritability, and much else. All
of these components can be broken down further into more fundamental
constituents. From the privileged vantage point of contemporary evolu-
tionism, Darwin’s attempts to do so obtained mixed results. At times, he was
remarkably successful, as when he distinguished between variations that
affect fitness and variations that do not. Other endeavors were more prob-
lematic, as we have seen with his attempts to unravel inheritance. Darwin
at first lacked a solution, and his subsequent proposal missed the mark.
Nevertheless, by viewing Darwin’s constructs as black boxes, as frames and
difference-makers represented in models, we can see how his theory is a mix
of true claims about genes and cytological gears, retained and developed
in subsequent elaborations, and false claims about pangenesis, soon to be
discarded.
§9.5.2. Genes: Mendelian, Molecular, or Manifold?
Genes provide another interesting case study for conceptual change. In the
course of the twentieth century, the reference potential of the term “gene”
has been significantly altered in response to experimental and theoretical
innovations. This raises interesting puzzles. On the one hand, geneticists since
Mendel have always been talking about the same things, namely, chromo-
somal segments. On the other hand, it is clear that, over decades of research,
tokens of “gene” refer to distinct entities. On various occasions, different
chromosomal segments may be picked out. The question becomes: what are
the principles that determine the specific referent of each token?
In his detailed analysis, Kitcher (1982) distinguishes two biological
characterizations of “genes.” The first strategy, prominent in Mendelian genetics,
identifies genes by focusing on their function in producing phenotypic effects.
In its early usage among Morgan’s Drosophila group, “gene” or “factor” referred
to a set of chromosomal segments, each of which plays a distal causal role in the
determination of phenotypic traits. Because of ambiguity in the specification of
these functional roles, the concept gene rapidly acquired a heterogeneous ref-
erence potential, where different tokens could pick out different segments, in
a hypothesis-relative fashion. This classical gene concept was molded into its
definitive form in the 1950s, when Benzer introduced “cistrons” into the pic-
ture. The second approach, common in molecular biology, identifies genes by
focusing on their proximate action. On this view—first articulated in the 1930s
in the context of Beadle’s “one gene–one enzyme hypothesis”—chromosomal
segments are pinpointed according to their functional roles at the earlier stages
of the process, as opposed to using relatively indirect mutation and recombina-
tion tests.
These well-known considerations, Kitcher argues, show the existence of
many concepts of gene, determined by alternative decisions at the pheno-
typic level. These concepts are not in competition. Different ways of dividing
chromosomes—on a spectrum ranging from codons coding for single amino
acids to lengthy DNA sequences coding for multi-polypeptides—will be best
suited to serve one’s interests depending on the particular research project at
hand. From this standpoint, asking which criterion of segmentation corres-
ponds to the concept of gene is an ill-posed question. As anticipated in section
9.3, the term is theory-laden and its reference varies across contexts.
The black- boxing strategy preserves the spirit of Kitcher’s historico-
philosophical reconstruction. In addition, it further explains how and why the
notion of reference potential captures Frege’s sense qua mode of presentation
without lapsing into the quagmire of radical incommensurability.
Alternative gene concepts underlie different theoretical and experimental
purposes. Classical genetics and molecular biology are not addressing the
242 Black Boxes
same questions.24 This becomes crystal-clear at the first framing stage of the
black-boxing recipe. Classical geneticists were perfectly aware that genes are
not sufficient causal bases for complex phenotypic traits such as eye color or
wing shape. From a Mendelian perspective, genes are supposed to account
for variation among members of a population. Molecular genetics, in con-
trast, presupposes an altogether different framework, which aims at unrav-
eling the mechanisms underlying ontogeny. With this in mind, it is hardly
surprising that genes do not play the same difference-making role in these
causal explanations. Classical genetic models represent genes as difference-
makers for phenotypic traits. Molecular biological models, in contrast, rep-
resent genes as difference-makers for the production of polypeptides. Hence,
these two accounts should not be viewed as competing. Classical and molec-
ular genetics should be assessed independently, on their own ground.
At this point, some attentive readers may have noted, we face a problem
of a different sort. If classical and molecular biology presuppose different
concepts of gene, then are they really talking about the same kind of stuff?
Following Putnam’s insight, developed in section 9.3, it may be tempting to
note that both theories refer to the same sets of entities, namely, genes. But
is this a plausible response? As Kitcher rightly notes, asking which concept
of segmentation corresponds to “the” concept of gene is an ill-posed ques-
tion. But, then, what does it mean to say that both theories are talking about
genes? What kind of genes? The issue, simply put, is that we are now led back
into the swamp of meaning incommensurability. If the meaning of theoret-
ical terms is theory-dependent, the meaning will change with the theory.
But this seems simply preposterous. As many authors have noted, molecular
biology extends and deepens the explanations of classical genetics. But this
presupposes that there is a common measure to assess them. What is it? Once
again, black boxes to the rescue.
Both classical genetics and molecular biology treat genes as black boxes.
Importantly, they are not the same black box. Morgan and colleagues frame
and represent genes as the units of mutation, recombination, and pheno-
typic function. Within the framework pioneered by Beadle and later refined
by Watson, Crick, and subsequent molecular biologists, genes are framed
and represented as functional units responsible for transcribing protein. At
the same time, at broader levels of description, viewed as the mechanisms
24 This point, presented cogently albeit in a different context, by Waters (1990, 2007), will be devel-
oped, in greater detail, in Chapter 10.
underlying inheritance and variation, all these units can be subsumed under
a common description. It is these broader frames and difference-makers
that constitute the trait d’union between theories that make very different
assumptions. What is the relation between these levels of description? This is
a question that will be addressed in Chapter 10. And, once again, the notion
of incommensurability will play a central role. Before getting there, however,
we still have a few more examples to cover.
§9.5.3. Where Did Behaviorism Go Wrong?
Black boxes capture conceptual progress in Darwin and Mendel. Can analo-
gous considerations shed light onto the strength and weaknesses of sophis-
ticated radical behaviorism? Skinner is frequently faulted for his decision
to black-box mental states. As we saw in section 6.4 of Chapter 6, this is a
mistake. Skinner did deliberately introduce abstractions and idealizations
in his psychological models. His admittedly oversimplified characteriza-
tion of stimuli and behavior is a good illustration. But this, in and of itself, is
no more problematic than, say, Darwin’s oft casual appeals to traits, fitness,
and the mechanisms of inheritance. Furthermore, unlike Darwin, Skinner’s
explanations do not involve any systematic introduction of nonexistent enti-
ties or non-referring terms. This basic parallel raises a third problematic ob-
servation. The principle of humanity permits a charitable reinterpretation of
some tokens of “phlogiston” and “gemmules” to refer to oxygen and genes,
respectively. Could one not rephrase Skinner’s behavioral states so as to in-
clude psychological and neural processes, as we understand them today? The
problem is that now it seems hard to celebrate Darwinism as a success story
while discarding behaviorism as an outdated theory of mind. But this is bla-
tantly absurd! Where did our analysis go wrong?
The appropriate reaction is not to throw in the towel and settle for an “an-
ything goes” relativism. Skinner was, indeed, guilty of mistakes that did not
undermine Darwin’s and Mendel’s approaches. But, in order to pinpoint
these shortcomings, we need to move beyond referential models of meaning,
even when reference is—rightly—understood in a context-sensitive fashion.
The lack of progress that eventually turned radical behaviorism into a re-
gressive research program involves not what the theory talks about. All its
key terms are perfectly referential. The problem pertains to the structure of
Skinner’s causal explanations of human conduct. Allow me to elaborate.
244 Black Boxes
Sophisticated behaviorism can be portrayed as an attempt to account for

the conduct of intelligent beings without appealing to irreducible mental
phenomena. Recall from section 6.4 the typical causal explanation in
psychology—E → M → B—constituted of environmental events (E), mental
events (M), and behavioral events (B). Skinner’s theoretician’s dilemma pur-
ported to show that the intermediate link M could be eliminated without
any substantial loss of explanatory power. This involves showing that mental
states are dispensable qua explanantia and qua explananda, since, in our
chain, they are explained by the environment and they explain behavior.
We can now state clearly why both claims are fundamentally problematic.
Skinner is correct that environmental stimuli play a decisive role in the pro-
duction of mentality. The cool breeze is partly responsible for my desire to
wear a sweater. His mistake was overlooking the irreplaceable role of log-
ical, cognitive, and linguistic competence, which are required to frame and
causally explain action and other forms of human conduct. All the right
ingredients are in place. Yet, the explanandum is poorly framed and the
choice of difference-makers is inadequate. This shows that scientific progress
goes well beyond successful reference. The causal representation must also
be adequate for the true advancement of science.
In sum, Skinner’s abstraction and idealization in psychological ex-
planation was perfectly acceptable; indeed, it was fruitful. Like the main
ingredients of Darwin’s evolutionary theory and Mendelian genetics, mental
and behavioral states are black boxes. Still, these placeholders were not ap-
propriately constructed. The box of mentality lacks cognitive structure, and
the box of behavior ignores the framing and difference-making role of psy-
chological states. This is where behaviorism went wrong. Black-boxing is
clearly ubiquitous in science. But not all black boxes are made equal.
§9.5.4. Two Concepts of Utility
Finally, consider the black box of neoclassical economics, which provides

the most interesting and controversial example because of its open status.
As discussed in section 6.5 of Chapter 6, Friedman and Savage offer a uni-
fied explanation of economic behavior by positing a single algorithm for
the maximization of utility. According to their influential proposal, the
difference-maker underlying all economic choices—including both risky
and riskless ones—is the agent’s assignment of utility to alternatives. Fans
of neoclassical economic theory praise the simplicity, elegance, and math-

ematical rigor of this simple model. Detractors criticize the lack of re-
alism. Rather than positing a highly idealized notion of utility and abstract
conditions for rationality, psycho-neural economists contend, we are better
off by looking at the mechanisms of behavior that are actually implemented
in the human brain.
If we look at the situation from the standpoint of theory change, the re-
lation between neoclassical and psycho-neural economic approaches might
appear like a textbook example of incommensurability. Both stances assign a
central role to utility. But is it the same construct? The notion of utility intro-
duced in neoclassical economics is an abstract mathematical value with no
physical counterpart. Following the groundbreaking insights of Pareto at the
dawn of the twentieth century, neoclassical economists view utility as an or-
dinal ranking revealed by preference, with no cardinal value. Psycho-neural
economists, in contrast, identify utility with an actual neural value, instanti-
ated in the brain, and measurable, at least in principle.25
These two theoretical constructs could hardly be more different. This raises
a quandary. As we saw in our discussion of genetics, asking which chromo-
somal segments correspond to “genes” is an ill-posed question. Wondering
about which definition of utility is the “correct” one fares no better. But then,
what are we to make of the heated debates in contemporary economics? The
reference of “utility” is clearly different, but the alternative theories nonethe-
less compete. Once again, this goes to show that commensurability, commu-
nication, and progress transcend shared denotation.
Reference potential is a first step toward a solution. The short answer is
that, depending on the context, tokens of the term “utility” may refer to var-
ious concepts across subfields of economics. “Utility” may pick out a math-
ematical construct, a neural mechanism, a set of beliefs and desires, or
something altogether distinct. This effectively captures how different theo-
ries posit a different conception of utility. But this is only half of the story.
The other side involves the shared measure, the common denominator that
makes these theories competitors. If they are talking different notions of
utility, in what sense are they disagreeing? Black-boxing provides a simple
and effective solution.
The common denominator between our two economic approaches is not
provided by shared reference—as just noted, different uses of “utility” pick out
25 These presuppositions, often left implicit, are discussed explicitly in Glimcher (2011).
246 Black Boxes
different concepts. The theoretical overlap is provided by the structure of the

black box. Viewed as a black box, utility may figure in economic explanations
both as explanans and explanandum. When we ask what determines my
selection of, say, coffee over tea, utility is an explanans, a difference-maker
that explains the choice itself. In contrast, when one inquires into the na-
ture of the mechanism underlying choice, utility becomes a frame, it is the
object of the explanation, the explanandum. With this distinction in mind,
it becomes clear that alternative theories provide competing explanations
of utility. Neoclassicists opt for a highly idealized “as if ” mathematical ap-
proach. Psycho-neural economists spell out a mechanistic model. These are
distinct difference-makers, represented in alternative models. At the same
time, both theories have a single objective in mind: an explanation of the
same economic behavior. The use of utility as an overarching frame to pin-
point the behavior under investigation is the common factor, making the
two frameworks different explanantia of the same core explananda. In short,
across neoclassical and psycho-neural economics, utility spells out the same
frame, but alternative difference-makers.
Time to take stock. Reference potential performs a role analogous to
Frege’s mode of presentation, making concepts commensurable across par-
adigm shifts. The black-boxing strategy explains how this can be done in
a rigorous fashion, without positing intensional entities, without lapsing
into meaning incommensurability, or other unpalatable forms of holism.
Gemmules, genes, mental states, and utility are all theory-laden concepts in
virtue of their heterogeneous reference potentials. What determines which
of these uses is preserved across theories and paradigm shifts is the structure
of these black boxes—framing, difference-making, and representation. We
can thus make sense of Hempel’s dictum that “concept formation and theory
formation go hand in hand.” Theories presuppose concepts, and concepts re-
quire a theoretical framework. Black-boxing shows how both points can be
held simultaneously, without getting us stuck in a vicious circle.
§9.6. Incommensurability, Progress,

and Black Boxes
In his 1982 article, “Commensurability, Comparability, Communicability,”

Kuhn asks: what is it that translation must preserve? Not merely reference,
he notes, since reference-preserving translations may well be incoherent
or incomprehensible. What about requiring that translation preserve sense

or intension? While Kuhn, at one time, endorsed this strategy, he now
recognizes that this cannot be quite right either. Without denying the im-
portance of addressing the thorny issue of meaning, Kuhn concludes the
aforementioned article by trying to avoid semantic talk altogether, replacing
it with a quasi-metaphorical discussion of how members of a language
community pick out the referents of the term they employ. How can it be,
Kuhn wonders, that people with very different criteria regularly pick out the
same reference for their terms? A preliminary answer notes that language is
adapted to our social and natural world, and the world is such that it does
not present objects and situations that would trigger different identifications.
But this begs a further, deeper question: what must speakers with disparate
reference-determining criteria share for them to count as speakers of the
same language, to be members of the same linguistic community?
By now it must be clear where, in my view, the invariants of translation

are to be sought. Unlike two members of the same language community,
speakers of mutually translatable languages need not share terms. “Rad” is
not “wheel.” But the referring expressions of one language must be match-
able to coreferential expressions in the other, and the lexical structures em-
ployed by speakers of the language must be the same, not only within each
language but also from one language to the other. Taxonomy must, in short,
be preserved to provide both shared categories and shared relationships be-
tween them. Where it is not, translation is impossible, an outcome precisely
illustrated by Kitcher’s valiant attempt to fit the phlogiston theory to the
taxonomy of modern chemistry. (Kuhn 1982, p. 683)
In my opinion, Kuhn is wise in eschewing talk of intentions, and he is correct

about the importance of preserving taxonomy. Yet, what exactly does it mean
to “preserve taxonomy”? On what grounds can we compare and contrast tax-
onomies across competing theories and paradigm shifts?
Kitcher’s notion of reference potential is a solid starting point in answering
this question. He advocates for the importance of examining the intentions
of speakers and to interpret them via a “principle of humanity.” But intentions
are notoriously slippery notions. I have tried to show that black-boxing
provides a strategy to refine these intuitions, making them more objective
and precise. Preserving taxonomy, from our perspective, is ultimately a
matter of identifying the right frame or difference-maker, while setting aside
248 Black Boxes
the precise nature of their realizers. Darwin, Mendel, and Crick have very
different conceptions of gene in mind—indeed, Darwin and Mendel did not
talk about “genes” at all. The same can be said about Priestley and Lavoisier.
These authors all share a coarse description of a set of explananda and the
underlying mechanisms. In short, Kuhn’s preservation of taxonomy can be
fruitfully understood as the unfolding of black boxes across theories. These
boxes capture the structure that remains constant across paradigm shifts, for
instance in the shift from the classical physics of Newton to Einstein’s relativ-
istic framework.
Time to tie up some loose ends. An account of progress is a central compo-
nent of any serious analysis of science. For a long time, philosophers assumed,
more or less implicitly, that the advancements of science could be depicted
as a gradual accumulation of truth and knowledge. Kuhn, Feyerabend, and
their colleagues are rightly credited with a decisive dismissal of this simplistic
view. However, their pars construens was not quite as persuasive as their pars
destruens. Referential models, the most popular solution to radical holism
and meaning incommensurability, are not structured enough to capture sci-
entific progress. Successful translation requires more than mere agreement
on reference. The black-boxing strategy attempts to blend the objectivity of
purely referential models with a mild—but inevitable—form of meaning ho-
lism to sketch a notion of progress that allows us to compare frameworks and
avoid the bogey of incommensurability.
In addition, a discussion of progress sheds further light on historical
examples. Darwin’s and Mendel’s work reveals how, as long as the struc-
ture of the black box is identified correctly, failure to pinpoint the under-
lying mechanisms does not affect the success and fruitfulness of a scientific
hypothesis. Skinner’s shortcomings warn of the danger of mis-framing and
mis-representing a causal explanation. Finally, the debate between classical
and psycho-neural economists shows that identity of reference is neither a
necessary nor a sufficient condition for progress. Utility is understood very
differently across these frameworks. But, at a coarse level of description, it
plays the same functional role as frame and difference-maker.
One final remark. Hopefully, the discussion in the previous three chapters
shows that my analysis of black boxes is not an attempt to rebuild philosophy
of science entirely from scratch. This prominent albeit neglected construct
captures a number of traditional themes—such as mechanisms, emergence,
and progress—and recasts them in a new light. At the same time, talk about
meaning, reference, and incommensurability does not necessarily drag us
back to hackneyed debates in the philosophy of language. Be that as it may,

we can now move on to the final leg of our long journey together. True sci-
entific achievement is a combination of decomposition, as emphasized by
reductionists, and autonomy, a trademark of antireductionism. It is time
to apply all of our insights to this central debate in current philosophy of
science.
10
Sailing through the Strait
Ich habe dir nicht nachgestellt

Bist du doch selbst ins Garn gegangen,
Der Teufel halte, wer ihn hält!
Er wird ihn nicht so bald zum zweiten Male fangen.
—J.W. Goethe, Faust, 1426–1429*
§10.1. Back to Bricks and Boxes
Our long excursus into the nature and structure of black boxes began with
an old dusty image, which was mainstream well into the twentieth century
and is still popular in some circles. This is the figurative depiction of science
as a slow, painstaking accumulation of truths. The goal of scientific research,
from this hallowed perspective, is to provide an accurate and complete de-
scription of the universe, or some portion of it. In this sense, the develop-
ment of science is akin to the erection of a wall or the tiling of a mosaic. The
building blocks of science are basic facts about the universe we inhabit.
Over the years, scholars—scientists, philosophers, historians, and many
others—have vocally denounced the misleading nature of this analogy. At
best, it is a drastic oversimplification. At worst, it utterly misses the mark.
Either way, it should not be taken too seriously. Popper’s suggestive image of
the scientist as a knight in shining armor battling the evil forces of darkness
has followed suit and faded away. Of course, truth, knowledge, and objec-
tivity remain important goals for the scientific enterprise. But they are only
one side of the story. The remainder involves what we do not know, what we
cannot grasp, what we get wrong. In a word, what is missing from the “brick-
by-brick” model of science is the productive role of ignorance.
* “You were not caught by my device //When you were snared like this tonight. //Who holds the
Devil hold him tight! //He can’t expect to catch him twice.” Translation by W. Kaufmann.
DOI: 10.1093/oso/9780190095482.003.0010
Sailing through the Strait 251
The cumulative image remains popular among the general public.

Journalists, politicians, entrepreneurs, and other non-specialists typically
presuppose, more or less consciously, the vision of science as a gradual ac-
cumulation of truths. The source of this misconception, I suggested at the
outset, is the standard packaging of the inquiry. Textbooks and lectures in-
troduce science as a bunch of laws, theories, facts, and other cut-and-dried
notions to be memorized and, eventually, internalized uncritically. This re-
silient stereotype trickles down from schools and universities to television
shows, newspapers, magazines, and other channels that perpetuate it among
the public. Only the small portion of students who attend graduate school
and pursue research careers is exposed to the true face of science. Real prac-
tice at the bench is quite different, and way more interesting, than the ossified
caricatures presented to outsiders in journals, books, and conferences.
At the same time, the textbooks from which students learn science are
written by those same specialists who, when pressed on the issue, adamantly
reject the brick-by-brick conception. This raises some questions. Why are
experts perpetuating an image of science that they themselves disavow? Are
there more fruitful ways of exposing the new generations to the fascinating
world of science? Can we do a better job at popularizing our research?
My answer can be broken down into a pars construens and a pars
destruens. The negative claim presents the two dominant models of science—
reductionism and antireductionism—as ultimately unsatisfactory. This is be-
cause, at a general level of description, neither resolves the tension between
two intuitive aspects of scientific practice. Reductionism captures the insight
that breaking a complex system down to a more fundamental level always
provides valuable information. But this seems to leave no room for produc-
tive ignorance, that is, the autonomy of higher-level explanations. Sure, from
a reductionist perspective, ignorance may be useful when it points out what
still needs to be done—in this sense, it may not be a hindrance. Still, knowing
is always better than not knowing, and replacing ignorance with more details
invariably constitutes a step forward. Antireductionism, on the other hand,
legitimizes the role of ignorance and autonomy in science, but has trouble
explaining how macro-descriptions can be objectively superior to their
micro-counterparts. When we focus on more local debates, modest reduc-
tionism and sophisticated antireductionism substantially overlap, making
the dispute largely semantic. After decades of debate, philosophy of science
appears to be stuck between Scylla and Charybdis. Can we find a different
route forward?
252 Black Boxes
My positive proposal is simple. Over the centuries, science has developed

a well-oiled technique for turning ignorance and failure into knowledge and
success. This practice hinges on the construction of black boxes.
References to black boxes are ubiquitous across the sciences and the
humanities. Intuitive familiarity with the expression may well be the reason
that compelled you to pick up this book in the first place. And the general
idea should be clear and intuitive enough. Black-boxing a phenomenon
involves isolating some of its core features in a way that it can be assumed
without further explanation, justification, or detailed structural description.
This, in a nutshell, is a black box. But that cannot be the entire story. Even a
cursory glance at the history of science makes it clear that not all black boxes
are the same. Some work perfectly well; others muddy the waters. Some are
constructed out of necessity; others are the product of error, laziness, or over-
sight. Some unify; others divide. In short, the well-known tales of Darwin,
Mendel, Skinner, and Friedman reveal that black-boxing is more complex,
nuanced, and powerful than is typically assumed.
At root, black boxes are placeholders. Yet, the role of placeholders in
science is more prominent, interesting, and subtle than people realize.
For one, there are (at least) two kinds of placeholders. Frames stand in for
behaviors in need of explanation. Difference-makers stand in for whatever
produces the behaviors in question. With this distinction in mind, I broke
down the construction of a black box into three constitutive stages: framing,
difference-making, and representation. This led to the definition of a black
box as a placeholder in a causal explanation represented in a model. The
simplicity of this one-liner is deceptive: clarifying the steps and underlying
presuppositions required long and technical discussions.
The second part of the book put this definition to use. First, I reassessed
our case histories in evolution, genetics, psychology, and economics from
the standpoint of black boxes. This motivated my contention that the his-
tory of science is a history of black boxes. Next, I argued that black-boxing
constitutes a “diet” form of mechanism. Finally, I applied black boxes to two
substantial philosophical issues: emergence and scientific progress.
It is time to wrap up our discussion by cashing out my last two promissory
notes. At the end of Chapter 1, I asked two related families of questions. First,
can we avoid both Scylla and Charybdis? Is there an alternative explanation
of the nature and advancements of science? What is it? Second, how does sci-
ence turn ignorance into knowledge? Black boxes are the key to answering
both sets of inquiries. The goal of this chapter is to substantiate these claims.
How does black-boxing reframe reductionism and antireductionism? How

does it capture the role of ignorance in science?
Here is the plan. Section 10.2 recaps what got us into the current stale-
mate. Section 10.3 argues that black-boxing captures the advantages of both
reductionism and antireductionism, while eschewing their most troubling
implications. Section 10.4 elucidates my claims by illustrating them with
our historical examples. I conclude that, as I mentioned at various stages
throughout the book, the history of science really is a history of black boxes.
Section 10.5 wraps up the discussion by cashing out the notion of productive
ignorance that is missing from our extant models. By that point we will be
ready for farewells.
§10.2. The Quagmire
Chapter 2 portrayed contemporary philosophy of science as figura-

tively sailing between two hazards: the Scylla of reductionism and the
Charybdis of antireductionism. Almost sixty years after the publication of
Nagel’s The Structure of Science—the epitome of the positivist ideal—the
battle between friends and foes of reductionism rages on. No winners or
losers are in sight, but both armies look fairly worn out. Before exploring
alternatives, it might be worth recalling what got us into this quagmire in
the first place.
Positivism maintained that all of science can be reduced to fundamental
physics, at least in principle. Reduction was conceived as a deductive re-
lation between theories, characterized as interpreted axiomatic systems.
This classical model of reduction has fallen on hard times, eroded by
powerful objections: the multiple-realizability of many prominent higher-
level kinds; the lack of appropriate laws, generalizations, and bridge
principles; and the shortcoming of “syntactic” conceptions of theories.
Taken together, they mount an overwhelming case against Nagel’s deriva-
tional model of reduction. The demise of classical reductionism divided
philosophers into opposite camps, depending on their methodological
inclinations.
One strategy aims at refining the reductive framework. The story goes
something like this. The classical model had the right ideal in mind, but
failed to implement it properly. Reduction should not be cashed out as a
logical relation between interpreted axiomatic systems. The main issue is
254 Black Boxes
a question about explanation. Lower-level descriptions, neo-reductionists

maintain, invariably deepen and enrich our understanding of systems,
technically rendering higher-level depictions disposable. Thus, all of sci-
ence can, in principle, be reduced to physics. But this is not because the-
ories in the special sciences can be logically deduced from physical laws,
or other underlying generalizations. They cannot. The reason is that, bar-
ring pragmatic considerations of computational tractability and conven-
ience, micro-descriptions provide the best explanations: the deeper, more
powerful, more comprehensive models. In its most radical, ambitious
formulations, physics regains its role as a true “theory of everything,” the
paradigm and limiting case of all of science, just like it has been for much
of its history.
The second, alternative strategy follows a diametrically opposite path,
eschewing reductionism tout court and treating the special sciences as “au-
tonomous.” More specifically, antireductionists agree with neo-reductionists
on the shortcomings of the classical model. However, they diverge on the
diagnosis and its implications. It is false, the classic antireductionist argu-
ment runs, that higher-level explanations are invariably empowered when
they are reassessed from a lower-level standpoint. Sure, in many cases, struc-
tural details are crucial for addressing some macro-explanandum. But this
downward trajectory does not extend all the way down to the foundations
of the discipline itself, let alone to physics. There is a threshold past which
adding structural information does not enhance higher-level explananda. If
anything, unnecessary details end up hindering connections and muddying
the waters. This, in a nutshell, is the autonomy of the special sciences.
Before moving on, it is useful to clarify, one final time, the nature of the
debate. What is at stake in the divide between friends and foes of reduction?
First and foremost, the contention is independent of the truth or fal-
sity of physicalism. Sure, a handful of philosophers still resist the
reduction of mental to physical states on the basis of Cartesian or Neo-
Cartesian dualistic assumptions. But most discussants, reductionists and
antireductionists alike, tend to be thoroughgoing materialists. The phys-
ical supervenience of biology, neuropsychology, economics, and other
special sciences is typically taken for granted. The crux of the matter is
what exactly follows from this assumption: should materialism be “reduc-
tive” or “non-reductive”?
Second, the fruitfulness of decomposing complex systems into
more basic constituents is not in question. This is not a contentious
methodological strategy.1 As new mechanists have helped clarify, it is a

truism that everyone, no matter their ontological inclinations, should take
for granted.
As I said, the heart of the contention concerns explanation. Do micro-
details invariably deepen macro-explananda? Are higher-level descriptions
autonomous from lower levels? Are physical accounts of Putnam’s square peg
deeper or overkill? Is his geometrical account incomplete, or does it strike
just the right balance? Is it true that all molecular findings enhance classical
genetics, or do we reach a point where further additions become unneces-
sary? These are the questions that have divided philosophers for decades.
So, who wins, the reductionist or the antireductionist? Is it going to be
the rock shoal or the whirlpool? To begin, note that pros and cons of both
stances, presented at the most general level, are unsurprisingly specular: one
is the mirror image of the other.
Reductionism provides a clear-cut account of how discoveries at the lower
levels may, in principle, inform and extend higher-level theories. Suppose
that one is trying to explain why table salt dissolves in tap water (cf. §2.3,
Chapter 2). Intuitively, the simplest strategy is to look at the chemical compo-
sition of these elements. But why should this be so? Why would an analysis of
H2O and NaCl shed light on the behavior of salt and water? Reductionism has
a prima facie compelling story to tell. If salt and water are reducible, in some
relevant sense of the term, to their molecular structure, then there is no mys-
tery why the nomological connection between H2O and NaCl explains their
respective behaviors. The apparent simplicity of this illustration should not
mislead us. Similar reasoning also sheds light on how biochemistry informs
functional biology or how neuroscience contributes to psychology.2 This is
what makes reductionism such an intuitive stance.
Antireductionism, in turn, is fueled by an equally compelling observa-
tion: the autonomy of higher-level models. Consider, once again, the example
just rehearsed. The relation between salt and water is correctly described
as one of solubility. This characterization appears to be perfectly adequate
and exhaustive. Furthermore, increased accuracy does not come for free.
Solubility is a relatively simple property. But it has the virtue of being quite
general. It captures the relation between salt and water, but it is also applicable
1 Thus, for instance, Mayr (2004), an outspoken antireductionist, calls this basic tenet “analysis”
and contrasts it explicitly with “reduction.”
2 I explore these issues, more systematically, in a series of articles: Nathan (2012, 2015b); Nathan
and Del Pinal (2016, 2017).
256 Black Boxes
to sugar and tea, stevia and soda, and many more substances. Details about
H2O and NaCl increase the depth of the explanation. But adding precision
will ipso facto impact its breadth, its generality. Given that salt and sugar have
different chemical structures, making the model too similar to the former
substance will make it inapplicable to the latter. In short, there is a trade-off
between accuracy and generality, and the golden mean is not invariably, or
even typically, found at the lowest, most fundamental levels. This, simply put,
fuels the methodological autonomy of the special sciences, the heart and soul
of contemporary antireductionism.
Viewed in this light, the choice between reductionism and antireduc-
tionism may well turn out to be an ill-posed one. Sophisticated versions of
both these influential stances stress different, equally legitimate aspects.
Philosophy tends to view them as incompatible, but this is a mistake. As we
saw, when we focus on concrete examples, initially apparent differences fade
away. Try asking whether current science better conforms to reductionist or
antireductionist standards, and the matter tends to lose substance.
The stalemate becomes especially evident in the biological sciences, where
much is known about the structural implementation of function. Can all bi-
ology be reduced to molecular biochemistry? The trouble with this question
is that it requires and presupposes a rigorous definition of “molecular pro-
perty” that we currently lack and will likely never be found. The crux of the
disagreement is whether our best biological models are molecular, and this
ultimately depends on whether we classify explanations as “molecular.” This
is a semantic, terminological issue, not a substantive one.
Philosophers of mind, psychology, neuroscience, and economics should ac-
cept the dire lesson from their colleagues working on biology. Empirical dis-
coveries are unlikely to solve long-standing philosophical disputes over the
mind-body problem or the relation between economics and psychology. The
reason is not the—indisputable—complexity of these fields. The issue is the na-
ture of reduction which, contrary to common wisdom, is murky. All of these
observations point to the same conclusion: it is time to move on.
§10.3. Rocking the Rock, Weathering the Whirlpool
Section 10.2 recapped the argument, originally spelled out in section 2.6 of
Chapter 2, that the current standoff between reductionism and antireduc-
tionism is not as substantive as it is typically taken to be. Our present goal is
to blaze an unbeaten trail, find an alternative way forward. My suggestion is

to reframe the current debate from a different perspective: the standpoint of
black boxes.
Traditionally, philosophers pitch autonomy and reduction against each
other. Following Putnam’s lead, autonomy is conventionally defined as the
rejection of reduction. Reduction thus ipso facto becomes the denial of au-
tonomy. As just presented, these two stances are obviously mutually incom-
patible, in virtue of their meaning alone. But could this be a mistake? Why not
accept both tenets simultaneously? Could we not maintain that descending
to a lower explanatory level invariably enhances explanatory power, without
thereby giving up on the autonomy of higher-level description? Can we not
have our cake and eat it too? I want to argue that, yes, we can.
How should the issue of (anti)reduction be repackaged so as to make it
more substantive and significant? This section sketches an answer to this
question, based on the black-boxing recipe developed in previous chapters.
The following section (§10.4) will illustrate the proposal by applying it to the
case studies that accompanied us during our journey together.
Let us consider the relation between higher and lower levels—intra-field
or inter-field, it does not matter.3 On the one hand, it is clear that micro-
depictions provide details that deepen macro-explanations in various ways.
This is hardly a shock: discovering the workings and implementation of a
complex system is key to understanding its behavior. This is the selling point
of reductionism. At the same time, it seems equally compelling that higher
levels are often autonomous, in the sense that macro-claims can be estab-
lished, corroborated, and explained without introducing micro-details. This
is the benchmark of antireductionism. Can both points be maintained si-
multaneously? Can higher-level descriptions be both reducible to and au-
tonomous from the underlying substrate? How can we use structural details
to advance macro-explanations without thereby threatening their indepen-
dence? The key lies in Putnam’s insights. He had the right intuition when he
stressed the centrality of autonomy. The mistake was turning a discussion of
autonomy into a debate about reduction.
Before delving into the argument, it seems only fair to note an analogy
between my proposal and an influential precursor. The possibility and
3 Recall from Chapter 1 that this characterization of levels as “coarse-grained” vs. “fine-grained,”
“micro” vs. “macro,” or “higher” vs. “lower,” should be understood as relativized to a specific choice of
explanandum.
258 Black Boxes
importance of finding some sort of middle ground between reductionism

and antireductionism has been discussed in the literature. Notably,
philosophers working within the new wave of mechanistic philosophy have
presented their approach as an attempt to move past the current dichotomy:
Our project began as an attempt to understand reductionistic research in

science, but we soon found that neither the classic [Nagelian] philosoph-
ical treatment of reduction [ . . . ] nor various modifications that followed
[ . . . ] offered an appropriate framework for the biological research cases we
had chosen to analyze. We ended up recharacterizing the science in terms
of mechanistic explanations in which systems are decomposed into their
component operations and dropping use of the word reduction altogether.
(Bechtel and Richardson 2010, p. xxxvii)
As stressed in Chapter 7, I am intellectually indebted to and sympathetic

with the broad mechanistic outlook. Yet, I believe that their blade hardly cuts
deep enough. Mechanistic philosophy has failed to transcend old divisions,
as witnessed by the observation that whether the new wave of mechanism
fits in better with a general reductive or antireductionist stance is a matter of
constant debate. Some see a connection, more or less explicit, with modest
forms of reductionism.4 At the same time, because of its focus on abstraction
and idealization, which confer to higher-level mechanistic description an au-
tonomy of sorts, others present mechanistic philosophy as an alternative to
reduction.5 What this shows is that mechanistic theory has not freed us from
the old dichotomy. I would like to take this one step further, and move away
from both stances, stressing how they can be effectively reconciled. With
these clarifications in mind, let us get back to business.
Here is the punchline. The problem with reductionism vs. antireductionism
debates is the more or less tacit presupposition that macro-explanations
4 For instance, Glennan (2017) notes: “Because New Mechanists emphasize the ways in which
the organized activities and interaction of parts explain the behavior of wholes, it might seem that
New Mechanist ontology is committed to a pluralist (and reductionist) view” (p. 56, italics added).
Glennan is adamant in stressing that his approach avoids radical “nothing-but” forms of reduc-
tionism. Still, it has a clear reductionist flavor. Similarly, Godfrey-Smith (2014) notes, “This kind of
[mechanistic] work is “reductionist,” in a low-key sense of that term: the properties of whole systems
are explained in terms of the properties of their parts, and how these parts are put together” (p. 16).
5 In the words of another prominent advocate of new mechanism, “I argue that the mosaic model
of the unity of neuroscience, based on the search for mechanistic explanations, is better suited than
reduction to the descriptive, explanatory, and epistemic projects for which these classic models were
designed” (Craver 2007, p. 233, italics added).
and their micro-counterparts have the same explanandum, the same object
of explanation. What most discussants fail to note is that, by transforming
higher-level questions into lower-level ones, we are not providing different,
competing answers to the same query. Inquiries framed at different levels are,
for all intents and purposes, different questions. Failure to acknowledge this
simple—albeit powerful—point leads to much confusion, including viewing
reduction and autonomy as antithetical, when they are not. This is the ful-
crum of my argument. I now need to put some meat on these bare bones.
What is the relation between hypotheses at different levels? Borrowing a
Kuhnian metaphor, one could say that explanations with substantially dif-
ferent scope are typically “incommensurable.” As we saw in Chapter 9,
incommensurability was one of the main tenets of Kuhn’s philosophy of sci-
ence. His claim that scientific paradigms are “incommensurable” is often
understood as a radical form of conceptual relativism. This, simply put, is
the claim that a paradigm shift alters the language of a theory so drastically
that superficially similar claims turn out not to be mutually translatable
across paradigms. This, as we discussed at length in the previous chapter,
is too strong. Paradigms are seldom, if ever, incommensurable in this ex-
treme sense.
Our excursus into the nature of black-boxing draws attention to a subtler,
more compelling, and less devastating form of incommensurability. This is
the claim that hypotheses and explanations framed at different levels cannot
be compared directly. They are always mediated by models, theories, and
other vehicles of representation. This context-relativity boils down to a sort
of testing holism, albeit one whose consequences are fairly mild. Basically,
it is always possible to adjudicate between competing hypotheses directly,
when these alternatives are embedded within the same model. Nevertheless,
whenever one is trying to contrast explanations pertaining to different
models, what is compared are not the individual hypotheses themselves, but
a broader theory or paradigm. I should stress that my point is not the rad-
ical claim that hypotheses can never be compared across models. That would
be tantamount to advocating a radical form of meaning incommensurability
and giving up on the progress of science. My suggestion is that when one
compares, say, a Newtonian explanation with its relativistic counterpart, one
is not assessing hypotheses directly. Rather, in a more Duhemian fashion,
one judges the explanatory forces of two entire paradigms. As we shall see in
the following section, a similar relation holds when we contrast genetic vs.
molecular explanations, or psychological vs. neuroscientific ones.
260 Black Boxes
These considerations raise an important question. What is the relation be-

tween paradigms? How do we assess two hypotheses or explanations pitched
at distinct levels or scale? Allow me to address this issue by introducing an
important—if sometimes forgotten—philosophical distinction.
In his classic “Empiricism, Semantics, and Ontology,” Carnap (1956a)
observes, “If someone wishes to speak in his language about a new kind of
entities, he has to introduce a system of new ways of speaking, subject to new
rules; we shall call this procedure the construction of a linguistic framework
for the new entities in question” (p. 206). Carnap then draws a divide be-
tween two kinds of questions concerning ontology, or what there is. On the
one hand, there are questions concerning the existence or reality of entities
within a framework—call these “internal” questions. On the other hand,
there are questions concerning the existence or reality of the system of enti-
ties as a whole. Carnap calls these latter questions “external.”
A few examples should clarify the main idea. Are numbers real? This ques-
tion, Carnap maintains, is subject to two interpretations. On the one hand, it
could be understood internally, that is, relative to the system of numbers of
basic arithmetic. From this perspective, the answer is trivially true: of course
there are numbers in the framework of arithmetic. Note, however, that not
all questions of this sort are obvious. Consider, for instance, the existence
of the largest prime number, or Goldbach’s conjecture, the currently unre-
solved hypothesis that every even integer greater than 2 can be expressed as
the sum of two primes. Here the answer is far from evident. Still, what all
these hypotheses have in common is the presupposition of a framework, the
system of arithmetic. This is what makes them internal.
On the other hand, questions concerning the existence of numbers can be
understood externally. Thus conceived, in asking whether numbers are real,
one is inquiring into the ontological status of numbers, that is, their “reality”
independent of arithmetic or any other linguistic framework.6 How is one
supposed to adjudicate ontological questions from this external perspective?
Carnap contends—correctly in my opinion—that such a question, strictly
speaking, has no cognitive content. Yet, this is not to dismiss it as gibberish.
The right interpretation is not as whether or not, say, numbers have the pro-
perty of existence. Rather, it is a pragmatic issue of whether or not we should
accept or introduce the framework of natural numbers, as opposed to a dif-
ferent one. The choice between frameworks is not one capable of being true
6 For an excellent example of this kind of “external” inquiry, see Benacerraf (1965).
or false. In this sense, Carnap claims, it lacks cognitive content. It is a prag-

matic matter of usefulness.
Hang on a second. Some readers may wonder whether they have missed
something. Since the very beginning of this work, I have declared positivism
dead and philosophy of science in need of a different guiding spirit. And,
now, am I really drawing upon the work of one of the main figures in log-
ical empiricism, namely, Rudolf Carnap? Short answer: yes, yes, and yes.
When critically assessing a historical movement, we must be careful not to
throw out the baby with the bath water. Logical positivism is no exception.
The naïve derivational reductionism and the cumulative conception of prog-
ress ought to be abandoned. In contrast, Carnap’s distinction between in-
ternal and external questions, as well as its connection to Quine’s attack on
the analytic-synthetic distinction, remains very important for contemporary
philosophy.7 All of this just goes to show how black boxes—and philosoph-
ical discussions thereof—are full of surprises. One can never predict in ad-
vance what is going to pop out once you open them.
So, where am I going with this? Carnap’s notion of linguistic framework,
and his related distinction between internal vs. external questions, can be
applied to the explanatory-relativity that is integral to the black-boxing ap-
proach. Allow me to elaborate. Recall from section 5.2 of Chapter 5 that all
requests for explanation presuppose a context. Once the backdrop is in place,
we can ask whether the explanation is plausible, correct, if it can be amended,
etc. These are all internal questions, since they occur relative to the choice of
a model—or, at least, a preliminary frame. Consider a simple example from
genetics. Suppose we are wondering which among two genes, call them g1
and g2, is responsible for transcribing protein p. This question is an empirical
one, in the sense that we have two candidate answers to the inquiry: which
gene transcribes p, and we need to figure out the solution.
Nevertheless, it often happens that competing explanations do not share
a background theory, framework, or paradigm. Consider, for instance, al-
ternative accounts of human behavior that appeal to genetics vs. upbringing
(nature vs. nurture), or psychological vs. neuroscientific characterizations of
behavior. Even if, at very general levels of description, such depictions con-
verge on the same explananda (“inheritance,” “behavior”) the explanations
7 Coming from a similar perspective, Bickle (2003, Ch. 1) resurrects a Carnap-inspired internal-
external distinction as part of meta-science. For a discussion of the revived importance of Carnap’s
distinction in contemporary ontology, see Chalmers et al. (2009).
262 Black Boxes
themselves cannot be compared directly, as they are framed quite differently.

As such, they may not share much in terms of precisely what patterns of events
they are trying to capture, as well as their standards and presuppositions. In
such cases, we need to compare entire models, not individual explanations.
In this sense, these explanations are incommensurable: they cannot be com-
pared directly; they require the mediating role of a model. But models are
useful or useless, clear or obscure, illuminating or not. They are not true or
false. Hence, the choice between models is not a cognitive one, in Carnap’s
sense. It is an external question, a choice involving which framework to
adopt. Note how well this fits in with the “diet” mechanism of Chapter 7, ac-
cording to which whether a system is depicted as an entity or as a box inher-
ently depends on the choice of model.
In conclusion, reductionists and antireductionists share the presupposi-
tion that explanations pitched at different levels provide competing explan-
antia of the same explananda. This is a mistake. They do not. Explanations
are context-relative. It should now become clear in what sense higher-level
explanations can be both reducible and autonomous from lower levels.
Reduction is a relational issue, concerning the reframing of a macro-
explanandum as micro-inquiry. Autonomy is an intrinsic property of an
explanation, pertaining to whether a model provides the resources to ade-
quately address its explananda. Thus conceived, these two properties are per-
fectly compatible.
Let me illustrate the coexistence of autonomy and reduction by focusing
on our toy example. Scientific scenarios will be covered in section 10.4.
In Putnam’s square-peg round-hole thought experiment, the core issue
is whether or not structural physical details are relevant to the explanation
of the behavior of the system and, if so, whether they increase explanatory
power. Unless we clarify what it is that we are trying to explain, such a ques-
tion is meaningless. If the explanandum is the observation that the square
peg does not pass through the round hole, then the physical details can—
indeed, must—be effectively black-boxed. In contrast, if the objective is
explaining why this state occurs, then looking at the physical structure of the
system, down to its atomic constituents, becomes relevant. In this case, the
black-box must be opened and replaced by a mechanism.
Putnam’s deep insight was recognizing that higher-level explanations
can be epistemically autonomous from lower-level ones. His error, which
turned out to be quite consequential in the wake of the ensuing debate, was
to spin this as a vindication of antireductionism. Putnam, like many other
philosophers thereafter, identified autonomy with antireductionism. He was

wrong. These concepts should not be conflated. The reason is that, prop-
erly conceived, the reductionism vs. antireductionism debate involves ex-
ternal questions concerning the choice of a model and levels of explanation.
Autonomy, in contrast, underlies an “internal” question concerning the in-
dependence of a causal explanation given the choice of a model.
§10.4. A History of Black Boxes
Section 10.3 argued that the black-boxing strategy reconciles autonomy and
reduction, the signature traits of antireductionism and reductionism, re-
spectively. Let us see how this works in practice by revisiting, one final time,
the case studies that accompanied us throughout our journey together.
§10.4.1. Beyond Reductionism and

Antireductionism: Evolution
To get started, take one last look at Darwin’s explanations. Recall that his ex-
plicit target is the distribution of organisms and traits around the world. His
evolutionary explanans, simply put, is descent with modification, fueled by
natural selection. Is this a story of reductionism or antireductionism?
Here is a reductionist rendition of the tale. Darwin is surely correct that
descent with modification is the central factor in explaining distributions
of organisms and traits across the globe. Evolution by natural selection is,
indeed, the principal frame and difference maker. But how should we un-
derstand evolution by natural selection? Darwin himself breaks down the
process into four key ingredients: variation, competition, fitness, and herit-
ability. This was an important insight. But it was only the beginning. In the
wake of Darwin’s groundbreaking work, progress was gradually achieved by
decomposing these broad concepts into more fundamental components.
This is precisely what one would expect from a reductionist perspective.
To further elaborate, consider Sober’s example of frequency changes in a
population of Drosophila. Initially, variation at the population level may be
explained by positing that type A flies are “fitter” than type B ones. As the
system is studied further, details emerge. In the imaginary case at hand, it
turns out that the fitter type A is characterized by a chromosome inversion
264 Black Boxes
that produces a thicker thorax, better insulating the organism, which makes
it more resistant to cold weather. As Sober notes, at this point, appeals to “fit-
ness” become disposable. Fitness attributions can be replaced by descriptions
of the mechanisms producing the relevant frequencies. And, indeed, this
more precise depiction provides a deeper account of changes at the popula-
tion level. In short, “fitness” is explanatory. But describing the mechanisms
responsible for differential survival is more explanatory. And further decom-
posing these genetic mechanisms into more fundamental constituents will
make the model even more powerful. This downward trajectory, the reduc-
tionist says, is captured by the reductive outlook.
Not so fast, the antireductionist retorts. Sure, the chromosomal descrip-
tion is more detailed than the preliminary ascription of fitness. Still, is this, in
and of itself, a vindication of reductionism? Accuracy is an important aspect
of explanation. But it is not the only factor at play. Shifts between levels in-
volve a trade-off in explanatory power. Micro-descriptions are more precise;
macro-depictions are more general. To wit, what do the fittest Bacteriophage
λ, the fittest Drosophila melanogaster, and the fittest Homo sapiens have in
common, which sets them apart from other members of their species? Good
luck answering this question by pinpointing shared biochemistry. And
insisting that, from a biophysical perspective these organisms have nothing
in common just goes to show that moving down on our levels of descriptions
is a compromise. Some features are gained, while others are lost. In short,
lower-level depictions are not invariably more powerful than higher-level
ones. They merely focus on different aspects. Fitness is just an instantiation
of a broader phenomenon. Darwin’s explanations in the Origin, which are
still widely accepted by evolutionists, vindicate the antireductionist story.
At this point, reductionists will presumably point out all the contributions
of molecular biology, broadly construed, to the study of evolution.
Sophisticated antireductionists should agree. But they also maintain that
what reductionists insist in dubbing “reduction” is really a case of multilevel
integration. Modest reductionists, in turn, will hold their ground, claiming
that this is all part of a molecular basis. And—here we are, back right where
we started! Both parties recognize Darwin’s remarkable contributions and
the progress that has been made since. The disagreement is whether this
should be recounted as a story of “reductionism” or “autonomy.”
How about trying something different? Consider the situation from the
perspective of black-boxing. First, note that variation, competition, fit-
ness, and heritability are placeholders in causal explanations framed in
models—they are black boxes. The construction of a black box begins by

sharpening the object of explanation. Next, one determines what brings
about this explanandum and embeds the causal story within a model.
The significance of the framing stage is that macro-explanations and their
micro-counterparts have distinct explananda, which presuppose sui generis
explanantia. If the target is an explanation of frequency changes in the fruit
fly population, fitness provides a perfectly adequate explanans. What fit-
ness alone does not explain is why type A flies are more resistant than type
B flies. This is an important question, albeit one that frames a very different
explanandum. From this perspective, reductionists are correct in stressing
that molecular biology deepens our evolutionary accounts. At the same time,
this is not to reject the autonomy of evolutionary biology. This is because
explanations are relativized to contexts. As a result, we cannot compare them
directly. What is compared are entire models, not individual hypotheses.
In sum, does the story of Darwin vindicate reductionism or antire-
ductionism? Both or, perhaps, neither. Either way, it hardly matters.
Reconstructing the situation in terms of black boxes shows that micro-and
macro-explanations are both autonomous and reducible. They are autono-
mous because, when appropriately framed, higher-level depictions stand
alone, independent of structural details. At the same time, higher-level ac-
counts are reducible because there are models that provide more detailed
and more powerful lower-level depictions. The mistake, common to both
reductionists and antireductionists, was assuming that these two stances are
incompatible. The root of the trouble is presupposing that the two explan-
antia have identical explananda. They do not. Asking whether lower-level
explanations are “better” than higher-level ones is an ill-posed question. The
explanations are incommensurable. What is compared are not individual
hypotheses, but the models, the paradigms that embed them. This compar-
ison is not a matter of truth or falsity. It corresponds to what Carnap dubbed
an “external” question, to be addressed on purely pragmatic grounds.

Antireductionism: Genetics
Our second case study focuses on the development of genetics, from Mendel’s
pioneering insights to the contemporary landscape. Originally showcasing
266 Black Boxes
the limitations of Nagel’s account, the relation between classical and molec-
ular genetics has grown into a poster child for antireductionism.
Mendel’s explicit targets are the observable inheritance patterns, some-
times referred to as “Mendelian ratios.” Each organism, Mendel hypothe-
sized, inherits two “factors”—subsequently called “genes”—one from each
parent. When these factors are different, one is always expressed preferen-
tially over the other. These factors are passed on, unchanged, to the next gen-
eration. Why would this well-known story support antireductionism?
To restrict the scope of the discussion, consider Mendel’s second law, the
“law of independent assortment,” which, simply put, states that genes located
on non-homologous chromosomes assort independently. This generaliza-
tion, as Kitcher maintains, is best explained at the cytological level. To be
sure, the process of meiosis can be described, in much greater detail, in bi-
ochemical terms. But does this knowledge enhance the original cytological
depiction? Antireductionists answer in the negative.
Many reductionists disagree, complaining that this perspective is simply
outdated. Molecular biology has greatly improved the study of gene repli-
cation, expression, mutation, and recombination. Insisting that explanatory
power is, in principle, immune from lower-level revision and enrichment
is simply wrongheaded. Mendel’s insight was on the right track. But much
remained to be done. What is the physical structure of a gene? How are these
traits inherited, transmitted, and expressed? We now have a much deeper
understanding of genes and other genetic processes. Mendel’s insights have
not just been vindicated. They have been expanded and explained.
The problem with this response, from an antireductionist perspective,
is that it sets up a straw man. Of course, biochemistry and other molecular
insights enhance our knowledge of genetics. This was already acknowledged
in Kitcher’s pioneering “1953” article, where molecular biology is presented
as an “explanatory extension” of classical genetics. The point, stressed by
antireductionists ever since, is that higher-level natural kinds do not count
as natural kinds at all from a lower-level standpoint. Recall how the classical
Mendelian concept of a gene was supposed to fulfill three roles. It was in-
tended as the unit of mutation, the unit of recombination, and the unit of
function. No single biochemical entity fulfills all three. Hence, reductionism
distorts the success of classical pre-molecular genetics. We need an antire-
ductionist outlook to do justice to these groundbreaking discoveries.
Now it is time for reductionists to complain. Multiple-realizability did
pose an inescapable trap for classical reduction, which required a series of
type-identities across levels. New-wave epistemic reductionism, in contrast,

does not fall prey to this objection. What if no single biochemical unit cor-
responds to Mendelian genes? Is it just that the microcosm is more complex
than Mendel himself and his early twentieth-century followers could sur-
mise? Furthermore, of course we need to employ functional language to cor-
rectly study genes, chromosomes, and the like. Molecular biology is not an
attempt to purge biology of all talk of teleology. Au contraire, it is an attempt
to “naturalize” the field, by uncovering the entities, processes, and activities
that instantiate these functions and dispositions. This process enhances ex-
planatory power, vindicating the reductionist maxim.
It should be clear by now where the discussion has gone astray. The issue is
not whether biochemistry can shed light on genetics, evolution, or the rest of
biology. It does, and everyone is well aware of this. The question underlying
the debate is whether the relation between classical and molecular genetics
better conforms to the reductionist or the antireductionist paradigm. And, as
noted in section 2.6 of Chapter 2, this ultimately hinges on the definition of
“purely molecular language” and other terminological conventions.
Let us try reassessing the situation from a different perspective. Genes
and other genetic constructs are black boxes. They are placeholders in causal
explanations framed and represented in models. In this regard, reductionists
are correct in pointing out that biochemistry provides much deeper and
more insightful analyses, compared to the relatively shallow accounts of clas-
sical genetics. Thus conceived, the realization that, from a biochemical stand-
point, the Mendelian gene does not exist as a single structure is a vindication
of the reductionist strategy, not a failure thereof. At the same time, this does
not entail a rejection of autonomy at higher levels. We still teach Mendel’s
laws to beginning biology students. This is because these generalizations cap-
ture important biological facts and convey them well, striking the right bal-
ance between precision and generality.
In sum, reductionism and antireductionism are both correct, but not
exclusively so. Reductionism rightly notes that the molecular paradigm
provides deeper, more accurate, and more detailed explanations. The insight
of antireductionism is the autonomy of higher-order paradigms. Both parties
ultimately miss the mark by tacitly presupposing that these explanations are
directly in competition with one another. They are not. Strictly speaking, the
explanations are incommensurable. This, however, does not mean that mo-
lecular biology is incomprehensible from the perspective of cytology, or vice
versa. The point is that what is compared is not individual hypotheses, but
268 Black Boxes
entire theories, frameworks, paradigms. And the choice between them is “ex-
ternal,” that is, pragmatic. Thus construed, Mendelian genetics is both auton-
omous and reducible to molecular biology.

Antireductionism: Behaviorism
Radical behaviorism in psychology provides a different, interesting case

study for a simple reason: it is a story of failure. Virtually no contemporary
scholar, in psychology or philosophy, flatly identifies with a full-fledged
behaviorist, at least when behaviorism is conceived as a theory of mind.
Consequently, the question underlying Skinner’s ultimately unsuccessful at-
tempt to provide new foundations for the field of psychology becomes: was
he guilty of a reductionist mistake or an antireductionist one?
Skinner’s target is a complex reconstruction of psychology that eschews
any appeal to the internal mental states of agents. As noted in section 6.4
of Chapter 6, Skinner’s rejection of mentalism is more complex than is
often assumed. The crux of his argument can be phrased in terms of the
theoretician’s dilemma. If, on the one hand, mental states link environmental
stimuli and behavior in a deterministic fashion, the story goes, mental states
become completely unnecessary for the explanation of behavior. If, on the
other hand, the mental-state-mediated links between stimuli and beha-
vior are indeterministic, the mental states are utterly useless. Either way,
references to mental states do not contribute to psychological explanations
of behavior and should therefore be omitted. Environmental inputs are all we
really need.
Skinner’s theoretician’s dilemma does not hold up to serious scrutiny. It is
controversial, to say the least, that deterministic links of the form E → M → B
make appeals to M unnecessary. After all, even in a fully deterministic uni-
verse, M could provide better, more accurate explanations of behavior than
E. This first horn of the dilemma requires further argument to become con-
vincing at all. But the real problem emerges when we shift to the second horn.
Why would indeterministic links make M disposable? Non-deterministic
generalizations may be perfectly explanatory, as witnessed by their exten-
sive use throughout the sciences, from quantum mechanics to evolutionary
theory, from biochemistry to economics.
In short, Skinner’s negative argument against mentalism leaves us

wanting. Moreover, all positive evidence points to an indispensable role of
mental states, broadly construed, within psychology. Skinner was wrong on
this score, or so it seems. But what moral should one draw from his mistakes?
Antireductionists will likely blame reduction. Skinner’s shortcoming lies
in his attempt to reduce mental states to complex behavioral dispositions.
Interestingly, from this perspective, the error is pervasive across psychology
and the philosophy of mind. Behaviorism was hardly the only reductive
theory of mind developed in the century just passed. U. T. Place and J. J.
C. Smart’s identity theory purported to reduce mental states to brain states,
a view subsequently developed by Francis Crick’s “astonishing hypothesis”
(1994) Eliminative materialism, a position advanced and advocated by
Patricia and Paul Churchland, maintains that mental states are theoretical
entities posited by folk psychology. Widespread and commonsensical as it
is, folk psychology, they argue, is a flawed theory of mind and, as such, it is
not a candidate for integration. Rather, it should be eliminated and replaced
by a mature neuroscience, which will be more predictive, more explanatory,
and better connected to extant research. Despite evident differences, from
an antireductionist perspective, behaviorism, identity theory, and elimina-
tive materialism partake of the same root problem: they purport to reduce
or replace mental states with something else. Contemporary neuroscientists
who follow suit and identify mental states with brain states are committing
the same kind of mistake as Skinner, and their proposals are doomed for
precisely the same reasons. The solution is to adopt an altogether different
theory of mind, one that does not dispense with mentality. Functionalism,
token-identity theory, and property dualism are all popular candidates.
Reductionists will surely provide a very different reading of the situation.
The trouble with radical behaviorism, this story goes, has nothing to do with
reduction per se. The mistake lies in the type of reduction. There is more to
mental states than behavioral dispositions. The key to mind-body reduction
is an appropriate neuroscientific foundation for psychology.
In sum, the quest for the nature of mind has turned into a battleground
for reduction. Can mental states be reduced to physical states? If so, how? If
not, why not? This reducibility of mental to physical states, or impossibility
thereof, lies at the core of twentieth-century philosophy of mind.
If reductionism is so central to contemporary philosophy of mind, we
asked back in section 2.6 of Chapter 2, where do we stand with respect to
psycho- neural reduction, after decades of extensive debate? As noted,
270 Black Boxes
not too much success can be boasted. To be sure, much progress has been
achieved in the unraveling of psychological and neural mechanisms under-
lying higher and, especially, lower cognition. But has this advanced the phil-
osophical debate over the reduction of mind? If so, the news has not yet been
broken, as there seems to be no more consensus on this matter than there was
in the 1950s.
It is worth recalling that, when confronted with this lack of resolution,
scientists and philosophers alike tend to respond by pointing to the com-
plexity of the human brain, with its astronomical number of connections
among billions of cells. I obviously do not question how hard it is to study
brains and minds. I am, however, skeptical that the complexity of the under-
lying structure is responsible for lack of tangible progress. The main culprit is
the notion of (anti)reductionism itself. This should be old news. What’s novel
is that we can now finally see how to overcome the false dichotomy.
Consider behaviorism from the perspective of black boxes. Watson and
Skinner’s attempt to place psychology on more secure methodological
grounds had a momentous impact on the entire field. First, it stressed the
tight connection between mental states and behavior. Second, and more
important, it contributed to making psychology more “scientific.” Their in-
sight to black-box mental states was indeed revolutionary. By setting psycho-
neural patterns and mechanisms aside, Skinner and colleagues were able to
draw attention to the importance of stimuli, operant conditioning, and other
forms of environmental effects on human conduct. At the same time, rad-
ical behaviorists believed that the content of this black box had no substan-
tial place in psychology. This was their crucial mistake. In the wake of the
cognitive revolution, it became increasingly clear that mental dispositions,
and other psychological states, play an irreplaceable role in the explanation
of human conduct. As we saw, Watson did embrace a simplistic and naïve
reductionism. Not so Skinner, who was guilty of neither reductionism nor
antireductionism, which are external, pragmatic stances. His error is best un-
derstood in terms of the black-boxing strategy. His explanandum is framed
incorrectly. Consequently, the difference makers are misidentified, and the
ensuing causal model of behavior is lacking. All of this is independent of re-
ductionism and autonomy. Skinner’s shortcomings lie in the details of his
model of behavior—the construction of his black box.

Antireductionism: Economics
The heart and soul of neoclassical economics is the attempt to predict and ex-
plain economic behavior on the basis of choice-related data. The ingredients
for this simple, yet controversial, recipe include formal postulates like the
weak and general axioms of revealed preference (“Warp” and “Garp”), as
well as the ideal of rationality at the core of Expected Utility Theory.
The past few decades have witnessed the rise of alternative approaches,
which I subsumed under the moniker “psycho-neural economics.” The basic
insight is that details concerning how human minds and brains actually
frame, compute, and resolve problems challenge fundamental assumptions
regarding the behavior of agents. For instance, behavioral economists stress
how people rely heavily on heuristics, biases, and reference points when
adjudicating potential outcomes, as opposed to computing expected utilities.
And, clearly, from the standpoint of psycho-neural economics, heuristics,
biases, and reference points are not the means by which rational agents ap-
proximate the predictions of expected utility theory. Certainly Friedman, or
someone of his behalf, could make the argument that psychological data of
various kinds do not disconfirm or otherwise call into question his “as if ”
approach.
An informed prediction of the trajectory of economics lies beyond
my interest and professional competence. The important point, for pre-
sent purposes, is that the debate between neoclassical and psycho-neural
economists can be reconstructed along the reductionism vs. antireduc-
tionism lines. But should it? Unsurprisingly, I offer a negative answer and an
alternative.
The reductionist story goes like this. Characterizing utility in terms of re-
vealed choice was an effective strategy early in the twentieth century, which
allowed economics to rest on more secure methodological grounds, making
it kosher from a scientific— read: positivist—standpoint. Specifically, it
allowed economists to transform utility ascriptions from spooky unquanti-
fiable mental states to empirically testable hypotheses. This effective strategy
faced some shortcomings. In particular, it was founded on a series of “as if ”
models, introducing various unrealistic psychological assumptions. This
was a necessary move decades ago, when little was known about how this
information is actually processed. The landscape, however, has changed
drastically. We now know a lot more about the psychological mechanisms
272 Black Boxes
underlying human decision-making. Consequently, economics can advance

by replacing “as if ” models with more realistic psycho-neural assumptions.8
Given the methodological ties linking early neoclassical economics to
positivism, which explicitly endorses the in-principle reducibility of all of
science to physics, it might seem paradoxical to present this branch of eco-
nomics as a form of antireductionism. Nonetheless, from our contemporary
perspective, Friedman’s influential methodology of positive economics bears
a striking resemblance to antireductionist arguments supporting autonomy.
Friedman defends the foundations of neoclassical economics from the
charge that it presupposes hypotheses known to be false. His discussion of
the law of free-falling bodies, leaves on trees, and professional billiard players
are intended to put economics on a par with other natural sciences, as well as
common sense. We know all these hypotheses to be false. And, yet, everyone
accepts them as unproblematic. Then why the fuss about individual firms
behaving as if they strive to maximize profit? Truth or falsity of assumptions,
Friedman famously concludes, is irrelevant with regard to the acceptance
or rejection of a scientific theory. What really matters is whether the hypo-
thesis is predictive and explanatory with respect to the class of phenomena
that it is intended to cover.9 Hence, the relevant question to ask about the
assumptions of a theory is not whether they are descriptively realistic. They
never are. The issue is whether the hypotheses in question are good enough
approximations for the purposes at hand.
These analogies spell out a powerful argument in favor of the autonomy of
economics from neighboring disciplines. Economics is, in principle, shielded
from “lower” sciences, psychology and neuroscience included. In the context
of neoclassical economics, choice is assumed, not explained. But is it really
true that psychological discoveries cannot have any impact on economics?
From the standpoint of black-boxing, both stories capture something im-
portant regarding the relation between economics, psychology, and neu-
roscience. And both leave out a significant piece. Neoclassical economists
are right to stress autonomy. Economics is neither psychology nor
8 Some neuroeconomists, such as Glimcher (2011), are explicit and unapologetic about their re-
ductive endeavors. Other authors prefer to replace talk of “reduction” with the “integration” of psy-
chology, economics, and neuroscience (Craver and Alexandrova 2008). Setting aside differences
among these strategies, the central point is that it is important to reconcile the claims of economics
with the best theories and data from psychology and neuroscience, as these fields have the potential
to mutually inform each other.
9 Recall from section 4.4 of Chapter 4 that, following the deductive-nomological model, Friedman
treated prediction and explanation as two sides of the same coin.
neuro-science. These disciplines have different objectives and demand dif-

ferent abstractions. But completely insulating economics from these fields
is unreasonable. Psycho-neural economists correctly note that psychology
and neuroscience may inform economics in various ways. The challenge is
how to reconcile these seemingly incompatible claims. How can we make
psychology and neuroscience relevant to economics while maintaining the
methodological autonomy of these disciplines? Black boxes to the rescue.
The solution is to construct utility as a black box. At a general level of
description, utility can be used to frame the notion of economic behavior,
which brings the two theories together. Neoclassical and psycho-neural ec-
onomics both study Robbins’s “human behavior as a relationship between
ends and scarce means which have alternative uses” (Robbins 1932, p. 16).
But when we start unpacking this black box, we find very different frames
and difference-makers. Psychology and neuroscience do a fine job unrav-
eling the mechanisms underlying decision-making. Yet, in doing so, they ad-
dress very different explananda, incommensurable to economic ones, which
makes the choice between them “external.” The integration of psychology
and economics cannot be achieved by uncovering underlying mechanisms.
It rather depends on finding broader, overarching frameworks that construct
black boxes spanning both fields.

Antireductionism: Phlogiston
Before moving on, let me briefly address our final example: the case of phlo-
giston. This is another peculiar case study since, differently from the others,
it involves an entity that has been purged from our scientific ontology. This
raises a host of distinctive questions and challenges.
Recall how phlogiston theory purported to provide an explanation of the
process of combustion. What happens when a log is burned? What is the na-
ture of the reaction that turns wood into ashes? What accounts for changes
in mass, alterations to the surrounding air, and so forth? The keystone of
phlogiston theory is the postulation of an entity—phlogiston—that is sup-
posedly emitted in all cases of combustion. This concept, we now all know, is
not a natural kind. There is no substance that is emitted in all cases of com-
bustion. There is no such thing as phlogiston. Still, as we saw in Chapter 9,
the progress of the theory itself raises interesting conundrums. The question
274 Black Boxes
that I want to address here is this: is the transition from phlogiston theory to
atomic chemistry a reductionist or an antireductionist story?
From a radical reductionist perspective, the natural conclusion to draw is
that phlogiston has been eliminated or reduced to concepts and entities pos-
tulated by modern atomic chemistry. The causal role played by phlogiston
in the old theory is now performed, for the most part, by the oxidation of
organic compounds. The problem, in brief, is that this seems to throw out the
baby with the bath water. Sure, we can all agree that there is no phlogiston
out there in the world. Still, phlogiston theory did make strikingly accurate
predictions. How does a nonexistent substance explain anything at all? How
do we account for the partial success of the theory?
Traditional antireductionism provides a diametrically opposite perspec-
tive, according to which phlogiston theory has been rephrased in terms of
modern chemistry. This solves the issue of success, which undermines the
eliminativist story. If what Priestley and colleagues called “phlogiston” actu-
ally refers to, say, oxygen and other chemical elements, then there is no mys-
tery as to why appeals to phlogiston are partially explanatory. But we now face
trouble explaining why the theory came to be discarded at all. Furthermore,
if not even phlogiston counts as an instance of elimination, does any theoret-
ical entity in science ever come to be thrown out?
In short, uncompromising forms of both reductionism and antireduc-
tionism miss the mark. Case histories like this one require a more nuanced
and subtle middle ground that recognizes how parts of a theory have been
eliminated or reduced, whereas other parts have been integrated and
rephrased in contemporary terms. The question is: which parts and why?
On Kitcher’s view, theory-laden terms, such as “phlogiston,” have a com-
plex reference potential. The meaning and denotation of these expressions
is context-dependent: they may refer to a plurality of entities, processes, and
activities, as well as to nothing at all, depending on the circumstances and the
intentions of both the speaker and the surrounding linguistic community.
Chapter 9 argued that Kitcher’s notion of reference potential and my black-
boxing strategy mutually reinforce each other. When does “phlogiston” refer
to oxygen? When does it pick out something else? When does it flatly fail
to refer? The key to answering these questions is to note that theory-laden
expressions like “phlogiston” are embedded in a rich and complex theoretical
thicket. As such, their framing is crucial. This process singles out the presence
or absence of phlogiston as the crucial factor in the process of combustion.
This is the difference-making stage, which then requires the representation

of these causal factors within a model.
In brief, the intentions of speakers and the presuppositions of linguistic
communities surely play a crucial role in the determination of reference.
But understanding whether the term in question is employed as frame or
difference-maker, and how it is represented in the context of a model, is also
of paramount importance for determining when the term “phlogiston” picks
out a behavior in need of explanation, whatever is responsible for producing
the behavior in question, or when it refers to nothing at all.
In sum, the story of phlogiston combines aspects of reductionism with
aspects of antireductionism. Reductionists are correct that progress was
achieved by breaking down phlogiston and rephrasing its explanatory
successes in terms of atomic chemistry. At the same time, antireductionists
rightly stress the autonomy of phlogiston theory, whose explanatory
success—or lack thereof—should be assessed on its own grounds. Once
again, all of this goes to show that autonomy and reduction are perfectly con-
sistent. The reason is that what we compare and assess are theories, not indi-
vidual hypotheses.
§10.5. The Devil in the Details?
Time to take stock. There is a common misconception underlying both tradi-

tional reductionism and classic antireductionism. This is the assumption that
the same questions can be raised at different levels. On the contrary, inquiries
are framed differently and thus do not correspond across theories, fields, and
paradigms. In this sense, they are incommensurable. Physical descriptions do
not answer the same questions posed in chemistry, biology, or higher steps
in the hierarchy. These inquiries are context-dependent. As such, we cannot
compare them directly. What we compare are entire models, and the choice
between frameworks themselves is a Carnapian “external” question. The
bottom line is that theoretical progress is not a matter of reductionism vs.
antireductionism or decomposition vs. autonomy. It is both, and much more.
The key to the advancement of science is a matter of framing the right inqui-
ries, identifying the appropriate difference-makers, and constructing useful
models. In a nutshell, it is a matter of packing and unpacking black boxes.
What becomes of the old methodological dichotomy between Scylla and
Charybdis? As discussed at length throughout the book, the position of
276 Black Boxes
modest reductionists and sophisticated antireductionists overlap. Still, this

does not mean that there are no controversial issues or substantial debates
at play. But these disagreements are better captured in terms of black boxes,
as opposed to drawing a less-than-substantive contrast between reduction
and autonomy. Here is a suggestion. Many philosophers with reductionist
inclinations clearly acknowledge that black boxes play a significant role in
science. Yet, they contend that their function is merely a temporary con-
venience. Black boxes, from this perspective, are disposable scaffoldings. As
science marches on, these placeholders are gradually eliminated. The regu-
lative ideal of science is an unqualified theory of everything, expressed in
an ontologically fundamental physical language—whatever that may look
like—where each black box has been opened and replaced. Antireductionist
supporters can retort that boxes have a more permanent status: placeholders
are here to stay. From this standpoint, our best scientific models, including
our most fundamental physical descriptions of the universe, necessarily in-
clude black boxes, and other forms of abstraction and idealization. A com-
prehensive theory of everything purged of boxes and other higher-level
placeholders is neither achievable nor desirable. This contrast captures some
core aspects of the old debate between reductionists and antireductionists.
At the same time, in my humble opinion, such paraphrase in terms of the
status of black boxes in scientific practice is clearer and more perspicuous
than the old muddled question: can physics explain everything?
Before bidding farewell, I still have one last promise to fulfill. All the way
back in Chapter 1, I criticized the cumulative model of science, and the cor-
responding way in which science is taught, on the grounds that it leaves
no room for ignorance. The antireductionist pathway does not provide a
promising alternative. This book spelled out a middle-ground solution that,
I promised, would fare better on this score. Can I back this up? Does the
black-boxing strategy shed light on the productive role of ignorance without
giving up on the insights of reductionism?
All ignorance is equal, but some ignorance is more equal than other.
Setting Orwellian references aside, back in Chapter 1, I briefly intro-
duced Firestein’s distinction between two kinds of ignorance. On the
one hand, there is “willful stupidity,” a stubborn devotion to uninformed
opinion and other situations where “the ignorant are unaware, unenlight-
ened, uninformed, and surprisingly often occupy elected offices.” At the
same time, Firestein observes, there is a different, less-pejorative sense
of ignorance. This is knowledgeable ignorance, perceptive ignorance, in-

sightful ignorance.
I would like to further elaborate on this distinction, by suggesting that
there are not two, but three kinds of ignorance at play in science, as well
as in ordinary settings. First, there are situations which correspond to
Firestein’s “willful stupidity,” where someone stubbornly resists getting
acquainted with the relevant, known facts. Call this disruptive ignorance,
because this kind of ignorance has no positive role whatsoever to play,
in science as in everyday life. Now, let us focus on knowledgeable, per-
ceptive, insightful ignorance. It is important to appreciate that this idea
can be cashed out in two distinct ways. First, there are situations where
we cannot provide an adequate account of an event simply because we do
not know enough about the event in question. But realizing what we do
not know may provide some steppingstones toward acquiring the know-
ledge in question. Call this weakly productive ignorance. Contrary to dis-
ruptive ignorance, weakly productive ignorance is blameless. It is not that
we refuse to acquire the relevant information; the relevant information is
just not there. As Firestein notes, when properly harnessed, this kind of
ignorance may become a portal to knowledge, to success. It paves the way
toward more knowledge. In this sense, it is productive. There is, however, a
third, less obvious kind of productive ignorance, which I call strongly pro-
ductive ignorance. This is an instance where ignorance is not only blame-
less, harmless, and potentially conducive to knowledge, as in the case of
weakly productive ignorance. With strongly productive ignorance, the
state of ignorance may actually be preferable to the corresponding state of
knowledge. Ignorance actually becomes a goal in itself. In sum, we can all
agree that disruptive ignorance is bad and productive ignorance is good.
Yet, productive ignorance comes in two varieties, weak and strong.10
The reason why this matters is that weakly productive ignorance fits in
well with all extant models of science. Antireductionists and reductionists
alike should agree that what we do not know may become an incentive to do
better, to learn more, to accomplish what we are striving for. When I claimed,
back in Chapter 1, that the reductionist image of science as erecting a wall
10 These considerations, ça va sans dire, hardly exhaust the deep and extensive issue of ignorance.
Another interesting question concerns whether there is ignorance that is, in principle, insoluble. In
other words, are there questions that we do not currently know, and there is no way of knowing—
something akin to Emil du Bois Reymond’s ignoramus et ignorabimus mentioned in Chapter 1? Or
do all human questions have an answer, at least in principle? This is a fascinating debate, albeit one
tangential to the scope of this work.
278 Black Boxes
or tiling a mosaic leaves no room for ignorance, I had a different notion of

productivity in mind. This is what I now call “strongly productive igno-
rance.” From a cumulative, brick-by-brick perspective, the very idea that ig-
norance may be preferable to the corresponding state of knowledge makes
no sense. The bigger the wall, the better! Every time we add something we
previously did not know, we have made progress. This, however, comes at a
cost. Without strongly productive ignorance, it becomes hard to see how any
higher-level explanation can be truly autonomous.
It should come as little surprise that antireductionist approaches fare much
better on this score. By stressing autonomy, traditional antireductionism has
no problem recognizing the productive role of ignorance, which goes beyond
the truism that some explanations do not require further details to stand
alone. What makes this ignorance strongly productive, is that, by adding fur-
ther detail we may end up losing generality and muddying the waters. Yet, as
discussed at length, antireductionism suffers from its own shortcomings. It
has trouble accounting for why adding information does not invariably en-
rich an explanation. After all, reductionists retort, do we not make progress
when we replace ignorance—negative or productive—with knowledge? Do
we not learn something that may matter now or in the future? Is the very idea
of ignorance being preferable to the corresponding state of knowledge not an
oxymoron?
The problem with extant debates, as we saw, is that reductionists and
antireductionists alike tend to share a restrictive conception of the interface
between levels of explanation. The relation between microcosm and macro-
cosm is much richer, and more interesting, than it is standardly depicted.
The black-boxing strategy allows us to maintain the positive features of
both approaches, while avoiding the respective shortcomings. Ignorance
has a constructive, strongly productive role to play. It promotes autonomy,
as antireductionists point out. At the same time, opening a black box always
constitutes progress, as emphasized by reductionists. The key is noting that
this operation transforms the old inquiry into a new problem. Opening a
black box introduces new explananda without necessarily threatening the in-
tegrity or autonomy of old ones. Medawar’s notion of overtaking the forces of
ignorance can be maintained without sacrificing the true autonomy of higher
levels. Thus characterized, the very idea of a strongly productive ignorance is
not an oxymoron, after all. It plays an important role in research.
These considerations suggest some respects in which we can do better at
presenting science to the younger generations. The dusty image of science as
a slow, painstaking accumulation of truth has dominated the scene for the
better part of the past century, and it is still dominant in textbooks. The anti-
reductionist image of a dappled universe might not be the appropriate sub-
stitute. Black-boxing suggests a different image. How about recharacterizing
the scientific world as a set of matryoshka, that is, nested Russian dolls
(Figure 10.1)? This suggests that the relation between explanatory levels is
neither one of layer-cake-style progressive reduction nor one of complete
autonomy. The right metaphor is one of dynamic containment, where each
discipline constitutes a system in and of itself that, however, is constrained
by what goes on “outside” and, in turn, constrains what goes on “inside.” This
interplay, I believe, is an important message to convey to young students of
science, in books as in class, and the educated public in general. What goes
on at lower levels does, indeed, constrain what happens at higher floors. But
the converse is also true. Framing the right macro-explananda is instru-
mental for raising the appropriate micro-questions. The general lesson to be
learned—and taught—is that the advancement of science requires coopera-
tion, not elimination.
In conclusion, this book sets the foundations and explores the boundaries
of what is—hopefully—a fecund and exciting research project. What I have
done here barely scratches the surface of a deep, unfathomed ocean. Much
remains to be done to adequately study all the workings and implications of
ECON O M IC S
NEUROPSYCH
BIO L O G Y
PH YSIC S
Figure 10.1. The matryoshka model of science.

280 Black Boxes
black boxes and related concepts. If my presentation inspires others, espe-

cially younger scholars and students, by showing how important historical,
scientific, and philosophical questions acquire new significance when re-
cast and addressed in the context of this strategy, my goal will have been met
in full. After all, the devil might not always be in the details. But as Goethe
reminds us in the quote selected as the epigraph for this final chapter, it may
be hard to catch him anyway.
References
Achinstein, P. (1968). Concepts of Science. Baltimore, MD: Johns Hopkins University

Press.
Allen, G. E. (1975). Life Science in the Twentieth Century. Cambridge: Cambridge University
Press.
Allen, G. E. (1978). Thomas Hunt Morgan: The Man and His Science. Princeton, NJ:
Princeton University Press.
Amundson, R. (1983). “E.C. Tolman and the Intervening Variable: A Study in the
Epistemological History of Psychology.” Philosophy of Science 50, 268–282.
Amundson, R. (1986). “The Unknown Epistemology of E.C. Tolman.” British Journal of
Psychology 77, 525–531.
Anderson, J. (1938). “The Problem of Causality.” Australasian Journal of Psychology and
Philosophy 16, pp. 127–142.
Appiah, K. A. (2017). As If: Idealizations and Ideals. Cambridge, MA: Harvard University
Press.
Barnes, E. (2012). “Emergence and Fundamentality.” Mind 121(484), 873–901.
Batterman, R. (2002). The Devil in the Details: Asymptotic Reasoning in Explanation,
Reduction and Emergence. Oxford: Oxford University Press.
Bechtel, W. (2011). “Mechanism and Biological Explanation.” Philosophy of Science 78,
533–557.
Bechtel, W., and R. C. Richardson. (2010). Discovering Complexity: Decomposition and
Localization as Strategies in Scientific Research (2nd ed.). Cambridge, MA: MIT Press.
Bedau, M. A. (1997). “Weak Emergence.” In J. Tomberlin (Ed.), Philosophical
Perspectives: Mind, Causation, and World, Vol. 11, pp. 375–399. Malden, MA: Blackwell.
Bedau, M. A. (2002). “Downward Causation and Autonomy of Weak Emergence.”
Principia 6, 5–50.
Bedau, M. A. (2012). “Weak Emergence and Computer Simulation.” In P. Humphreys
and C. Imbert (Eds.), Models, Simulations, and Representations, pp. 91– 114.
New York: Routledge.
Behe, M. J. (1996). Darwin’s Black Box: The Biochemical Challenge to Evolution. New York:
Free Press.
Benacerraf, P. (1965). “What Numbers Could Not Be.” Philosophical Review 74, 47–73.
Bickle, J. (1998). Psychoneural Reduction: The New Wave. Cambridge, MA: MIT Press.
Bickle, J. (2003). Philosophy and Neuroscience: A Ruthlessly Reductive Account. Dordrecht:
Kluwer.
Bogen, J. (2008). “Causally Productive Activities.” Studies in History and Philosophy of
Science 39(1), 112–123.
Boniolo, G. (2013). “On Molecular Mechanisms and Contexts of Physical Explanation.”
Biological Theory 7(3), 256–265.
Boring, E. G. (1950). A History of Developmental Psychology (2nd ed.). New York:
Appleton-Century-Crofts.
282 References
Box, G. E. (1976). “Science and Statistics.” Journal of the American Statistical Association
71(356), 791–799.
Burge, T. (2013). “Modest Dualism.” In Cognition through Understanding: Philosophical
Essays, Vol. 3, pp. 471–488. Oxford: Oxford University Press.
Camerer, C. F. (2010). “The Case for Mindful Economics.” In A. Caplin and A. Schotter
(Eds.), The Foundations of Positive and Normative Economics, pp. 43– 69.
New York: Oxford University Press.
Camerer, C. F., G. Loewenstein, and D. Prelec. (2005). “Neuroeconomics: How
Neuroscience Can Inform Economics.” Journal of Economic Literature 43, 9–64.
Carey, S. (1999 [1991]). “Knowledge Acquisition: Enrichment or Conceptual Change?”
In E. Margolis and S. Laurence (Eds.), Concepts: Core Readings, pp. 459– 487.
Cambridge: Bradford.
Carey, S. (2009). The Origin of Concepts. New York: Oxford University Press.
Carnap, R. (1938). “Logical Foundations of the Unity of Science.” In O. Neurath, R.
Carnap, and C. Morris (Eds.), International Encyclopedia of Unified Science, pp. 42–62.
Chicago: University of Chicago Press.
Carnap, R. (1956a). “Empiricism, Semantics, and Ontology.” In Carnap, Meaning and
Necessity (2nd ed.), pp. 205–221. Chicago: University of Chicago Press.
Carnap, R. (1956b). Meaning and Necessity (2nd ed.). Chicago: University of Chicago Press.
Carroll, S. B. (2005). Endless Forms Most Beautiful: The New Science of Evo Devo.
New York: Norton.
Cartwright, N. (1980). “The Truth Doesn’t Explain Much.” American Philosophical
Quarterly 17(2), 159–163.
Cartwright, N. (1983). How the Laws of Physics Lie. Oxford: Clarendon.
Cartwright, N. (1999). The Dappled World: A Study of the Boundaries of Science. Cambridge:
Cambridge University Press.
Chalmers, D. J. (2006). “Strong and Weak Emergence.” In P. Davies and P. Clayton (Eds.),
The Re-Emergence of Emergence, pp. 244–254. New York: Oxford University Press.
Chalmers, D. J. (2012). Constructing the World. Oxford: Oxford University Press.
Chalmers, D. J., D. Manley, and D. Wasserman (Eds.) (2009). Metametaphysics: New
Essays in the Foundations of Ontology. Oxford: Oxford University Press.
Chomsky, N. (1959). “Review of Verbal Behavior by B.F. Skinner.” Language 35(1), 26–58.
Chomsky, N. (2012). The Science of Language: Interviews with James McGilvray. New York:
Churchland, P. (1986). Neurophilosophy. Cambridge, MA: MIT Press.
Churchland, P. M. (1979). Scientific Realism and the Plasticity of Mind.
Cambridge: Cambridge University Press.
Craver, C. F. (2007). Explaining the Brain: Mechanisms and the Mosaic Unity of Neuroscience.
Craver, C. F., and A. Alexandrova. (2008). “No Revolution Necessary: Neural Mechanisms
for Economics.” Economics and Philosophy 24, 381–406.
Craver, C. F., and L. Darden. (2013). In Search of Mechanisms: Discoveries across the Life
Sciences. Chicago: University of Chicago Press.
Craver, C. F., and D. M. Kaplan. (2020). “Are More Details Better? On the Norms of
Completeness for Mechanistic Explanation.” British Journal for the Philosophy of
Science 71(1), 287–319.
Crick, F.H. (1994). The Astonishing Hypothesis: The Scientific Search for the Soul,
New York: Scribners.
References 283
Culp, S., and P. Kitcher. (1989). “Theory Structure and Theory Change in Contemporary
Molecular Biology.” British Journal for the Philosophy of Science 40, 459–483.
Darden, L. (1991). Theory Change in Science: Strategies from Mendelian Genetics.
Oxford: Oxford University Press.
Darden, L., and N. Maull. (1977). “Interfield Theories.” Philosophy of Science 44, 43–64.
Darwin, C. (1859). On the Origin of Species (2008 ed.). New York: Oxford University Press.
Davidson, D. (1970). “Mental Events.” In L. Foster and J. Swanson (Eds.), Experience and
Theory, pp. 79–101. London: Duckworth.
Davidson, D. (1974). “On the Very Idea of a Conceptual Scheme.” Proceedings and
Addresses of the American Philosophical Association 47, 183–198.
DeMartino, G. F. (2000). Global Economy, Global Justice: Theoretical Objections and Policy
Alternatives to Neoliberalism. New York: Routledge.
Dennett, D. C. (1981). “Skinner Skinned.” In Dennett, Brainstorms: Philosophical Essays
on Mind and Psychology, pp. 53–70. Cambridge, MA: Bradford, MIT Press.
Dennett, D. C. (1987). The Intentional Stance. Cambridge, MA: Bradford, MIT Press.
Dennett, D. C. (1991). Consciousness Explained. Boston: Little, Brown.
Dizadji-Bahmani, F., R. Frigg, and S. Hartmann. (2010). “Who’s Afraid of Nagelian
Reduction?” Erkenntnis 73, 393–412.
Donnellan, K. (1970). “Proper Names and Identifying Descriptions.” Synthese 21,
335–358.
Donnellan, K. (1974). “Speaking of Nothing.” Philosophical Review 83, 3–31.
Dretske, F. (1973). “Contrastive Statements.” Philosophical Review 82, 411–437.
Duhem, P. M. (1954). The Aim and Structure of Physical Theory. Princeton, NJ: Princeton
University Press.
Dupré, J. (1993). The Disorder of Things. Cambridge, MA: Harvard University Press.
Dupré, J. (2012). Processes of Life: Essays in the Philosophy of Biology. New York: Oxford
University Press.
Falk, R. (2009). Genetic Analysis: A History of Genetic Thinking. Cambridge: Cambridge
University Press.
Fazekas, P. (2009). “Reconsidering the Role of Bridge Laws in Inter-Theoretic Relations.”
Erkenntnis 71, 303–322.
Feyerabend, P. K. (1993 [1975]). Against Method (3rd ed.). London and New York: Verso.
Firestein, S. (2012). Ignorance: How It Drives Science. New York: Oxford University Press.
Flanagan, O. (1991). The Science of the Mind (2nd ed.). Cambridge, MA: MIT Press.
Fodor, J. (1974). “Special Sciences (Or: The Disunity of Science as a Working Hypothesis).”
Synthese 28, 97–115.
Fodor, J. A. (1999). “Let Your Brain Alone.” London Review of Books 21.
Franklin-Hall, L. R. (2008). From a Microbiological Point of View. Ph.D. thesis, Columbia
University, New York.
Franklin-Hall, L. R. (2016). “New Mechanistic Explanation and the Need for Explanatory
Constraints.” In K. Aizawa and C. Gillett (Eds.), Scientific Composition and Metaphysical
Ground: New Directions in the Philosophy of Science, pp. 41–74. London: Palgrave
MacMillan.
Franklin- Hall, L. R. (forthcoming). “The Causal Economy Account of Scientific
Explanation.” Minnesota Studies in the Philosophy of Science.
Frege, G. (1892). “On Sinn and Bedeutung.” In M. Beaney (Ed.), The Frege Reader, pp.
251–271. Oxford: Blackwell.
284 References
Friedman, M. (1953). “The Methodology of Positive Economics.” In Friedman, Essays in

Positive Economics, pp. 3–43. Chicago and London: University of Chicago Press.
Friedman, M. (2001). Dynamics of Reason. Stanford, CA: CSLI Publications.
Friedman, M., and L. J. Savage. (1948). “The Utility Analysis of Choices Involving Risk.”
Journal of Political Economy 56(4), 279–304.
Garfinkel, A. (1981). Forms of Explanation. New Haven, CT: Yale University Press.
Giere, R. N. (1988). Explaining Science: A Cognitive Approach. Chicago: University of
Chicago Press.
Gigerenzer, G. (2007). Gut Feelings: The Intelligence of the Unconscious. New York: Viking
Penguin.
Gigerenzer, G. (2008). Rationality for Mortals: How People Cope with Uncertainty.
Gillett, C. (2016). Reduction and Emergence in Science and Philosophy. Cambridge:
Glennan, S. (2016). “Mechanisms and Mechanical Philosophy.” In P. Humphreys (Ed.),
The Oxford Handbook of Philosophy of Science, pp. 798–816. Oxford University Press.
Glennan, S. (2017). The New Mechanical Philosophy. Oxford: Oxford University Press.
Glimcher, P. W. (2011). Foundations of Neuroeconomic Analysis. New York: Oxford
University Press.
Godfrey-Smith, P. (2003). Theory and Reality: An Introduction to the Philosophy of Science.
Godfrey-Smith, P. (2014). Philosophy of Biology. Princeton, NJ, and Oxford: Princeton
University Press.
Goodman, N. (1955). Fact, Fiction, and Forecast. Cambridge, MA: Harvard University
Press.
Gould, S. J. (1977). Ontogeny and Phylogeny. Cambridge, MA: Belknap.
Grandy, R. (1973). “Reference, Meaning, and Belief.” Journal of Philosophy 70, 439–452.
Grene, M., and D. Depew. (2004). The Philosophy of Biology: An Episodic History.
Cambridge: Cambridge University Press.
Griffiths, P., and K. Stotz. (2013). Genetics and Philosophy: An Introduction. Cambridge:
Gul, F., and W. Pesendorfer. (2008). “The Case for Mindless Economics.” In A. Caplin
and A. Schotter (Eds.), The Foundations of Positive and Normative Economics, pp. 3–39.
Hacking, I. (1983). Representing and Intervening: Introductory Topics in the Philosophy of
Natural Science. Cambridge: Cambridge University Press.
Hall, R., and C. Hitch. (1939). “Price Theory and Business Behavior.” Oxford Economic
Papers 2, 12–45.
Hanson, N. R. (1963). The Concept of the Positron: A Philosophical Analysis. Cambridge:
Harrison, G. W. (2008). “Neuroeconomics: A Critical Reconsideration.” Economics and
Philosophy 24, 303–344.
Hatfield, G. (2002). “Psychology, Philosophy, and Cognitive Science: Reflections on
the History and Philosophy of Experimental Psychology.” Mind and Language 17(3),
207–232.
Hausman, D. (1992). The Inexact and Separate Science of Economics. Cambridge:
References 285
Hempel, C. G. (1945, January). “Studies in the Logic of Confirmation (I).” Mind: A

Quarterly Review of Psychology and Philosophy 54(213), 1–26.
Hempel, C. G. (1965). Aspects of Scientific Explanation and Other Essays in the Philosophy
of Science. New York: Free Press.
Hempel, C. G. (1966). Philosophy of Natural Science. Englewood Cliffs, NJ: Prentice-Hall.
Hempel, C. G., and P. Oppenheim. (1965). “On the Idea of Emergence.” In C. G. Hempel
(Ed.), Aspects of Scientific Explanation and Other Essays in the Philosophy of Science, pp.
258–264. New York: Free Press.
Hooker, C. A. (1981). “Towards a General Theory of Reduction. Part III: Cross-
Categorical Reductions.” Dialogue 20, 496–529.
Hull, D. L. (1974). Philosophy of Biological Science. Englewood Cliffs, NJ: Prentice-Hall.
Humphreys, P. (2016). Emergence: A Philosophical Account. New York: Oxford
University Press.
Hutchison, T. W. (1938). The Significance and Basic Postulates of Economic Theory. London:
Macmillan.
Hütteman, A., and A. C. Love. (2016). “Reduction.” In P. Humphreys (Ed.), The Oxford
Handbook of Philosophy of Science, pp. 460–484. New York: Oxford University Press.
Huxley, J. (1942). Evolution: The Modern Synthesis. London: Allen and Unwin.
Kahneman, D. (2011). Thinking, Fast and Slow. New York: Farrar, Straus and Giroux.
Kaplan, D. M., and C. F. Craver. (2011). “The Explanatory Force of Dynamical and
Mathematical Models in Neuroscience: A Mechanistic Perspective.” Philosophy of
Science 78, 601–627.
Kim, J. (1999). Mind in a Physical World. Cambridge, MA: MIT Press.
Kim, J. (2005). Physicalism, or Something Near Enough. Princeton, NJ: Princeton University
Press.
Kincaid, H. (1990). “Molecular Biology and the Unity of Science.” Philosophy of Science
57, 575–593.
Kitcher, P. (1978). “Theories, Theorists, and Theoretical Change.” The Philosophical
Review 87(4), 519–547.
Kitcher, P. (1982). “Genes.” British Journal for the Philosophy of Science 33, 337–359.
Kitcher, P. (1984). “1953 and All That: A Tale of Two Sciences.” The Philosophical Review
96, 335–373.
Kitcher, P. (1985). “Darwin’s Achievement.” In N. Rescher (Ed.), Reason and Rationality in
Natural Science, pp. 127–189. Lanham, MD: University Press of America.
Kitcher, P. (1993). The Advancement of Science. New York: Oxford University Press.
Kitcher, P. (1999). “The Hegemony of Molecular Biology.” Biology and Philosophy 14,
195–210.
Kitcher, P., and W. C. Salmon. (1987). “Van Fraassen on Explanation.” The Journal of
Philosophy 84(6), 315–330.
Klein, C. (2009). “Reduction Without Reductionism: A Defense of Nagel on
Connectability.” The Philosophical Quarterly 59(234), 39–53.
Kordig, C. R. (1971). The Justification of Scientific Change. Dordrecht: Reidel.
Kripke, S. A. (1972). Naming and Necessity. Cambridge, MA: Harvard University Press.
Kuhn, T. S. (1962). The Structure of Scientific Revolutions (1st ed.). Chicago: University of
Chicago Press.
Kuhn, T. S. (1977). “Objectivity, Value Judgment, and Theory Choice.” In Kuhn, The
Essential Tension: Selected Studies in Scientific Tradition and Change, pp. 320–339.
286 References
Kuhn, T. S. (1982). “Commensurability, Comparability, Communicability.” In P. Asquith

and T. Nickles (Eds.), Proceedings of the 1982 Biennial Meeting of the Philosophy of
Science Association, pp. 669–688. East Lansing, MI: Philosophy of Science Association.
Lakatos, I. (1970). “Falsification and the Methodology of Scientific Research Programmes.”
In I. Lakatos and A. Musgrave (Eds.), Criticism and the Growth of Knowledge, pp. 91–
195. Cambridge: Cambridge University Press.
Lakatos, I. (1976). Proofs and Refutations. Cambridge: Cambridge University Press.
Laudan, L. (1977). Progress and Its Problems: Toward a Theory of Scientific Growth.
Berkeley and Los Angeles: University of California Press.
Leahey, T. H. (1980). A History of Psychology. Englewood Cliffs, NJ: Prentice-Hall.
Leahey, T. H. (2018). A History of Psychology: From Antiquity to Modernity (8th ed.).
New York: Routledge.
Lester, R. A. (1946). “Shortcomings of Marginal Analysis for Wage-Unemployment
Problems.” American Economic Review 36, 63–82.
Lewis, D. K. (1972). “Psychophysical and Theoretical Identifications.” Australasian Journal
of Philosophy 50, 249–258.
Lewis, D. K. (1986). “Causal Explanation.” In Lewis, Philosophical Papers, Vol. II, pp. 214–
240. New York: Oxford University Press.
Lloyd, E. A. (1988). The Structure and Confirmation of Evolutionary Theory. Princeton,
NJ: Princeton University Press.
Love, A. C., and M. J. Nathan. (2015). “The Idealization of Causation in Mechanistic
Explanation.” Philosophy of Science 82, 761–774.
Machamer, P. K. (2004). “Activities and Causation: The Metaphysics and Epistemology of
Mechanisms.” International Studies in the Philosophy of Science 18, 27–39.
Mackie, J. L. (1974). The Cement of the Universe. Oxford: Oxford University Press.
Maull, N. L. (1977). “Unifying Science without Reduction.” Studies in History and
Philosophy of Science 8, 143–162.
Mayr, E. (1982). The Growth of Biological Thought. Cambridge, MA, and London: Belknap
Harvard.
Mayr, E. (1988). “The Challenge of Darwinism.” In Mayr, Toward a New Philosophy of
Biology. Observations of an Evolutionist, pp. 185–195. Cambridge, MA: Harvard
University Press.
Mayr, E. (1991). One Long Argument: Charles Darwin and the Genesis of Modern
Evolutionary Thought. Cambridge, MA: Harvard University Press.
Mayr, E. (2004). What Makes Biology Unique? Cambridge: Cambridge University Press.
Mayr, E., and W. B. Provine. (1980). The Evolutionary Synthesis. Cambridge, MA: Harvard
University Press.
McCabe, K. A. (2008). “Neuroconomics and the Economic Sciences.” Economics and
McLaughlin, B. (1992). “The Rise and Fall of British Emergentism.” In A. Beckermann,
J. Kim, and H. Flohr (Eds.), Emergence or Reduction?, pp. 49–93. Berlin: De Gruyter.
Medawar, P. B. (1969). The Art of the Soluble. Harmondsworth, UK: Penguin.
Mills, S. K., and J. H. Beatty. (1979). “The Propensity Interpretation of Fitness.” Philosophy
of Science 46, 263–286.
Morange, M. (1998). A History of Molecular Biology. Cambridge, MA: Harvard
University Press.
Morgan, M., and M. Morrison (Eds.) (1999). Models as Mediators: Perspectives on the
Natural and Social Sciences. Cambridge: Cambridge University Press.
References 287
Morrison, M. (2012). “Emergent Physics and Micro-Ontology.” Philosophy of Science 79,

141–166.
Nagel, E. (1961). The Structure of Science. New York: Harcourt Brace.
Nagel, E. (1963). “Assumptions in Economic Theory.” The American Economic Review
53(2), 211–219.
Nagel, T. (2012). Mind and Cosmos: Why the Materialist Neo-Darwinian Conception of
Nature Is Almost Certainly False. New York: Oxford University Press.
Nathan, M. J. (2012). “The Varieties of Molecular Explanation.” Philosophy of Science
79(2), 233–254.
Nathan, M. J. (2014). “Causation by Concentration.” British Journal for the Philosophy of
Science 65, 191–212.
Nathan, M. J. (2015a). “A Simulacrum Account of Dispositional Properties.” Nôus 49(2),
253–274.
Nathan, M. J. (2015b). “Unificatory Explanation.” British Journal for the Philosophy of
Science 68, 163–186.
Nathan, M. J., and G. Del Pinal. (2016). “Mapping the Mind: Bridge Laws and the Psycho-
Neural Interface.” Synthese 193(2), 637–657.
Nathan, M. J., and G. Del Pinal. (2017). “The Future of Cognitive Neuroscience? Reverse
Inference in Focus.” Philosophy Compass 12(7), 1–11.
Nicholson, D. J. (2012). “The Concept of Mechanism in Biology.” Studies in History and
Philosophy of Biological and Biomedical Sciences 43, 152–163.
Oppenheim, P., and H. Putnam. (1958). “The Unity of Science as a Working Hypothesis.”
In H. Feigl, M. Scriven, and G. Maxwell (Eds.), Minnesota Studies in the Philosophy of
Science, Vol. 2, pp. 3–36. Minneapolis: Minnesota University Press.
Orzack, S. H. (2008). “Testing Adaptive Hypotheses, Optimality Models, and
Adaptationism.” In M. Ruse (Ed.), The Oxford Handbook of Philosophy of Biology, pp.
87–112. New York: Oxford University Press.
Piccinini, G., and C. F. Craver. (2011). “Integrating Psychology and Neuroscience:
Functional Analyses and Mechanism Sketches.” Synthese 183, 283–311.
Pinker, S. (1997). How the Mind Works. New York: Norton.
Putnam, H. (1973). “Meaning and Reference.” Journal of Philosophy 70, 699–711.
Putnam, H. (1975a). “Philosophy and Our Mental Life.” In Putnam, Mind, Language, and
Reality, pp. 291–303. New York: Cambridge University Press.
Putnam, H. (1975b). “The Meaning of Meaning.” In Putnam, Mind, Language, and Reality,
pp. 215–271. Cambridge: Cambridge University Press.
Putnam, H. (1990). Realism with a Human Face. Cambridge, MA: Harvard University
Press.
Quine, W. V. O. (1953). From A Logical Point of View. Cambridge, MA: Harvard University
Press.
Quine, W. V. O. (1970). “On the Reasons for Indeterminacy of Translation.” The Journal of
Railton, P. (1981). “Probability, Explanation, and Information.” Synthese 48(2), 233–256.
Robbins, L. (1932 [1984]). An Essay on the Nature and Significance of Economic Science
(3rd ed.). London: Macmillan.
Rorty, R. (1979). Philosophy and the Mirror of Nature. Princeton, NJ: Princeton University
Press.
Rosenberg, A. (1983). “Fitness.” Journal of Philosophy 80, 437–447.
288 References
Rosenberg, A. (1985). The Structure of Biological Science. Cambridge: Cambridge

University Press.
Rosenberg, A. (1992). Economics–Mathematical Politics or Science of Diminishing Returns?
Rosenberg, A. (2006). Darwinian Reductionism: Or How to Stop Worrying and Love
Molecular Biology. Chicago: University of Chicago Press.
Russell, B. (1913). “On the Notion of Cause.” Proceedings of the Aristotelian Society
13, 1–26.
Russell, B. (1927). The Analysis of Matter. London: Allen and Unwin.
Salmon, W. C. (1984). Scientific Explanation and the Causal Structure of the World.
Princeton, NJ: Princeton University Press.
Sarkar, S. (1998). Genetics and Reductionism. Cambridge: Cambridge University Press.
Schaffner, K. F. (1967). “Approaches to Reduction.” Philosophy of Science 34, 137–147.
Schaffner, K. F. (1993). Discovery and Explanation in Biology and Medicine. Chicago:
University of Chicago Press.
Schaffner, K. F. (2006). “Reduction: The Cheshire Cat Problem and a Return to Roots.”
Synthese 151, 377–402.
Scheffler, I. (1967). Science and Subjectivity. Indianapolis: Bobbs-Merrill.
Sen, A. K. (1977). “Rational Fools: A Critique of the Behavioral Foundation of Economic
Theory.” Philosophy and Public Affairs 6(4), 317–344.
Shapere, D. (1966). “Meaning and Scientific Change.” In R. Colodny (Ed.), Mind
and Cosmos: Essays in Contemporary Science and Philosophy, pp. 41– 85.
Pittsburgh: University of Pittsburgh Press.
Shapere, D. (1974). “Scientific Theories and Their Domains.” In F. Suppe (Ed.), The
Structure of Scientific Theories. Urbana: University of Illinois Press.
Silberstein, M., and A. Chemero. (2013). “Constraints on Localization and Decomposition
as Explanatory Strategies in the Biological Sciences.” Philosophy of Science 80(5),
958–970.
Simon, H. (1963). “Problems of Methodology: Discussion.” American Economic Review:
Papers and Proceedings 53, 229–231.
Skinner, B. F. (1956). “Critique of Psychoanalytic Concepts and Theories.” Minnesota
Studies in the Philosophy of Science 1, 77–87.
Skinner, B. F. (1957). Verbal Behavior. New York: Appleton-Century-Crofts.
Skinner, B. F. (1974). About Behaviorism. (1976 ed.). New York: Vintage Books.
Skipper, R. A., and R. L. Millstein. (2005). “Thinking about Evolutionary Mechanisms:
Natural Selection.” Studies in History and Philosophy of Biological and Biomedical
Sciences 36, 327–347.
Sklar, L. (1993). Physics and Chance: Philosophical Issues in the Foundations of Statistical
Mechanics. Cambridge: Cambridge University Press.
Smith, L. D. (1986). Behaviorism and Logical Positivism: A Reassessment of the Alliance.
Stanford, CA: Stanford University Press.
Sober, E. (1984). The Nature of Selection: Evolutionary Theory in Philosophical Focus.
Cambridge, MA: Bradford, MIT Press.
Sober, E. (1999). “The Multiple Realizability Argument against Reductionism.” Philosophy
of Science 66, 542–564.
Sober, E. (2000). Philosophy of Biology (2nd ed.). Boulder, CO: Westview.
Sober, E. (2015). Ockham’s Razor: A User’s Manual. Cambridge: Cambridge University
Press.
References 289
Sporns, O. (2011). Networks of the Brain. Cambridge, MA: MIT Press.

Sporns, O. (2012). Discovering the Human Connectome. Cambridge, MA: MIT Press.
Sporns, O. (2015). “Network Neuroscience.” In G. Marcus and J. Freeman (Eds.), The
Future of the Brain: Essays by the World’s Leading Neuroscientists. Princeton, NJ:
Princeton University Press.
Strevens, M. (2008). Depth: An Account of Scientific Explanation. Cambridge, MA:
Harvard University Press.
Strevens, M. (2016). “Special-Science Autonomy and the Division of Labor.” In M. Couch
and J. Pfeifer (Eds.), The Philosophy of Philip Kitcher, pp. 153–181. New York: Oxford
University Press.
Suppes, P. (1960). “A Comparison of the Meaning and Uses of Models in Mathematics
and the Empirical Sciences.” Synthese 12, 287–301.
Taylor, E. (2015). “An Explication of Emergence.” Philosophical Studies 172, 653–669.
van Fraassen, B. C. (1980). The Scientific Image. Oxford: Clarendon.
van Fraassen, B. C. (2008). Scientific Representation. New York: Oxford University Press.
van Strien, M. (2014). “On the Origins and Foundations of Laplacian Determinism.”
Studies in History and Philosophy of Science 45, 24–31.
Waters, C. K. (1990). “Why the Anti-Reductionist Consensus Won’t Survive the Case
of Classical Mendelian Genetics.” Proceedings to the 1990 Biennial Meeting of the
Philosophy of Science Association, pp. 125–139. East Lansing, MI: Philosophy of Science
Association.
Waters, C. K. (2004). “What Was Classical Genetics?” Studies in History and Philosophy of
Science 35, 783–809.
Waters, C. K. (2006). “A Pluralist Interpretation of Gene-Centered Biology.” In S.
Kellert, H. Longino, and C. K. Waters (Eds.), Scientific Pluralism. Minnesota
Studies in Philosophy of Science, Vol. XIX, pp. 190–214. Minneapolis: University of
Minnesota Press.
Waters, C. K. (2007). “Causes That Make a Difference.” The Journal of Philosophy 104(11),
551–579.
Watson, J. B. (1913). “Psychology as the Behaviorist Views It.” Psychological Review 20,
158–177.
Weber, M. (2005). Philosophy of Experimental Biology. Cambridge: Cambridge University
Press.
Weed, D. L. (1998). “Beyond Black Box Epidemiology.” American Journal of Public Health
88(1), 12–14.
Weinberg, S. (1992). Dreams of a Final Theory. New York: Pantheon.
Weisberg, M. (2013). Simulation and Similarity: Using Models to Understand the World.
Wilson, J. (2021). Metaphysical Emergence. Oxford: Oxford University Press.
Wimsatt, W. C. (2007). Re-Engineering Philosophy for Limited Beings. Cambridge,
MA: Harvard University Press.
Wittgenstein, L. (2001). Wittgenstein’s Lectures, Cambridge 1932–1935, A. Ambrose (Ed.),
New York: Prometheus.
Wright, C., and D. Van Eck. (2018). “Ontic Explanation Is Either Ontic or Explanatory,
but Not Both.” Ergo 5(38), 997–1029.
Zador, A. (2015). “The Connectome as a DNA Sequencing Problem.” In G. Marcus and J.
Freeman (Eds.), The Future of the Brain: Essays by the World’s Leading Neuroscientists,
pp. 40–49. Princeton, NJ: Princeton University Press.
Index
For the benefit of digital users, indexed terms that span two pages (e.g., 52–53) may, on
occasion, appear on only one of those pages.
abstraction, 15, 26, 70–72, 75, 109–10, 116, Bickle, John, 37n.16, 40–41n.22,
120–34, 136–37, 170–76, 187–90, 218n.3, 261n.7
202–11, 236, 243–45, 258, 272– biochemistry. See chemistry
73, 275–76 biogeography, 138–40
Achinstein, Peter, 224n.9 biology, 7–8, 16–17, 34–46, 50–61, 66, 77–
advancement of science. See scientific 80, 84–88, 91–93, 115–16, 138–49,
progress 172–74, 238–43, 256, 263–68
Alexander, Samuel, 192 biometrics, 148–49
Alexandrova, Anna, 272n.8 Bogen, Jim, 168–69
Allais, Maurice, 72–73 Bohr, Niels, 226
allele. See gene Boniolo, Giovanni, 186n.16
Allen, Garland E., 144n.4 Boring, Edwin G., 62n.10
Amundson, Ron, 63n.11 Box, George E.P., 186
Anderson, John, 125n.10 bridge laws, 26–29, 35, 36–37, 45–
antireductionism, 6–12, 15–16, 19–22, 46, 253
24–48, 49, 82–83, 99, 107–8, 163, bridge principles. See bridge laws
191–92, 201, 206–7, 209, 213–14, Broad, Charlie Dunbar, 192
248–49, 251–79 Bromberger, Sylvain, 112n.1
Appiah, Kwame Anthony, 71–72n.20 Burge, Tyler, 9n.5
Aristotle, 73–74, 192
autonomy. See epistemic autonomy Camerer, Colin, 74, 75–76, 76n.27
capacity. See disposition
Barnes, Elizabeth, 196n.8, 211–12 Carey, Susan, 233n.20
Bateson, William, 144, 146–47 Carnap, Rudolf, 22n.8, 28n.6, 62–63, 216,
Batterman, Robert. 107n.9, 212–13 218–19, 223n.8, 260–62, 265
Beadle, George W., 241, 242–43 Carroll, Sean, 58–59, 60–61
Beatty John, 86, 89, 91 Cartesian dualism. See dualism
Becher, Johann J., 103–4 Cartwright, Nancy, 28n.5, 32n.12,
Bechtel, William, 164, 168–69, 170, 175– 127n.12, 131n.18, 174–75, 190, 191
76, 185, 258 causal explanation. See scientific
Bedau, Mark, 196, 200, 209–10 explanation
behaviorism. See psychology causal mechanism. See mechanism
Behe, Michael, 143n.3 causal power, 90–91, 97, 106–7, 182–84,
Benacerraf, Paul, 260n.6 194–96, 198–203, 211–13
Benzer, Seymour, 241 Cavendish, Henry, 103–4, 227, 236
Bernoulli, Daniel, 157 Chalmers, David, 9n.5, 196n.8, 261n.7
Bernoulli, Nicolaus, 157n.12 Chemero, Anthony, 185
292 Index
chemistry, 62, 105, 199, 232–35, 273–75 determinism, 8–9, 268

biochemistry, 30, 39–40, 95–96, 202–4, development. See ontogeny
255–56, 267–68 Developmental Synthesis, 16–17, 59–61,
choice. See economic choice 66–67, 77–79, 82–83
Chomsky, Noam, 64–65, 66–67, 71– difference-maker, 17–18, 44–46, 83,
72n.20, 154n.11 94–108, 109–10, 121–26, 132–35,
Churchland, Patricia, 40–41n.22, 269 136–37, 143–61, 166–89, 203–7, 211,
Churchland, Paul M., 40–41n.22, 269 236–48, 252, 263, 273–75
combustion, 103–5, 227, 234–37, 273–75 difference-making stage, 18, 109–10, 118–
competition, 50, 143, 160, 244, 263–65 26, 130–31, 132–35, 136–37, 141,
principle of competition, 50–51 147, 162, 167–71, 175n.8, 181–82,
conceptual relativity. See 189, 237–46, 252, 274–75
incommensurability Darwin’s difference-making stage, 140–42
connectome, 199 Friedman’s difference-making
consciousness, 13–14, 61–62, 63–64, 78, stage, 156–59
99, 149–50, 152–53, 194, 204–6, Mendel’s difference-making stage, 147
208, 211 Skinner’s difference-making
Conway’s game of life, 196 stage, 150–51
Copernicus, Nicolaus, 222–23 disposition, 17–18, 43–44, 83–84, 86,
covering-law explanation. See deductive- 88–94, 96–100, 103–6, 203–4,
nomological explanation 234, 266–70
Craver, Carl, 122n.8, 164–65, 166–69, 170, dispositional property. See disposition
175–76, 179–85, 258n.5, 272n.8 Dizadji-Bahmani, Foad, 31n.11
Crick, Francis, 242–43, 247–48, 269 Dobzhansky, Theodosius, 57
critical state, 123, 198–200, 211 Donnellan, Keith, 228nn.15–16
Culp, Sylvia, 44n.26 dormitive virtue. See virtus dormitiva
cumulative model of science. See scientific drepanocytosis. See sickle-cell anemia
progress Dretske, Fred, 111–12, 113–14,
117, 131–32
Dante, 23 Drosophila, 44, 57, 59–60, 85–91, 123, 144,
Darden, Lindley, 116n.5, 144n.4, 149n.7, 241, 263–65
164–65, 166–68, 179 dualism, 38, 99
Darwin, Charles, 16–19, 50–61, 65–67, Cartesian substance dualism, 63–64,
77–78, 79–80, 82–83, 84, 88, 98, 124, 151, 210
135, 138–44, 147–49, 159–61, 162– property dualism, 38, 269
63, 179–80, 237–40, 243–44, 247–48, Du Bois Reymond, Emil, 13–14, 277n.10
252, 263–65 Duhem, Pierre M. M., 222–23, 259
Davidson, Donald, 93–94, 224n.9 Dupré, John, 28n.5, 32n.12, 144n.4
deductive-nomological explanation , 102,
112, 165–66, 174–75, 272n.9 economic choice, 68, 72–76, 78–79, 101–3,
Del Pinal, Guillermo, 26n.2, 155–60, 208, 244–46, 271–73
38n.17, 255n.2 rational choice theory, 72–73, 155–59, 271
De Vries, Hugo, 147 risky vs. risk-less choice , 155–
DeMartino, George, 72n.21, 130n.14 60, 244–45
Democritus, 226 economics, 7–8, 67–76, 77–80, 82–83,
Dennett, Daniel, 9n.5, 13–14, 151, 154n.11 100–3, 154–61, 205, 208, 244–46,
Depew, David, 140–41 252, 256, 271–73
Descartes, René, 40, 63–64, 98, 151–53, neoclassical economics, 16–19, 70–73,
210, 254 75–76, 78–80, 82–83, 100–3, 115,
Index 293
137, 155–60, 161, 162–63, 208, 237– Flanagan, Owen, 61n.9, 152–53
38, 244–46, 248, 271–73 Fodor, Jerry A., 27n.4, 29, 36–38n.18
“psychoneural” economics, 18–19, fragility, 89–91, 92–93
74–76, 78 ––80, 102–3, 159, 161, 208, frame, 17–18, 83, 94–107, 109–10, 116–18,
244–46, 248, 271–73 124, 126–27, 132–35, 136–37, 140–
Edgeworth, Francis Y., 78–79 60, 166–69, 177–79, 203–8, 236–37,
Einstein, Albert, 223, 247–48 239–40, 242–48, 252, 261–68, 273–75
eliminative materialism, 40–41n.22, 269 framing stage, 18, 109–10, 111–18, 124–
Ellsberg, Daniel, 72–73 26, 132–35, 136–37, 141–43, 150,
emergence, 19–22, 32n.12, 163, 191–214, 160, 162, 166–71, 181–82, 188–89,
215, 248–49, 252 207, 237, 238, 246, 252, 264–65, 274–
epistemic vs. metaphysical emergence, 75, 278–79
193–97, 199, 201–2, 208–9, 213–14 Darwin’s framing stage, 140–41
strong emergence, 32n.12, 194–95, 201 Friedman’s framing stage, 155–59
weak emergence, 194–95, 200, 201 Mendel’s framing stage, 145–47, 241–42
epistemic autonomy 8–21, 25, 31–48, 76, Skinner’s framing stage, 150–51
108, 133–34, 136–37, 191–96, 204– Franklin-Hall, Laura R., 44n.26,
14, 248–49, 251–79 131n.17, 188–89
evolution by natural selection, 50–55, 58, Frege, Gottlob, 225, 226–27, 233–34, 237,
60–61, 77, 84–86, 92–93, 115, 137, 241, 246
138–43, 160, 238–40, 244, 263–65 Fregean sense, 225–27, 233–34, 237,
principle of natural selection, 50–51, 85, 241, 246–47
138, 139 Freud, Sigmund, 61, 152–53
evolutionary-developmental biology. See Friedman, Michael, 11n.6
developmental synthesis Friedman, Milton, 70–75, 82–83, 102, 135,
expected utility theory. See utility 155–61, 238, 244–45, 252, 271, 272
explanatory autonomy. See epistemic fruit fly. See Drosophila
autonomy functional organization. See functionalism
explanatory extension, 37, 232, 266 functionalism, 13–14, 40, 42–46, 64–67,
explanatory relativity, 110–18, 125–26, 98–99, 129–30, 155, 180, 194–96,
175–76, 261 236–37, 241–43, 248
explanatory relevance, 36n.13, 112–16, functionalization. See functionalism
119–26, 169, 184
external question. See internal vs. external Galilei, Galileo, 71–72n.20
questions Galison, Peter, 32n.12
Galton, Francis, 56–57, 148–49
Falk, Raphael, 144n.4, 149n.7 Garfinkel, Alan, 37n.14, 113–14,
Fazekas, Peter, 31n.11 117, 131–32
fertility, 17–18, 84–88, 90–92, 96–98, 145 gemmules, 52–53, 147–48, 238–40,
Feyerabend, Paul K., 136, 216–17, 221n.6, 243, 246
222, 224n.10, 224, 227, 230–32, 248 gene, 16–17, 18–19, 35, 40, 42–46, 56–61,
Firestein, Stuart, 1, 3, 4–5, 48, 115, 134– 77–78, 115–16, 129–30, 144–49,
35, 276–77 159–60, 171–74, 207–8, 230, 239–48,
Fisher, Ronald, 57 261, 266–68
fitness, 17–18, 50–53, 83–94, 96–99, 101, genetics, 16–19, 30–31, 34–36, 40, 51n.4,
104–8, 123, 140–43, 160–61, 184, 53, 55–61, 66–67, 77–79, 115, 129–
208, 239–40, 243, 263–65 30, 137, 144–49, 159–60, 207–8, 230,
propensity interpretation of, 17–18, 83, 240–43, 244, 245, 252, 255, 261–
86–92, 96–98, 105–6 62, 265–68
294 Index
genotype, 97–98, 148–49, 230 Hume, David, 120

Giere, Ronald N., 127–28, 131 Humphreys, Paul, 193n.1, 196n.7, 209n.13
Gigerenzer, Gerd, 3, 14 Hutchison, Terence, 69
Gillet, Carl, 11n.7, 196n.7, 207n.12 Hütteman, Andreas, 31n.10, 38n.18
Glennan, Stuart, 164n.1, 164–65, 165n.2, Huxley, Julian, 57–58
168n.3, 168–69, 170, 175n.7, 258n.4 Huxley, Thomas Henry, 58
Glimcher, Paul, 245n.25, 272n.8
Godfrey-Smith, Peter, 85n.2, idealization, 3, 15–18, 26, 71–72, 103, 126–
238n.23, 258n.4 34, 136–37, 170–76, 187–90, 202–11,
Goethe Johann Wolfgang, 250, 279–80 236, 243–46, 258, 275–76
Goldbach’s conjecture, 260 identity theory. See philosophy of mind
Goodman, Nelson, 90n.6, 124–25, ignorance, 1–6, 11–13, 15–17, 21, 23, 26,
125n.9, 131–32 48, 49, 67, 79, 109, 132–33, 179–80,
Gould, Stephen Jay, 59 183–84, 185–86, 250–53, 276–78
Grandy, Richard, 229n.17 disruptive ignorance, 4, 276–77
Grene, Marjorie, 140–41 productive ignorance, 2–3, 4–6, 11–13,
Griffiths, Paul E., 144n.4, 149n.7, 230n.18 16, 21, 23, 26, 48, 49, 82–83, 109, 250,
Gul, Faruk, 75n.26, 75–76, 76n.27 251, 253, 276–78
strongly productive ignorance, 277–78
Hacking, Ian, 32n.12, 216n.1, 218–19, weakly productive ignorance, 277–78
220n.5, 221–23, 224n.10, 226 incommensurability, 217–49, 259, 275
Haldane, John B.S., 57 inheritance. See heredity
Hall, R.L., 69n.17 intension. See sense
Hanson, Norbert R., 13–14, 179, 180, 181– intention. See speaker’s intention
82, 216–17 interdisciplinarity, 7–8, 25, 38, 42, 197
Harrison, Glenn W., 75n.25 interfield theory, 115–16
Hatfield, Gary, 61n.9 internal vs. external questions, 260–63,
Hayek, Friedrich, 69–70 265, 267–68, 270, 273, 275
Hempel, Carl G., 62–63, 93–94, 96n.8, 112,
123, 195–96, 199, 205–6, 216, 218– James, William, 61
19, 223n.8, 233n.21, 246 Johannsen, Wilhelm, 144, 147, 148–49
heredity, 16–17, 18–19, 50–57, 58–59,
66–67, 77–79, 115–16, 134, 138–49, Kahneman, Daniel, 3
159–60, 162–63, 180, 238–43, 261– kairetic model of explanation, 120–24,
62, 263, 264–65, 266 132–33, 141, 182
inheritance of acquired characters, 51– Kaplan, David M., 170, 181–82
52, 145, 147–48 Kauffman, Stuart, 164
principle of inheritance, 50–51, 79, 144 Kim, Jaegwon, 40–41n.22, 46n.27,
soft inheritance, 51–52, 147–48 194, 213
Herschel, William F., 140–41 Kincaid, Harold, 37n.14, 44n.26
Hitch, C.J. 69n.17 Kitcher, Philip, 22n.9, 34–41, 44n.26,
holism, 31n.10, 217–18, 233n.21, 246, 50n.2, 50–51, 113n.2, 217n.2, 221n.6,
248, 259 226–42, 247–48, 266, 274–75
meaning holism, 31n.10, 217–18, 248 Klein, Colin, 31n.11
testing holism, 259 Knight, Frank, 69–70
Hooker, Clifford A., 41n.23 knowledge. See scientific knowledge
Hull, Clark L., 62–63, 150 Kordig, Carl R., 224n.9
Hull, David, 34–35, 37n.14, 40n.21 Kripke, Saul, 228nn.15–16
Index 295
Kuhn, Thomas, 6, 20–21, 22n.8, 30, 60–61, meaning incommensurability (see

216–17, 218–24, 226–38, 246–48, 259 incommensurability)
mechanism, 15–19, 26, 34–36, 51–61,
Lakatos, Imre, 12, 217n.2 65–67, 73–80, 83, 91–103, 105–7,
Lamarck, Jean Baptiste, 51–52, 142 109–10, 121–25, 131–34, 136–37,
LaPlace, Pierre Simon de, 9, 226 142–43, 147–49, 153–54, 159–61,
Laplacian Demon, 9, 42 162–90, 206–9, 215, 236–49, 252,
Latour, Bruno, 216–17 258, 261–64, 269–73
Laudan, Larry, 30, 217n.2 new wave of mechanistic philosophy,
Lavoisier, Antoine, 230–31, 235, 247–48 19, 162–90, 215, 257–58
laws. See scientific laws Medawar, Peter, 1–4, 278
Leahey, Thomas H., 62n.10, 152–53, Mendel, Gregor, 16–19, 35, 53, 55–61, 65–
154n.10 67, 77–80, 82–83, 115–16, 117, 124,
Lester, Richard A., 69n.17 135, 144–49, 155, 159–61, 162–63,
levels of explanation, 7–8, 15–16, 19–20, 180, 237–44, 247–48, 252, 265–68
23–48, 88, 103, 107, 122n.8, 176–77, Mendel’s laws, 34–35, 56, 145–47, 160,
178–79, 184–85, 191–214, 242–43, 208, 266–67
255–64, 266–67, 275–80 Mendelian factor, 144, 146–47, 149, 160,
Lewis, David K., 40–41n.22, 118–19n.6 241, 266
life sciences, 46, 57, 60–61 Mendelian ratios, 144–46, 149, 160, 266
linguistic framework, 260–63, 267–68, 275 Menger, Carl, 69–70
Lloyd, Elizabeth, 30n.8 mental state. See psychological state
logical empiricism. See logical positivism Mill, John Stuart, 78–79, 192
logical positivism, 11, 20–22, 24–32, 37, Millikan, Robert A., 226
42–43, 62–63, 68–74, 102, 112, 155, Mills, Susan, 86, 89, 91
192, 215–25, 253, 261, 271–72 Millstein, Roberta, 164n.1
Lorentz, Hendrik, 226 mind-body problem, 32, 40, 40–41n.22,
Love, Alan, 31n.10, 38n.18, 173n.5, 175n.9 256, 269–70
Lovejoy, Arthur O., 192 misrepresentation. See idealization
Lyell, Charles, 140–42 mode of presentation. See Frege
model. See scientific model
Mach, Ernst W.J.W., 148 modern synthesis, 16–17, 58–61, 66–67,
Machamer, Peter, 164, 168–69 77–79, 82–83
Mackie, John L., 90n.6, 125–26, 131–32 Molière, 82, 89–91, 106–7
Malthus, Thomas R., 50, 140–41 Morange, Michel, 144n.4
Marshall, Alfred, 78–79, 157 Morgan, C. Lloyd, 192
materialism. See physicalism Morgan, Mary S., 131nn.18–19
matryoshka model of science, 278–79, 279f Morgan, Thomas H., 146–47, 149,
Maull, Nancy, 31n.9, 116n.5 241, 242–43
Mayr, Ernst, 9n.5, 50n.1, 50n.3, 51n.4, Morgenstern, Oskar, 74–75, 157–158n.13
51n.5, 52n.6, 53n.7, 54–55, 56–57, Morrison, Margaret, 131nn.18–
58n.8, 60–61, 129–30, 139, 143n.3, 19, 212–13
145–46, 255n.1 multiple realizability, 24–25, 29–31, 34–
McCabe, Kevin A., 58n.8 35, 40, 182–83, 253, 266–67
McDougall, William, 62
McLaughlin, Peter, 194n.2 Nagel, Ernest, 16, 26–35, 71n.19, 216, 219–
meaning, 217–49, 274 23, 253, 265–66
meaning holism (see holism) Nagel, Thomas, 9n.5, 37n.15
296 Index
natural kind, 27n.4, 29, 36, 92–93, 190, 222–23, 226, 229, 247–48, 253–55,
225–36, 266, 273–74 272, 275–76
natural selection. See evolution by natural physical property, 84–97, 105–8, 128–
selection 33, 194, 204, 205–6, 209, 262, 269
neural network, 197–204 Piccinini, Gualtiero, 180
neural state. See psychological state Pigou, Arthur C. 78–79
neural synchronization, 123, 198–200, 211 Pillsbury, Walter, 62
Neurath, Otto, 218–19 Pinker, Steven, 13–14
neuropsychology. See neuroscience Place, Ullin Thomas, 40, 269
neuroscience, 7–8, 39–40, 45–46, 66–67, placeholder, 17–20, 77–78, 82–108, 109–
73–75, 78–79, 80, 154, 168–69, 10, 117–18, 122–24, 132–35, 136–61,
181n.12, 189, 193, 197–201, 204, 162–69, 177–90, 202–9, 236–44, 252,
212–13, 217, 254–56, 258n.5, 272–73 264–67, 275–76
systems neuroscience, 197–201, 204 Plato, 73–74
Newton, Isaac, 8–9, 77, 226, 247–48, 259 Popper, Karl R., 80, 216, 218–20, 250
Nicholson, Daniel, 187–88 Pott, Johann H., 103–4
normal science vs. revolutionary powers. See causal powers
science, 219–21 preference, 67–68, 72–248
Priestley, Joseph, 103–4, 227–28, 230–31,
ontogeny, 52–54, 58–61, 67, 77–78, 147– 235–37, 238–39, 274
49, 203, 238–39, 241–42 principle of charity. See principle of
Oppenheim, Paul, 28n.6, 123, 195–96, humanity
199, 205–6 principle of humanity, 229, 234, 239–40,
Orzack, Steven, 14 243, 247–48
progress. See scientific progress
pangenesis, 52–53, 147–48, 238–39, 240 propensity. See disposition; fitness
Paracelsus, 222 protein, 43–44, 60, 172–73, 199–211, 242–
paradigm. See scientific paradigm 43, 261
Pareto, Vilfredo, 245 Provine, William B., 58n.8
Pavlov, Ivan, 62, 64n.12 psychological state, 16–17, 18–19, 34, 40,
Pearson, Karl, 148–49 45–46, 61–62, 66–67, 68–69, 72–73,
Pesendorfer, Wolfgang, 75n.26, 75– 75, 78–80, 82–83, 98–101, 106–7,
76, 76n.27 151–55, 160, 162–63, 194, 198–200,
phenotype, 40n.21, 51n.5, 144–49, 204–5, 211, 229, 234, 243–44, 246,
230, 241–42 268–70, 271–72
philosophy of mind, 13–14, 36–37, 40, 45– Psychology, 3, 7–8, 27, 36–42, 45–46,
46, 98, 206, 269–70 61–68, 73–80, 82–83, 98–100, 128,
philosophy of science, 6–11, 19–22, 24–26, 137, 149–55, 189, 193, 212–13,
31, 38–39, 49, 80, 102, 107–8, 112, 220–21, 243–44, 252–56, 268–
131, 163–67, 189, 213–14, 215–32, 70, 272–73
248–49, 251–53, 259–61 psychological behaviorism, 61–67,
phlogiston, 103–7, 226–39, 243, 78–80, 98, 100, 115, 137, 149–55, 160,
247, 273–75 162–63, 237–38, 243–44, 268–70
phylogeny, 53–54, 57, 59, 77–78 Ptolemy, 222–23
physicalism, 25, 27n.4, 32–33, 36–39, 40, Punnett, Reginald C., 146–47
62, 64, 99, 192, 194–95, 210, 254, 269 Putnam, Hilary, 9n.5, 28n.6, 29, 32–37,
physics, 7–12, 24–43, 47, 56, 62–65, 69– 38–39, 41, 133, 137, 215, 225–26, 228,
72, 123, 128, 147, 154, 189, 211–13, 242, 255, 257, 262–63
Index 297
qualia. See consciousness Schaffner, Kenneth, 31n.11, 41n.23, 44n.26

Quine, Willard Van Orman, 13–14, 90n.6, Scheffler, Israel, 226n.13
222–23, 233n.21, 261 Schlick, Moritz, 62–63, 218–19
scientific knowledge, 1–6, 11–13, 16,
Railton, Peter, 119–20 20–21, 23, 48, 49, 67, 88–89, 109, 115,
rationality. See scientific rationality 127, 162, 179–80, 192–93, 195–96,
reduction. See reductionism 200–1, 216, 219–21, 248, 250–
reductionism, 6–12, 15–16, 19–22, 24–48, 53, 276–79
49, 82–83, 99, 107–8, 154, 163, 165– scientific laws, 5, 8–10, 26–31, 33–37, 39–
66, 191–92, 194–95, 201, 206–7, 209, 40, 45–46, 51–52, 56, 61–64, 70–72,
213–14, 248–49, 251–79 78, 102, 112n.1, 115, 125–32, 145–47,
classical model of reduction , 16, 24–31, 165–66, 174–77, 192, 220–21, 223,
41, 62–63, 210, 253–54, 261 251–54, 267
molecular reductionism, 42–44, 46 scientific model, 18–20, 26, 109–11, 116–
reference, 218, 225–49, 274 35, 136–37, 140–61, 162–90, 192,
direct reference, 228 202–5, 209, 225–32, 236, 239–42,
reference potential, 229–48, 274–75 243, 252, 253–54, 255–56, 259, 261–
referential change, 217–18, 225–27, 63, 265, 267, 275–80
232–33, 237 mechanistic model, 103, 109–10, 162–
referential models of meaning, 217–18, 90, 245–46
224–32, 233, 243, 248 scientific paradigm, 30, 218–38, 246–
Reichenbach, Hans, 216, 218–19 48, 259–75
relevance. See explanatory relevance scientific progress, 20–21, 22, 25, 29, 30–
Rensch, Bernhard, 57 31, 49, 61, 103–5, 213–14, 215–49,
representation. See scientific model 252–53, 259, 261, 269–79
representation stage, 18, 109–10, 126–35, cumulative model of scientific progress,
136–37, 162, 169–71, 175n.8, 237– 1–6, 11–12, 16, 20–21, 23, 30, 48,
38, 252 49, 216, 218–21, 248, 250–51,
Darwin’s representation stage, 142–43 261, 276–79
Friedman’s representation stage, 159–60 scientific rationality, 216–17, 220–24
Mendel’s representation stage, 147–49 scientific revolution, 219–32, 238
Skinner’s representation stage, 152–54 scientific theory, 6–12, 26–31, 40, 70–71,
reproduction. See fertility 127, 131–32, 196, 202, 207, 216,
revealed preference. See preference 219–24, 233, 253–54, 259, 261–62,
Richardson, Robert, 164, 168–69, 175– 272, 275–76
76, 258 Scriven, Michael, 112n.1
Robbins, Lionel, 67–68, 72, 74, 103, 273 sense. See Frege
Rorty, Richard, 13–14 Shapere, Dudley, 30, 217n.2, 224n.9, 224
Rosenberg, Alexander, 39n.20, 40n.21, sickle-cell anemia, 43–44
41n.23, 42n.24, 44n.26, 72n.23, 86n.3 Silberstein, Michael, 185n.15
Russell, Bertrand, 125n.10, 201 Simon, Herbert A., 71n.18
Russian doll, See matryoshka Simpson, George G., 57
Ryle, Gilbert, 90n.6, 100 Skinner, Burrhus F., 18–19, 62–67,
79–80, 82–83, 100, 135, 149–55,
Salmon, Wesley, 113n.2, 119–20, 162, 181 160–61, 162–63, 238, 243–44, 248,
Sarkar, Sahotra, 42n.24 252, 268–70
Saunders, William, 146–47 Skipper, Robert, 164n.1
Savage, Leonard, 102, 155–59, 244–45 Sklar, Lawrence, 29n.7
298 Index
Smart, John Jamison Carswell, 40, 269 utility, 72–79, 100–3, 106–7, 155–60, 208,
Smith, Adam, 78–79 211, 244–46, 248, 271–73
Smith, Laurence D., 63n.11 expected utility theory, 72–74, 156–59,
Sober, Elliott, 39n.19, 85n.2, 86–88, 208, 271
89, 91, 96–97, 98, 106–8, 123, marginal utility, 100–2, 157
154n.11, 263–64 revealed preference view of utility, 72–
social sciences, 7–8, 36–37, 161, 197 76, 78–79, 102, 155, 159, 245, 271
solubility, 89–93, 95–96, 98–101, 104–
7, 255–56 van Eck, Dingmar, 175n.9
speaker’s intention, 175n.7, 228–29, 234, van Fraassen, Bas, 30n.8, 112–13, 125n.9,
236, 237–38, 247–48, 275 131n.16
special sciences, 8–11, 30, 36–41, 93, 107– van Strien, Marij, 9n.4
8, 187, 253–56 variation, 18–19, 50–60, 66–67, 77–80,
Spencer, Herbert, 85 84–97, 138–47, 160, 238–43, 263–65
Sperry, Roger Walcott, 192 principle of variation, 50–51
Sporns, Olaf, 197–99, 200, 201, 204 principle of variation in fitness, 50–51
square-peg-round-hole example 33–36, viability, 17–18, 84–88, 90–92, 96–
39, 47, 133, 137, 255, 262–63 98, 105–6
St. Petersburg paradox, 157 Virtus dormitiva, 82, 87, 89–94, 97, 99–
Stahl, George E., 103–4 100, 104, 106–7
Stebbins, George L., 57 von Mises, Ludwig, 69–70
Stotz, Karola, 144n.4, 149n.7, 230n.18 von Neumann, John, 74–75, 157, 158n.13
Strevens, Michael, 36n.13, 41n.23, 114–
15n.4, 119–22, 125–26, 130n.14, 131– Waters, C. Kenneth, 40n.21, 144n.4,
33, 141, 179, 182–86 149n.7, 242n.24
supervenience, 17–18, 25, 37–39, 83–108, Watson, James D., 242–43
210, 254 Watson, John B., 62, 63–64, 79–80, 149–
supervenience emergentism, 194 51, 154, 155, 270
Suppes, Patrick, 30n.8, 131n.18 Weber, Marcel, 144n.4, 149n.7
survival. See viability “wedding cake” model of science, 7–
8, 193–94
Taylor, Elanor, 196n.6 Weed, Douglas L., 14
tendency. See disposition Weinberg, Steven, 9n.5
Theoretician’s dilemma, 153–54, 244, 268 Weisberg, Michael, 128–31, 175n.8
theory change. See scientific progress Weismann, August, 53, 56–57, 147–
Theory-ladenness, 196–97, 222–23, 229– 49, 238–39
31, 237–46, 274–75 William of Occam, 186
Thomson, Joseph J., 226 Wilson, Charles E., 131n.17
Thorndyke, Edward, 62 Wilson, Jessica, 196n.8
Tolman, Edward C., 62–63, 150 Wimsatt, William, 3–4, 11n.7,
Toulmin, Stephen, 30 131n.18, 164
transmission of traits. See heredity Wittgenstein, Ludwig, 47, 100
truth, 1–3, 5, 6, 70, 127–28, 174, 190, 215– Wright, Cory, 175n.9
16, 248, 250–51, 272, 278–79 Wright, Sewall, 57
Wundt, Wilhelm, 61
unification. See unity of science
unity of science, 28–29, 30–31, 231 Zador, Anthony, 199n.10

Black Boxes - How Science Turns Ignorance Into Knowledge

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

Black Boxes - How Science Turns Ignorance Into Knowledge

Uploaded by

Copyright:

Available Formats

Published in the United States of America by Oxford University Press

© Oxford University Press 2021

All rights reserved. No part of this publication may be reproduced, stored in

You must not circulate this work in any other form

Library of Congress Cataloging-in-Publication Data

1. Bricks and Boxes  1

5.4. Step Three: The Representation Stage  126

Practice. I am indebted to all the students who shared thoughts, comments,

Knowledge is a big subject.

§1.1. The Wall

perspective, is to provide an accurate and complete description of the uni-

1 See, for instance, Gigerenzer (2007) and Kahneman (2011).

in well with the image of science as a cumulative, brick-​by-​brick endeavor,

§1.2. A Theory of Everything?

The previous section concluded by posing a puzzle. Why do textbooks pre-

Classic and contemporary philosophy of science has been molded by a

Figure 1.1. The “wedding cake” layering of reality.

under the moniker “neuropsychology,” knowing perfectly well that their

precisely from any comprehensive description of a past state, in conjunction

theory of everything, a grand characterization of the world that, in principle,

able to successfully capture the interplay between knowledge and ignorance

§1.3. Pandora’s Gift to Science: The Black Box

Section 1.2 introduced my pars destruens: the debate between reductionism

a comprehensive analysis will be undertaken throughout the monograph, let

These issues will be developed throughout the monograph. Meanwhile,

§1.4. Structure and Synopsis

§1.5. Aim and Scope

traditional reductionism vs. antireductionism divide and recast these impor-

Come fa l’onda là sovra Cariddi,

Our discussion began, in Chapter 1, with a well-​known image of science.

§2.2. The Rock Shoal and the Whirlpool

In Greek mythology, sailors who dared to navigate across the Strait of

an “antireductionist consensus” was forged. On this view, progressive reduc-

If this old-​fashioned debate is not a promising way forward, in what

§2.3. The Rise and Fall of Classical Reductionism

derivation to get off the ground, a second condition of “connectability” must

(R1) S1x ↔ P1x

How should the biconditional be interpreted, for bridge principles like R1

5 Prominent examples are Dupré (1993) and Cartwright (1999).

(R3) S1x ↔ P1x ∨ . . . ∨ Pnx

For most higher-​level predicates S, the number of corresponding lower-​level

A second difficulty with classical reductionism concerns the underlying

genetics, notoriously fail (§2.4). Finally, classical reductionism is not only

§2.4. Antireductionism Strikes Back

If antireductionism were tantamount to the rejection of Nagel’s model

9 For an excellent discussion of these points, see Maull (1977).

(a) If humans are purely material entities, then there is a physical

(a) If humans are purely material entities, then there is a physical

Figure 2.1. Putnam’s square-​peg-​round-​hole scenario.

various ways. Nevertheless, appeals to molecular biology do not deepen cy-

This point deserves to be clarified. In general, richer biochemical

§2.5. The Reductionists Return

As discussed in the previous section, antireductionism emerged from

15 For a recent discussion of these issues, see Nagel (2012).

16 This objection is presented and discussed in detail by Bickle (1998, 2003).

corresponding change in the underlying micro-​structure. Yet, qua anti-

19 For a response along these lines, see Sober (1999, 2000).

Related worries have been raised against Kitcher’s arguments developed

§2.6. Scylla or Charybdis?

several functional biological concepts have already been characterized from

25 For a more detailed discussion of these examples, see Nathan (2012).

and functional notions in structural terms would be anywhere from frus-

It is important to stress that the issue is independent of biology. Analogous

1. Bricks and Boxes 1

5.4. Step Three: The Representation Stage 126

in well with the image of science as a cumulative, brick-by-brick endeavor,

Our discussion began, in Chapter 1, with a well-known image of science.

If this old-fashioned debate is not a promising way forward, in what

For most higher-level predicates S, the number of corresponding lower-level

Figure 2.1. Putnam’s square-peg-round-hole scenario.

corresponding change in the underlying micro-structure. Yet, qua anti-

27 From a functional perspective, to “reduce” a higher-level theory to a lower-level one is to provide

rather lies in Darwin’s pioneering understanding of the far- reaching

This grand union—epitomized by Julian Huxley’s 1942 book Evolution: The

§3.4. Absent-Minded: Psychological Behaviorism

12 Skinner is often portrayed as advocating a naive “stimulus-response psychology,” analyzing be-

processes, Skinner purported to dismiss—or, more charitably, to minimize—

24 For the sake of convenience, I introduce the moniker “psycho-neuro-economics” to lump

neuro-psychological mechanisms, which played no role, in principle or in

The previous section argued that macro-properties supervening on micro-