ARTIFICIAL INTELLIGENCE

CS60045 Course Introduction

PROF. PALLAB DASGUPTA

INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR
pallab@cse.iitkgp.ernet.in
http://cse.iitkgp.ac.in/~pallab

Department of Computer Science and Engineering

 Human Cognition Abilities

LEFT BRAIN FUNCTIONS RIGHT BRAIN FUNCTIONS

Logic Creativity
Analysis Imagination
Sequencing Holistic Thinking
Mathematics Intuition
Language Arts (Motor Skill)
Facts Rhythm (beats)
Words of songs Visualization
Computation Tune of songs
Daydreaming

Modern AI is the
Symbolic methods: Classical AI Statistical methods: ML
convergence of these two

INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR 2

 What is Intelligence?

Reasoning Learning

Knowledge Communication

Planning Perception

Kinesthetic

3
IIT KHARAGPUR
 What is “Artificial Intelligence”?

Turing Test (1950)

• The computer is interrogated by a human via
a teletype.
• It passes if the human cannot tell if there is a
computer or human at the other end

IIT KHARAGPUR 4
 Evolution of AI

1956 1980 1997 2011 2012 2015 2016

1950 1969 1987-1993 2008 2011/12 2014 2016 2018

IIT KHARAGPUR
 Kasparov versus Deep Blue

On May 11, 1997, it won a 6-game

match by 2 wins to 1 with 3 draws

Today, we have power to evaluate

200 million moves per second !!

IIT KHARAGPUR
 What are the green fields for AI in India?

• Preventive and affordable Healthcare

• Agriculture and Rural Development
• Smart Mobility and Intelligent Transportation Systems
• Retail
• Manufacturing
• Energy management
• Smart Cities
• Education and Skilling

Source: Niti Aayog Discussion Paper on AI, June 2018

IIT KHARAGPUR
 Artificial Intelligence

CORE CAPABILITIES
• The ability to solve problems
• Constraint satisfaction, Optimization,
Search
• The ability to plan
• Abstraction
• The ability to deduce
• Logic, Reasoning algorithms
• The ability to learn
• Models, Data, Learning algorithms
• The ability to handle uncertainty
• The ability to interface with the real world

IIT KHARAGPUR Source: Accenture

 Constraints and Optimization

Path Finding
• I wish to find a shortest path
• I wish to find a path with minimum congestion
• I wish to find a path with combination of transportation
options (metro, bus, taxi)
• I wish to find a path which goes past a medicine shop
• I wish to find a path which minimizes energy
consumption from my battery in a e-vehicle
When the size and complexity becomes too big we use
“heuristic functions” to cut out unnecessary parts.

In the lack of domain knowledge, we can statistically learn

the best way (reinforcement learning) by exploration.

Modern AI aims to combine learning from data with

structured use of domain knowledge.

IIT KHARAGPUR 9
 AI Planning
Elements of a Planning Problem
• A set of states (worlds) described in terms of predicates
• A set of actions which transforms some parts of one world to
take us to another world
• An initial world
• A goal in terms of the predicates that must hold in the final
world

Planning is widely used in robotics and automated control

Modern AI explores techniques that combine planning with

machine learning
• Autonomous driving is one of many areas where such
combinations are highly relevant

IIT KHARAGPUR 10
 Logical Reasoning
• Automated ways to use what is known to reason
about something which is not explicitly known.
• Automated Reasoning:
• Deduction
Rule: All the marbles in this bag are blue
Case: These marbles are from this bag
Inference: These marbles are blue Five color theorem: All maps can be colored with five
colors, where neighboring countries get different colors
• Abduction [Proved in 1800s]
Rule: All the marbles in this bag are blue
Four color theorem: All maps can be colored with four
Observation: These marbles are blue
colors, where neighboring countries get different colors
Case: These marbles are from this bag [Proved in 1976 with help of computers]

• Induction Applications of Logical Reasoning

• Automated Theorem Proving
Case: These marbles are from this bag
• Rule-based Systems
Observation: These marbles are blue • Complexity Analysis
Rule: All the marbles in this bag are blue
IIT KHARAGPUR
 Logical Reasoning with Other Fun Things

father( x, z ), father( z, y ) Who is the maternal great uncle of Freya?

⇒ grandfather( x, y )

mother( x, z ), father( z, y )
⇒ maternalgrandfather( x, y )

mother( x, z ), mother( z, y )
⇒ maternalgrandmother( x, y )

father( x, z ), mother( z, y )
⇒ grandmother( x, y )

maternalgrandmother( x, z ),
mother( z, p ), son( p, y )
⇒ maternalgreatuncle( x, y )
maternalgrandmother( Freya, Charlotte ),
mother( Charlotte, Lindsey ), son( Lindsey, Fergus )
⇒ maternalgreatuncle( Freya, Fergus )

We could build this into a social media platform which suggests Freya to post a picture of Fergus on
the Maternal-Great-Uncle day !!
IIT KHARAGPUR 12
 Reasoning under Uncertainty
P(Obesity)
0.24 History of History of viral History P(Chronic
Obesity
alcohol abuse hepatitis of VH Hepatitis)
T 0.3
F 0.05
Hepatitic steatosis Cirrhosis Chronic Hepatitis

Triglycerides AST ALT Total bilirubin Fatigue

AST: Aspartate aminotransferase
ALT: Alanine aminotransferase

Jaundice
• We know the relations between events in probabilistic terms
• We are given a set of known events
Challenge: Sometimes we do not know
• Goal: Find the probability of some other event
which is the cause and which is the effect. 13
IIT KHARAGPUR
 Machine Learning

IIT KHARAGPUR
 Deep Learning and the Problem of Explainability

Test Image Evidence of Husky Evidence of Flute

Husky Classified Explanation based Same pixels used to explain two different classifications !!
as Wolf on saliency

It has evidence for cat as well as

evidence for bird, but it is neither

… but it is also not easy to define the features that

separate a dog from a cat, even if we can paint both
IIT KHARAGPUR 15
 The book that we will follow mostly for this and
many other topics:

Artificial Intelligence – A Modern Approach

Stuart J Russell, Peter Norvig

Pearson Education India

16
INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR