INDIAN INSTITUTE OF TECHNOLOGY ROORKEE

MIN-516: Artificial Intelligence

Dr. Anuj Bisht

Department of Mechanical and Industrial Engineering, IIT Roorkee
Syllabus/Brief Outline
• Artificial Intelligence: Definition, scope, brief history
• State spaces
• Knowledge representation
• Reasoning and inference strategies
• Expert system and their applications
Complete Outline

2
Book

Bible of AI
• S. Russell (Berkeley) and P. Norvig (Google): “Artificial
Intelligence: A Modern Approach”, 3rd edition, Prentice-
Hall, 2009

• E. Rich, K. Knight, SB Nair: “Artificial Intelligence”

• TL Dean, J Allen, Y Aloimonos: “ Artificial Intelligence: Theory
and Practice”
• MR Genesereth and N Nilsson: “Logical Foundation of
Artificial Intelligence”

3
Marks Distribution

Attendance/class interaction 10
CWS 25
MTE 25
ETE 40

CWS: Assignment, class quiz, etc

4
AI Breadth

• AI very broad field, big community

– courses topics taught in different departments of the world may be
limited
• AAAI conference (Association of advancement of Artificial
Intelligence): 8-9 parallel track

5
Goals

• Brief intro to the philosophy of AI

• Give you idea about the breadth of ideas in AI

• Traditional AI vs modern AI (60+ years of history)

• Build general thinking for new problem solving

• Better understanding while opting for specialized advanced

course

6
AI evolution

Flavors of:
• Search oriented view
• Knowledge representation-oriented view
• Modern AI: Machine learning oriented view
– Statistical and Neural model-based AI

7
Theory vs Modeling vs Algorithm vs Applications

• Lectures inclined towards modeling

• Assignments exams tilted towards applications
• Not much proof/theorem establishment

8
INDIAN INSTITUTE OF TECHNOLOGY ROORKEE

History
1946: ENIAC computer

• First electronic general

purpose computer
• Turing complete: “any real-
world general-purpose
computer or computer
language can approximately
simulate the computational
aspects of any other real-
world general-purpose
computer or computer
language”
• Can solve that times large
class of numerical problems
• Used to Design artillery firing • Human computer: Do manual
table for US army. Can computations
calculate projectile trajectory
in 60 s
10
1950: Turing asks the question
“Can Machine think?”
– Alan Turing, 1950
Thought experiment
• Can a person think?
• Paralysis: How to know if brain perfectly
working or not?
• Parkinson’s disease: Stephen Hawking
• Communication to the world: perceiving
the environment and giving output is
critically important to judge. Brain is
meaningless in absence of communicators

• Ask machine questions and judge based on decisions

• Turing test: a method of inquiry in artificial intelligence (AI) for determining
whether or not a computer is capable of thinking like a human being.
11
Turing test (imitation game)
• It is a test of a machine's ability to exhibit intelligent behaviour equivalent to, or
indistinguishable from, that of a human. if it can mimic human responses
under specific conditions.
Original Turing Test Questioner: interrogates the
Respondents

respondents within a specific subject

area, using a specified format and
context.

After a preset length of time or number

Questioner

of questions, the questioner is then

asked to decide which respondent was
human and which was a computer.

• The test is repeated many times. If the questioner makes the correct determination in half
of the test runs or less, the computer is considered to have artificial intelligence because
the questioner regards it as "just as human" as the human respondent.

12
1956: New field is born

• 1956 Dartmouth Summer Founding Fathers of AI

Research Project on
Artificial Intelligence: 2
month, 10 man study of
Artificial Intelligence

• John McCarthy, Marvin

Minsky, Nathaniel John McCarthy
Rochester, Claude Shannon Chess, Lisp Allen Newell, Herbert Simon
• Allen Newell, Herbert Simon Stanford Logic theorist
CMU
• AI researchers are highly
optimist Marvin Minsky
Neural network learning
machine: SNARC
MIT
13
1956-1966

• 1950: Turing test for Machine

Intelligence

• 1956: AI born at Dartmouth

workshop

• 1964: Eliza – first chatbot

psychotherapist

• 1966: Shakey – general purpose

mobile robot

14
 AI Winters
• 1974-1980: Winter 1
– Failure of machine translation
– Negative results in neural nets
– Poor speech understanding

• 1987 - 1993 : Winter 2

– Decline of LISP
– Decline of specialized hardware
for expert systems

• Lasting effects
– Ecomomist07 “AI associated with
systems who have failed to live
up to their promises
– BT06 “word “AI” may hurt
chances of getting funding”
15
1996: EQP algorithm

Robbins conjecture was unsolved for decades

EQP (an automated theorem-proving program

for equatorial logic) proves that Robbin’s
Algebras are all boolean

16
1997 – Deep Blue defeats human in Chess

Deep Blue defeats Gary Kasparov, World Chess Champion

I could feel human-level intelligence across the room -GK
Is Deep Blue really an AI!?
Or is it some algorithm (alpha-beta search algorithm)!

“If it works, its not AI”

17
 1999: Remote Agent based system
• 1998: Deep Space (mission) 1
– To fly by Asteroid (millions or
more of miles away)

• “Remote Agent” based AI

system supposed to manage
the spacecraft
• Diagnosis and repair system
to identify and rectify any fault
onboard spacecraft
autonomously

• Remote Agent did its job in

simulation onboard.
18
 2004 and 2009 – Mars Rover
• First generations: did not had
any onboard planning
– Scientist round table planned
schedule for the workday

• Very time consuming/wasting

• Rover sitting idle for most time

• Enable it to make some level

of autonomous decision

• AI Planner software to
automatically generate plan
based on inputs
19
2005: DARPA Grand challenge
• First working demonstration
of long trip by self-driving car

• Stanley and three other cars

self drove 132 miles
mountain road in Nevada

• Winner: Standford, Prof.

Sebastian group, Stanley

• Runner’s up: CMU, Prof.

Red, 2 entries: Sandstorm
and H1ghlander
20
2011: IBM’s Watson

• Watson won “Jeopardy” (TV’s smartest quiz show)

• Display of natural language processing
• DeepQA computer
21
 INDIAN INSTITUTE OF TECHNOLOGY ROORKEE

Present: 2007 onwards

• First generation of AI: Logic based AI, Symbolic AI, Search AI

• Second generation (late 90s-early 2010): Probabilistic AI
(Judea Pearl, got turing award)
• Third generation, 2007 onwards: neural models overtaken
and overshadowed everyone.
2016: AlphaGO

• GO: 19*19 board

game
• Harder with many
other complications
• Lee Sedol from
korea lost by 4:1

• Deep Mind: UK based small researcher (David Silver) group

purchased by alphabet (google) (2014-15) for 500 million
USD
• Earliest high quality deep-learning technology based
neural models specialized for specific area of reinforced
learning
23
• Atari game

• A single algorithm
based on deep
neural networks,
which learns playing
games and
specializes playing
them

24
What changed!?

25
Object Recognition: Image Recognition

• CIFAR-10
• Canadian Institute
for Advances
Research

Network Error Layer

AlexNet 16.0% 8
ZFNet 11.2% 8
VGGNet 7.3% 19
GoogLeNet 6.7% 22
MS ResNet 3.6% 152

26
Image and Text

27
28
AI doomsday scenario

29
INDIAN INSTITUTE OF TECHNOLOGY ROORKEE

What is AI!?
Science of AI

• AI: what is the nature of intelligent thought ?

– Can I mechanize thought?
– how thoughts happen and put it in machine

31
What is intelligence?

• Dictionary.com: capacity for learning, reasoning,

understanding, and similar forms of mental activity

• Ability to perceive and act in the world

• Planning: take decisions
• Learning and Adaptation: recommend movies, learn traffic
patterns
• Reasoning: proving theorems, medical diagnosis
• Understanding: text, speech, visual scene

32
Intelligence vs humans

• Are humans intelligent?

– Replicating human behavior early sign of intelligence

• Are humans always intelligent?

• Can non-human behavior be intelligent?

33
What is artificial intelligence?

Human-like vs Rational
(automation of) activities The study of mental
that we associate with faculties through the use
Thought human thinking, activities of computational models
such as decision making, (Charniak & McDertmott
problem solving, learning 1985)
vs …. (Bellman 1978)

The study of how to The branch of computer

Behavior make computers do science that is concerned
things at which, at the with the automation of
moment, people are intelligent behavior
better (Rich and Knight (Luger & Stubblefield
1991) 1993)

34
What is artificial intelligence?

Human-like vs Rational

Thought Systems that think Systems that think

like humans rationally
vs

Systems that act like Systems that act

Behavior
humans rationally

35
Turing test (imitation game)

Human or machine?
Respondents

If the human cannot tell

whether the responses
from the other side of the
wall are coming from
human or computer, then
Questioner

the computer is
intelligent.

36
Thinking Humanly
• Cognitive Science
– Very hard to understand how humans think
– Post-facto rationalizations, irrationality of human thinking
• Do we want a machine that beats humans in chess or a
machine that thinks like humans while beating humans in
chess?
– DeepBlue supposedly Doesn’t think like humans
• Thinking like humans important in Cognitive Science
applications
– Intelligent tutoring
– Expressing emotions in interfaces
• The goal of aeronautical engineer is not to fool pigeons in
flying!

37
Thinking Rationally: Laws of Thought

• Aristotle: What are correct arguments/thought

processes?
– Logic

• Problems
– Not all intelligent behavior is mediated by logical
deliberation. Many are reflexes
– What is the purpose of thinking?

38
Acting Humanly

• Loebner Prize (Turing test)

– Every year in Boston
– Expertise-dependent tests: limited conversation

• What if people call a human a machine?

– Shakespeare expert
– Make human-like errors
– Mistakes make us human

• Problems
– Not reproducible, constructive or mathematically
analyzable
39
Acting Rationally

• Rational behavior: doing the right thing (by any

process?)
• Need not always be deliberative
– Reflexive
• Aristotle (Nicomachean ethics)
– Every art and every inquiry, and similarly every action and
every pursuit is thought to aim at some good

40
Acting → Thinking

• Weak AI Hypothesis vs Strong AI hypothesis

– Weak Hypothesis: machines could act as if they are
intelligent
– Strong Hypothesis: machines that act intelligent have to
think intelligently too

41
Rationality and Rational Agents view of AI

• An agent should strive to do the right thing, based

on what it can perceive and the actions it can
perform. The right action is the one that will cause
the agent to be most successful
• Performance measure: an objective criterion for
success of an agent’s behavior
• Ex: performance measure of a vacuum-cleaner
agent could be amount of dirt cleaned up, amount of
time taken, amount of electricity consumed, amount
of noise generated, etc.

42
Ideal Rational Agent

“For each possible percept sequence, does

whatever action is expected to maximize its
performance measure on the basis of evidence
perceived so-far and built-in knowledge.”

• Rationality vs omniscience?
• Acting in order to obtain valuable information

43
Examples: Formal Cognitive Tasks

• Games
– Chess
– Checkers
– Othello

• Mathematics
– Logic
– Geometry
– Calculus
– Proving properties of programs

44
Examples: Expert Tasks

• Engineering
– Desing
– Fault finding
– Manufacturing planning

• Medical
– Diagnosis
– Medical image analysis

• Financial
– Stock market predictions

45
Examples: Perceptual Tasks

• Perception
– Vision
– Speech

• Natural Language
– Understanding
– Generation
– Translation

• Robot control

46
What is artificial intelligence (algorithmic view)

• A large number of problems are NP hard

• AI develops a set of tools, heuristics, …..

– To solve such problems in practice
– For naturally occurring instances

• Search
• Game playing
• Planning
• Etc ……

47
Reading assignment

• What is NP hard problem

48
Recurrent Themes

• Weak vs Knowledge-based methods

– Weak: general search methods (ex.: A* search)
• Primarily for problem solving
• Not motivated by achieving human-level performance

– Strong AI: knowledge intensive (ex.: expert systems)

• More knowledge …… less computation
• Achieve better performance in specific tasks

– How to combine weak & strong methods seamlessly?

49
Recurrent Themes

• Logica vs Probabilistic vs Neural

• In 1950s, logic dominates
– Attempts to extend logic

• 1988, Bayesian networks

– Efficient computational framework

• 2013, Deep Neural Networks

– Powerful representation across modalities

50
Topic of course

• Phase 1: Search, Constraint Satisfaction, Logic,

Games

• Phase 2: Uncertainty (Decision theory, probabilistic

knowledge representation), Learning
(reinforcement)

• Phase 3: Deep Neural Networks

51