Transcript Introduction

Lecture 1: Introduction
Heshaam Faili
[email protected]
University of Tehran
What is AI?
Foundations of AI
The History of AI
State of the Art
Definitions of AI




Develop programs/systems that
perform/act like humans
Develop programs/systems that
perform/act rationally
Understand human intelligence
Formalize the laws of thought and
action
INTELLIGENT AGENTS
2
What is AI?
Acting Humanly:The Turing Test
HUMAN
COMPUTER/
HUMAN
- types in questions
- receives answers on screen
- processes questions
- returns answers
If the human cannot tell if it is a computer or a
human, the program exhibits intelligence
3
Turing Test

Simple
test have
involve
AI Turing
researchers
devoted
little
NLP effort to passing the Turing test,
believing
that it isrepresentation
more important to study the
 Knowledge
underlying
principles
of in- intelligence than to
 Automated
reasoning
duplicate an exemplar.
 Machine learning
The quest for "artificial flight" succeeded when the
 To enhance should have
Wright brothers and others stopped imitating
 Computer vision
birds and learned about
 robotics
aerodynamics.

4
Thinking humanly

Cognitive modeling



Computer model together experimental
technique from psychology
We will not attempt to describe what is
known of human cognition
We will occasionally comment on
similarities or differences between AI
techniques and human cognition.
5
Thinking rationally


The "laws of thought" approach
Aristotle’s “right thinking”



Pattern for argument structure yield correct
conclusion
E.g : "Socrates is a man; all men are
mortal; therefore, Socrates is mortal."
Logic
6
Acting rationally



An agent is just something that acts
computer agents are expected to have
other attributes that distinguish them
from mere "programs,
A rational agent is one that acts so as
to achieve the best outcome or, when
there is uncertainty, the best expected
outcome.
7
Examples of task for AI

Play games


Process natural language


control tower conversation, stock market
briefs
Industrial applications


tic-tac-toe, chess, backgammon, poker
plant diagnostics, plan for manufacturing
Expert-level performance

molecular biology, computer configuration
8
Why is AI different than
conventional programming?

Strive for






GENERALITY
EXTENSIBILITY
Capture rational deduction patterns
Tackle problems with no algorithmic solution
Represent and manipulate KNOWLEDGE, rather
than DATA
A new set of representation and programming
techniques: HEURISTICS
9
Example: TIC-TAC-TOE
10
Program 1: hard wired



Code a table of all possible board
positions and the transitions between
them (state diagram)
Given a position, look in the table for
the next move and return
Properties:


time efficient, requires lots of storage
not extensible: requires a table for other
games
11
Program 2: less hard wired

Use procedures designed for the game:






try to place two marks in a row
if opponent has two marks in a row, place
mark in third space
Pattern matching to recognize board
positions
Can encode different playing strategies
Better space efficiency, less time efficiency
Still game-dependent
12
Program 3: AI-like

Represent the state of the game:



Use an evaluation function:



current board position
next legal positions
Rate the next move according to how likely it
will lead to a win
look-ahead of possible oponent moves
More general because it embodies a
general strategy.
13
Foundations of AI

Philosophy:





Aristotle: the first one worked on I: way of
thinking
mechanistic
views:
of
behavior
•Can formal rules be used to draw valid conclusions?
•How does the
mind arise
from a physical
materialism
or mental
dualism:
of mind
brain?
• Where
does
knowledge a
come
from?
Empiricism:
for
generate
knowledge
• How does knowledge lead to action?
Logical Positivism: all knowledge can be
connected to gather logically
14
Foundations of AI

Mathematics:
algorithms,
 logic,
 formalization of mathematics,
•What are the formal rules to draw
 Incompleteness, NP-completeness,
valid conclusions?
 decision
theory
• What can
be computed?

•How do we reason with uncertain
information?
15
How do humans and
animals of
think
Foundations
AIand act?
• How does language relate to
thought?

Psychology: behaviorism, cognitive

science. • How can we build an
efficient
computer?
Linguistics:
grammars,
syntax and
semantics.

Computer Science: computers,

software, theory
Others: neuroscience, economics, game
theory.
16
A brief history of AI (1)

birth of AI: 1956
Gestation (43-56):
"computational rationally”



automata theory, neural networks, checkers, theorem
"a
physical symbol system has the
proving.
necessary and sufficient means for
Shannon,
Von
Neumann, Newell and Simon,
generalTuring,
intelligent
action."
Minsky, McCarthy, Darmouth Workshop.
Great expectations (52-69):




computers can do more than arithmetic!
Physical symbol system
General Problem Solver (GPS), better checkers
LISP (LISt Processing language): AI programming
language
17
Minsky supervised a series of students
who chose limited problems that
appeared to require intelligence to solve.
A brief history of AI (2)

Microworlds: ANALOGY, blocks world
18
A brief history of AI (3)

A dose of reality (66-74):




ELIZA: human-like conversation.
limitations of neural networks, genetic algorithms,
machine evolution.
acting in the real world: robotics.
Knowledge-based systems (69-79):



All previous methods are weak methods !!
domain focus: experts systems vs. General
Problem Solvers.
DENDRAL(in Chemical experiment),
MYCIN(medical), XCON, etc.
19
A brief history of AI (4)

Commercial AI: the ‘80s boom (8090)




DEC’s R1 computer configuration program:
saving 40$ million in year
many expert systems tools companies
(mostly defunct): Symbolic, Teknowledge,
etc.
Japan’s 5th generation project: PROLOG.
limited success in autonomous robotics and
20
vision systems.
A brief history of AI (5)

The 90’s: specialization, quiet progress







neural networks, genetic algorithms
probabilistic reasoning and uncertainty
learning
planning and constraint solving
agents
autonomous robotics: NAV autonomous driving van,
crater exploration, robot soccer
IBM’s Deep Blue beats Kasparov!
21
State of the Art

Embedded AI: many use AI techniques without
saying it is AI!




Credit card approval (American Express)
Consumer electronics (fuzzy logic)
Healthy research in many areas: intelligent
agents, machine learning, man-machine
interfaces, etc.
More integrative view: acting in the real world
(robots, self diagnosing machines)
22
?
23