Introduction to Artificial Intelligence ITK 340, Spring 2012 For - - PowerPoint PPT Presentation

Introduction to Artificial Intelligence

ITK 340, Spring 2012

For Thursday

• Read Russell and Norvig, chapter 1
• Do chapter 1, exs 1 and 9

– There’s no single right answer for these. I’m looking for thoughtful multiple sentence responses.

Due Tuesday

• Send email to mecaliff@ilstu.edu from your

preferred email address

• Student information sheet

Course Info

• Instructor
• Textbook
• Syllabus
• Students

What is AI, anyway?

• Artificial Intelligence
• The artificial part is easy--we’re building

machines and computer programs

• Intelligence, however, is not well-defined
• Some things that require great intelligence

in human being are easy for computers

• Other things that are easy for most (all?)

humans are very difficult for computers

Categorizing the Definitions

• Acting or thinking

– Some definitions focus on thinking and reasoning, on the “mind” of the machine – Others focus on acting, on the behavior of the machine (whether there’s real thought behind it may not matter?)

• Human or rational

– Some definitions measure the computer against humans – Others focus on rationality--an ideal concept of intelligence

Thinking Humanly

• “The exciting new effort to make computers

think … machines with minds, in the full and literal sense” (Haugeland, 1985)

• “[The automation of] activities that we

associate with human thinking, activities such as decision-making, problem solving, learning …” (Bellman, 1978)

Thinking Humanly

• The cognitive modeling approach
• Interested not only in solving the problem,

but also in mimicking human thought processes

• This is where AI is most closely related to

cognitive science

Acting Humanly

• “The art of creating machines that perform

functions that require intelligence when performed by people” (Kurzweil, 1990)

• “The study of how to make computers do

things at which, at the moment, people are better” (Rich and Knight, 1991)

Acting Humanly

• The “Turing Test” Approach
• Focus is on how the system behaves, not

how it works inside

• Performance is measured against human

performance

• Biggest problem is the question of the value
• f the test--but we can’t pass it yet
• Development of practical systems

Thinking Rationally

• “The study of mental faculties through the

use of computational models” (Charniak and McDermott, 1985)

• “The study of the computations that make it

possible to perceive, reason, and act” (Winston, 1992)

Thinking Rationally

• The laws of thought approach
• Focus on logic--making correct inferences
• Problems

– Difficulty of formulating some types of knowledge logically – Solving in principal vs. solving in practice

• Strong contributions in reasoning and

knowledge representation

Acting Rationally

• “A field of study that seeks to explain and

emulate intelligent behavior in terms of computational processes” (Schalkoff, 1990)

• “The branch of computer science that is

concerned with the automation of intelligent behavior” (Luger and Stubblefield, 1993)

Acting Rationally

• The rational agent approach
• Instead of thinking the right way, focuses
• n doing the right thing
• More general than laws of thought
• More testable than comparing to human

behavior

• Approach taken by your text

What Do You Know?

• Examples of artificial intelligence in your

life?

• Can you name any of the areas of AI?

Foundations of AI

Foundations of AI

• Philosophy
• Mathematics
• Economics
• Neuroscience
• Psychology
• Computer engineering
• Control theory and cybernetics
• Linguistics

The Birth of AI

• McCulloch and Pitts(1943) theory of neurons

as competing circuits followed up by Hebb’s work on learning

• Work in early 1950’s on game playing by

Turing and Shannon and Minsky’s work on neural networks

• Dartmouth Conference

– Organizer: John McCarthy – Attendees: Minsky, Allen Newell, Herb Simon – Coined term artificial intelligence

Early Years

• What was the mood of the early years?

Early Years

• Development of the General Problem

Solver by Newell and Simon in 1960s.

• Arthur Samuel’s work on checkers in

1950s.

• Frank Rosenblatt’s Perceptron (1962) for

training simple networks

At MIT

• Marvin Minsky and John McCarthy
• Development of LISP
• SAINT: solved freshman calculus problems
• ANALOGY: solved IQ test analogy

problems

• SIR: answered simple questions in English
• STUDENT: solved algebra story problems
• SHRDLU: obeyed simple English

commands in the blocks world

Early Limitations

• Solved toy problems in ways that did not

scale to realistic problems

– Knowledge representation issues – Combinatorial explosion

• Limitations of the perceptron were

demonstrated by Minsky and Papert (1969)

Knowledge Is Power: The Rise of Expert Systems

• Discovery that detailed knowledge of the

specific domain can help control search and lead to expert level performance for restricted tasks

• First expert system was DENDRAL. It

interpreted mass spectogram data to determine molecular structure. Developed by Buchanan, Feigenbaum and Lederberg (1969).

Other Early Expert Systems

• MYCIN: Diagnosis of bacterial infection

(1975)

• PROSPECTOR: Found molybdenum

deposit based on geological data (1979)

• R1: Configured computers for DEC (1982)

AI Becomes an Industry

• Numerous expert systems developed in 80s
• Estimated $2 billion by 1988
• Japanese Fifth Generation project started in

1981.

• MCC founded in 1984 to counter Japanese.
• Limitations become apparent: prediction of

AI Winter

– Brittleness and domain specificity – Knowledge acquisition bottleneck

Rebirth of Neural Networks

• New algorithms (re)discovered for training

more complex networks (1986)

• Cognitive modeling
• Industrial applications:

– Character and hand-writing recognition – Speech recognition – Processing credit card applications – Financial prediction – Chemical process control

AI Becomes a Science

• Empirical experiments the norm
• Theoretical underpinnings are important
• The “See what I can do” approach is no

longer an acceptable method for doing research

• Some movement toward learning/statistical

methods.

Rise of Intelligent Agents

• Why?

Popular Tasks of Today

• Data mining
• Intelligent agents and internet applications

– softbots – believable agents – intelligent information access

• Scheduling applications
• Configuration applications

State of the Art

• Deep Blue beats Kasparov
• Sojourner, Spirit and Opportunity explore

Mars

• NASA Remote Agent in Deep Space I

explores solar system

• DARPA grand challenge: Autonomous

vehicle navigates across desert and then urban environment.

• Usable machine translation thru Google.

State of the Art

• iRobot Roomba automated vacuum cleaner,

and PackBot used in Afghanistan and Iraq wars

• Automated speech/language systems on

telephone.

• Fairly accurate speech recognition
• Spam filters using machine learning.
• Question answering systems automatically

answer factoid questions.

Views of AI

• Weak vs. strong
• Scruffy vs. neat
• Engineering vs. cognitive