Download CS440 - Introduction to Artificial Intelligence

Textbook
CS440 - Introduction to
Artificial Intelligence
 
Textbook: S. Russell and
P. Norvig. Artificial
Intelligence: A Modern
Approach. Prentice Hall,
2003, 2nd edition.
1
Course website
 
 
2
Course staff
Web Site: www.cs.colostate.edu/~cs440
What you can find there:
 
 
 
 
Instructor: Asa Ben-Hur
 
All slides (after or shortly before class)
All homework assignments
Important announcements
 
 
 
Office: 448
Office Hours: TBD, or by appt.
Email: asa at cs dot colostate dot edu
Teaching Assistant:
 
Doug Hains
3
Workload
 
 
 
4
Grading
Written/programming assignments (~6)
Programming in python
Project
Exams (midterm/final)
 
 
 
5
Exams: midterm: 15%
final: 25%
Assignments: 30%
Project: 30%
6
1
Course Outline
 
Search: How to explore the space of potential
solutions to a problem.
Logic: How to make inferences from stored/
learned knowledge.
Learning - How can a computer learn from data.
 
+ brief discussion of other topics
 
 
What is AI?
Definitions according to several AI
textbooks….
7
AI: Think Like Humans
 
 
8
AI: Think Like Humans
“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
 
How do humans think?
 
 
 
Requires understanding of brain activity (cognitive
model).
Cognitive science vs. cognitive neuroscience.
The available theories do not explain anything
resembling human intelligence!
9
AI: Act Like Humans
 
 
10
The Turing Test
“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
 
When does a system behave intelligently?
Turing (1950) Computing Machinery and Intelligence
  Operational test of intelligence.
 
 
 
11
Test still relevant now, yet might be the wrong question.
Requires the successful application of major fields of AI:
knowledge representation, reasoning, natural language
processing, machine learning
12
2
Role of the Turing Test
 
 
 
To articulate a performance goal.
To avoid defining intelligence.
How significant is the Turing test?
 
 
 
 
Eliza
Eliza, the computer therapist (J. Weizenbaum at
MIT, 1966).
How would you administer it?
What would you ask?
Would we all agree on the outcome?
How close are we?
13
14
AI: Think Rationally
 
Thinking rationally
“The study of mental faculties through the
use of computational models” Charniak and
McDermott, 1985
 
The logicist approach.
 
 
Logic: notation and rules of derivation for thoughts
Problems:
 
 
Not all intelligent behavior is mediated by logical
deliberation
What is the purpose of thinking? What thoughts
should I (bother to) have?
15
16
AI: Acting Rationally
 
 
 
“A field of study that seeks to explain and emulate
intelligent behavior in terms of computation
processes” Schalkoff, 1990
“The branch of computer science that is concerned
with the automation of intelligent behavior” Luger
and Stubblefield
Rational behavior: doing the right thing
 
 
What is AI?
Definitions of artificial intelligence:
The “right thing” is that which is expected to maximize goal
given the available information.
 
Our focus: rational agents, and how to construct
them.
Systems that think
rationally
Systems that act like
humans
Systems that act
rationally
The definitions vary by:
 
 
17
Systems that think like
humans
Thought processes vs. action
Judged according to human standards vs. success according to an
ideal concept of intelligence: rationality.
18
3
Tools
 
Lisp
 
 
Application areas
 
The traditional AI language
 
Prolog
 
 
Logic programming: fundamentally different!
 
Planning: What to do when.
Computer Vision: Seeing is knowing.
Speech Recognition: What words are
spoken.
Speech Understanding (NLP): What do the
words mean.
19
CSU AI Faculty
 
Darrell Whitley
Adele Howe
Ross Beveridge
Bruce Draper
Charles Anderson
 
Asa Ben-Hur
 
 
 
 
20
AI in everyday life
Genetic algorithms
Planning
Computer vision (face recognition)
Computer vision (biomimetics)
Machine learning /
computational neuroscience
Applications of machine
learning in bioinformatics
AI systems we use on a daily basis:
 
 
 
 
 
 
Scanning documents (OCR).
Talking with customer service or your cellphone
(voice recognition, NLP).
Applying for a loan.
Online map services.
Computer games (chess etc.)
Web searches
21
22
AI Systems
 
The Sojourner Mars rover (1996)
Deep Space 1
 
Deep Blue
 
DARPA grand challenge 2005: a 130 mile race of
driverless cars in the mojave desert.
 
 
 
AI planner controls space probe - NASA 1999.
Defeats Kasparov, Chess Grand Master - IBM 1997
Sandstorm
The entries from CMU
pictures from: http://www.grandchallenge.org/
The Sojourner rover
23
Highlander
24
4
Mundane Versus Expert Tasks
 
Mundane
 
 
 
 
Expert
 
 
 
 
The Stanford winning entry: Stanley
Identifying objects in an image
Answering a question
Picking up an arbitrary object
Chess
Medical diagnosis
Configuring computer hardware (circuit layout)
Special purpose robots
picture from http://stanfordracing.org
25
AI: How Close Are We to the Goal?
 
 
 
 
 
Military systems, space probes…
 
Machine learning:
 
 
 
Medical diagnosis, fault detection…
Planning:
 
 
Foundations of AI
Expert systems:
 
 
Philosophy: Logic, reasoning, rationality.
Mathematics: Logic, computability, tractability
Psychology: understanding how humans think and act.
Neuroscience: how do brains process information?
Economics: theory of rational decisions, game theory.
Computer Engineering: building the hardware and software
that make AI
Speech recognition, adaptive control…
 
Computer Vision:
 
26
 
Manufacturing, medicine, security…
Linguistics: how to deal with language
…
27
Foundations of AI: Mathematics
 
 
 
 
Logic has limits
The leap to formal science needs
 
Computation
Logic
Probability
 
Algorithms
Symbolic Inference
Quantifying “Maybe”
 
 
Hilbert: “Truth of any logical proposition regarding
natural numbers … may be proven using logic.”
Godel’s Incompleteness Theorem:
In 1931, the Czech-born mathematician Kurt Gödel demonstrated that within any given branch of
mathematics, there would always be some propositions that couldn't be proven either true or false
using the rules and axioms ... of that mathematical branch itself. You might be able to prove every
conceivable statement about numbers within a system by going outside the system in order to come
up with new rules an axioms, but by doing so you'll only create a larger system with its own
unprovable statements. The implication is that all logical system of any complexity are, by definition,
incomplete; each of them contains, at any given time, more true statements than it can possibly
prove according to its own defining set of rules. *
Key Mathematical Questions:
 
28
Just how far does logic take us?
What is computable?
*From
Jones and Wilson, An Incomplete Education,
http://www.miskatonic.org/godel.html
29
30
5
Beware of combinatorics!
 
 
“Solvable in Principle”: little help in practice
Beware of intractability…
 
 
 
Foundations of AI: Neuroscience
Use ideas from neuroscience to design
computer architectures that “learn”.
Considering all possibilities often leads to correct,
but intractable, algorithms.
Intractable means exponential time to solution.
NP-Complete Problems
 
Class of intractable problems
One View: AI proposes imperfect, but practical,
algorithms to solve NP-Complete problems.
Artist’s depiction of a neural
network
31
A Brief History of AI
 
The beginning of AI
 
Early enthusiasm
A dose of reality
 
 
1952 - 1969
1966 - 1974
 
- Progress was slower than expected: Unrealistic predictions.
- Some systems lacked scalability: combinatorial explosion
- Fundamental limitations on techniques and representations: Minsky and
Papert (1969) Perceptrons.
 
Knowledge is power
 
 
 
1969 - 1979
- Domain-specific systems - expert systems
- MYCIN: medical diagnosis of blood infections using ~450 rules.
 
 
AI becomes an industry
AI becomes a science
http://en.wikipedia.org/wiki/Neural_network
32
Primary Areas of AI
1943 - 1956
1950: Alan Turing’s “Computing Machinery and Intelligence”
The term “AI” proposed at the Dartmouth workshop (1956)
 
http://www.bitspin.net/images/neuron.jpg
Abstraction as an artificial
neural network
 
 
1980 – 1988
1988 - …
 
Knowledge Representation
Automated Reasoning
Game Playing
Planning
Machine Learning
Search and Optimization
Computer Vision
Robotics
Natural Language Processing
- Neats vs. scruffies
33
34
6