Transcript CSCI 5582 Artificial Intelligence

CS 2710, ISSP 2610
Foundations of Artificial Intelligence
introduction
1
Welcome!
• Syllabus
2
What is Artificial Intelligence?
• 4 views on what makes an AI system (next slide; R&N 1.1)
• Depends on what matters more in the evaluation
• Reasoning vs. Behavior [diagram]
• Human performance vs. Rationality
– Human model
– Rational model
3
4 Views of AI
Think like humans
Think rationally
Act like humans
Act rationally
4
Acting Like Humans?
• Turing (1950)
• “Can machines think?”  “Can machines behave
intelligently?”
• Operational test for intelligent behavior: the Imitation
Game
5
Player C tries to determine which of A and B
are human/computer
6
Turing Test
• Suggested the major components of AI:
– NLP (communicate in language)
– Knowledge representation (to store what it knows)
– Automated reasoning (use knowledge to answer questions and
draw conclusions)
– Machine learning (adapt to new circumstances; prediction)
– Vision
– Robotics (manipulate objects and move around)
7
Turing Test
• Not how we evaluate AI methods; Why? [board + next
slide]
8
Sample Eliza Dialogue
Joseph Weizenbaum ~1965
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
young woman: Men are all alike.
eliza: In what way?
young woman: They're always bugging us about something specific or other.
eliza: Can you think of a specific example?
young woman: Well, my boyfriend made me come here.
eliza: Your boyfriend made you come here?
young woman: He says I'm depressed much of the time.
eliza: I am sorry to hear you are depressed.
young woman: It's true. I'm unhappy.
eliza: Do you think coming here will help you not to be unhappy?
young woman: I need some help, that much seems certain
.eliza: What would it mean to you if you got some help?
young woman: Perhaps I could learn to get along with my mother
.eliza: Tell me more about your family
.young woman: My mother takes care of me
.eliza: Who else in your family takes care of you?
young woman: My father
…
9
Thinking Like Humans?
• For this approach, need a way to determine how humans
think
– Introspection (rejected by researchers as too subjective)
– Psychological experiments: predicting (based on cognitive
theory) and then testing, behavior of human subjects; matched
predictions give evidence for the theory
– Brain imaging – observing the brain in action
• Cognitive Science:
– Express a theory as a computer program; input/output behavior
matching human behavior is evidence for the theory
– Computer models from AI and experimental techniques from
Psychology; also
– Neurophysiological evidence incorporated into computational
models, e.g. vision
10
Thinking Like Humans
• AI and Cognitive Science are now largely distinct research
areas [board]
11
Thinking Rationally?
12
The “Laws of thought” approach
•
Logicist tradition:
•
Problems:
•
–
–
–
Logic: notation and rules of derivation for thoughts
Aristotle: what are correct arguments/thought processes?
Direct line through mathematics, philosophy, to modern AI
–
–
–
Not all intelligent behavior is mediated by logical deliberation
It is difficult to express informal knowledge in logic
It is not sufficient:
• Need a search process to go down fruitful reasoning paths
• logical systems tend to do the wrong thing in the presence of uncertainty
Logic is important in AI; but a pure logicist approach (early AI history)
to intelligence is not effective
• That leaves us with ….
13
Acting Rationally: Our Basic Framework
• Getting computers to do the right thing based on their
circumstances and what they know.
– Irrational != insane; irrationality is sub-optimal action
– Rational != successful; the most rational action may not succeed
due to some circumstance beyond our control or due to
incomplete knowledge
– Make the best choice, given the options
• Rational agents [board]
14
19401950: Early days
1943: McCulloch & Pitts: Boolean circuit model of brain
1950: Turing's “Computing Machinery and Intelligence”
1950—70: Excitement: Look, Ma, no hands!
1950s: Early AI programs, including Samuel's checkers program, N
ewell & Simon's Logic Theorist, Gelernter's Geometry Engine
1956: Dartmouth meeting: “Artificial Intelligence” adopted
1965: Robinson's complete algorithm for logical reasoning
1970—88: Knowledgebased approaches
1969—79: Early development of knowledgebased systems
1980—88: Expert systems industry booms
1988—93: Expert systems industry busts: “AI Winter”
1988—: Statistical approaches
Resurgence of probability, focus on uncertainty
General increase in technical depth
15
Agents and learning systems… “AI Spring”?
AI applications
AI techniques are used in many common applications; just a sample
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
Intelligent user interfaces
Search Engines
Spell/grammar checkers
Context sensitive help systems
Medical diagnosis systems
Regulating/Controlling hardware devices and processes (e.g, in automobiles)
Voice/image recognition (more generally, pattern recognition)
Scheduling systems (airlines, hotels, manufacturing)
Error detection/correction in electronic communication
Program verification / compiler and programming language design
Web search engines / Web spiders
Web personalization and Recommender systems (collaborative/content
filtering)
Personal agents
Customer relationship management
Credit card verification in e-commerce / fraud detection
Data mining and knowledge discovery in databases
Computer games
16
What to expect
Abstractive thinking/imagination sometimes needed
Extreme range of problem domains (as we just saw on the sample of
applications). We need to look for frameworks that apply to a
hugely diverse range of problem domains. Abstract distinctions
abound.
Real problem domains are often so complex we need to work with
simpler ones, and imagine what would be needed in a realistic
domain
Not a definitive answer about which method is best; depends on
the problem!
AI problems are those that we really don’t know how to solve.
Otherwise, we would use a direct solution (and it would not be
considered AI anymore)
Real AI systems are often mixtures of various algorithms/techniques,
experimentally determined
17
Course Topics
Four major areas:
• Problem solving and search.
– Formulating a search problem, uninformed and informed search;
constraint satisfaction, optimization, and adversarial search.
• Logic and knowledge representation
– First-order logic; reasoning; knowledge representation schemes
• Planning
– Situation calculus, STRIPS, Partial-order planning, GraphPlan
and SAT planners
• Uncertainty and Learning
– Modeling uncertainty, Bayesian belief networks, decision
theory, classification, density estimation
18
Wrap Up
• Chapter 1:
– You will not be tested on Sections 1.2 and 1.3
(history; foundations). But it’s interesting!
– Be able to explain the different possible
approaches to AI and why AI has settled on the
rational action approach
• Chapter 2:
– Will be covered on homework 1; Any explicit
exam question will be related to its coverage on
homework 1
19