Introduction to AI - Sixth Lecture

Post on 19-Jun-2015

572 views 0 download

Tags:

Transcript of Introduction to AI - Sixth Lecture

Introduction to AI6th Lecture

1980’s – Expert Systems

Wouter Beekme@wouterbeek.com

13 October 2010

Part I

1980’s, The decade of expert systems

1980’s, the decade of expert systems

0Funding in AI returned.0Applications become not wider but deeper.

0 Still within a very narrow domain.0 But no longer toy problems.

0Solutions for the common-sense knowledge problem were found.

0Successful applications that meet expectations are realized.

1982, 5th generation project

0FGCS, Fifth Generation Computer Systems project.0 Japan's Ministry of International Trade and Industry.

Computer generations:00th generation: 500 B.C., mechanical gears.01st generation: 1940’s, vacuum tubes.02nd generation: 1950’s, transistors.03rd generation: 1960’s, integrated circuits (ICs).04th generation: Microprocessors.

1983-1993, Reactions to 5th generation project

01982, MCC, Microelectronics and Computer Technology Corporation0 American computer manufacturers cooperate on

research.01983-1987, Alvey

0 British government project.01983-1993, SCI, Strategic Computing Initiative

0 DARPA’s response to FGCS0 $ 1.000.000.000

0 Remember the Sputnik launch in 1959, research funding is often reactive!

Expert systems

Main characteristics:1. Provides expert-level solutions to complex problems.

0 Solutions are generated in a heuristic way.

2. Give solutions that are understandable.0 Solutions are couched in qualitative terms (i.e.

concepts).

3. Flexible to accommodate new knowledge.0 Decoupling of reasoning and knowledge.

[1] Algorithm VS heuristic

Properties of algorithmic problem-solving:0Guaranteed to find a

solution.0The found solution is the

correct one.0The solution is found in

finite time.

Properties of heuristic problem-solving:0Probable to find a

solution.0The found solution is an

acceptable one.0The solution is found in

practical time.

[2] Quantitative VS qualitative

Quantitative reasoning:0Numerical data0Data-processing0Mathematical0Syntactical

Qualitative reasoning:0Conceptual data0Symbol-processing0Logical0Semantical

[3] Knowledge VS reasoning

0Mutilated chessboard problem0Suppose a standard 8x8

chessboard has two diagonally opposite corners removed, leaving 62 squares. Is it possible to place 31 dominoes of size 2x1 so as to cover all of these squares? [Gamow & Stern 1958]

0Representation (partially) obsoletes reasoning / knowledge (partially) captures reasoning.

[3] Knowledge engineering

0 “KE is an engineering discipline that involves integrating knowledge into computer systems in order to solve complex problems normally requiring a high level of human expertise.” [Feigenbaum & McCorduck 1983]

0 “Mapping an expert’s knowledge into a program’s knowledge base.” [Buchanan & Shortliffe 1983, p. 5]

Cyc (1984-present)

0Assembles a comprehensive ontology and knowledge base of everyday common sense knowledge.

0Allowing machines to overcome the common-sense problem.

0Started in 1984 by Douglas Lenat at MCC.0Currently developed by Cycorp.0OpenCyc is the open-source spin-off.

0Wordnet, 1985-present, lexical ontology

Part II

MYCIN

General and specific knowledge

0Production rules represent general knowledge.0Clinical parameters represent specific knowledge:

0 object0 attribute0 value

0Monitoring method: match the conditions in a production rule with clinical parameters.

0Find-out method:0 Infer unknown clinical parameters by using other

production rules.0 Query the user for unknown clinical parameters.

Production rules / Horn clauses0A literal is an atomic formula, or the negation of an atomic

formula.0 E.g. p, q, r, , walks(John), loves(John, Mary),

0A clause is a disjunction of literals.0 E.g. , ,

0A Horn clause is a clause with at most one positive literal.0A definite clause is a Horn clause with at least one positive

literal.0 E.g.

0Definite clauses are logically equivalent to implications.0 E.g.

Reasoning with production rules

0Backward chaining: for each conclusion, find the matching parameters.

0Forward chaining: for a set of parameters, find the conclusions that follow.

Explanation / transparency box

0The trace of a reasoning task is also an explanation.0 What rules were used to derive the result?0 In which order were the rules applied?0 Which parameters were used?

0The machine reasons like a human.

Knowledge hypothesis

0Expresses the relation between:0 the complexity of the world0 the required functionality0 the role of knowledge

0To achieve a high level of problem-solving competence, a symbol system must use a great deal of domain-specific, task-specific, and case-specific knowledge.