Transcript Expert Systems Outline

Expert Systems
Outline:
Various Objectives in Creating Expert Systems
Integration of AI Techniques into Applications
MYCIN, AM
Shells for Rule-based Expert Systems
Engineering of Expert Systems
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
1
Reasons for Creating Expert
Systems
Real-world applications serve various roles:
1. Automating services formerly performed by
human experts.
2. Boosting the productivity of experts by creating
initial solutions that must later be refined.
3. Providing services that would have been
impractical to offer without AI (e.g., intelligent web
search).
4. Capturing and immortalizing corporate knowledge
in order to preserve it as a permanent asset.
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
2
Integration of AI Techniques into
Applications
Individual AI techniques usually do not solve the full set of
problems faced by an expert.
Different AI techniques are needed for different parts of the
system.
Typical components of an expert system are:
Knowledge acquisition, knowledge representation, inference,
language understanding.
Designing each of these so that they all work together can be
viewed as an engineering problem.
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
3
MYCIN
Developed by Edward Shortliffe in 1976 at Stanford
University in the artificial intelligence in medicine
group.
Carried on a text-based dialog with a physician.
Contained knowledge about infectious diseases
(bacterial infections), represented in IF-THEN rules with
certainty values.
Did not explicitly diagnose diseases but prescribed
combinations of medications to cover for the likely
causes.
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
4
AM
Developed by Douglas Lenat at Stanford in 1977.
Simulated the activities of a mathematician, exploring
number theory, coming up with conjectures and testing
them empirically.
Contained knowledge encoded as production rules.
Came up with an explicit representation for the concept
of prime numbers.
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
5
Shells for Expert Systems
In order to facilitate developing more expert systems,
the domain-independent parts of MYCIN were made
into a separate system: EMYCIN.
Emptied of its knowledge, Empty MYCIN was ready to
receive other knowledge.
It was the first expert system “shell.”
It provided a rule representation, inference engine
capable of working with certainty values as well as
logical statements, and a simple interface.
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
6
Details of an Expert System
Shell
Rule language:
2 types of rules: IF-THEN rules and query rules.
Inference engine:
Forward chaining
Backward chaining
Interface:
Poses questions and interprets answers.
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
7
Sample Rules
dot-pointillism-rule
IF (texture (? painting) dots (? cf))
THEN (style (? painting) pointillism)
WITH-CERTAINTY (* (? cf) 0.8)
normal rule
abdom-pain?
IF (health-problem (? cf))
THEN
(“Do you have abdominal pain?”
(abdominal-pain)
(knowledge-of-abd-pain))
WITH-CERTAINTY 1.0
query rule
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
8
Chaining
Forward Chaining:
Starting with the premises or given data, apply rules
to derive more and more consequences.
Backward Chaining:
Starting with the goal, determine what subgoals must
be achieved in order to attain the goal, and recursively
attack the subgoals, until the subgoals are just
premises or obvious conditions.
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
9
Engineering of Expert Systems
The 1980’s model:
Create a team consisting of a knowledge
engineer (KE) and a domain expert (DE);.
The KE works with the DE to build a
knowledge base consisting of rules.
The rules are debugged by continual
testing and refinement.
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
10
Pedagogical Agents
CAI: Computer-Assisted Instruction
(automated presentation and drill)
ITS: Intelligent Tutoring Systems
(use natural language dialog and
reasoning to enhance the experience)
PA: Pedagogical Agents
(enhances the tutor with additional agentlike qualities, such as an onscreen
presence -- e.g., a talking head)
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
11
Pedagogical Agent Architecture
Domain
Knowledge
Student
Model
Executive
Pedagogical
Knowledge
Interface
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
12
Pedagogical Agent Components
Domain Knowledge:
Knowledge about the subject to be taught and learned.
Pedagogical Knowledge:
Knowledge about teaching strategies, optimal order of presentation, how
to diagnose and fix misconceptions, etc.
Student Model:
A representation of the student’s beliefs, knowledge, motivations, learning
style, and possibly past experiences.
Interface:
A means of communicating with the student, possibly including natural
language understanding and generation, graphical display and sketch or
gesture recognition, etc.
Executive:
An engine that follows a pedagogical plan, invoking the resources of the
other components.
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
13
Expert Systems: Summary
Expert Systems perform services and/or boost productivity.
Expert Systems encapsulate and “immortalize” knowledge.
Expert Systems are delivery modules for AI techniques.
Expert Systems may be “brittle” if they lack common sense (as
they all-too-often do).
Expert Systems are often constructed using special
“knowledge-engineering” tools such as: ES shells and
expertise transfer programs.
CSE 415 -- (c) S. Tanimoto, 2004
Expert Systems
14