Expert Systems

1
Expert Systems
Lila Rao Graham
2
Definition

• A program which achieves human-like performance
  for problem solving within a limited domain of
  expertise.
• Usually constructed by a knowledge engineer
• Engineer interacts with human expert to determine
  how they solve the particular tasks in the domain
  of interest.

3
The Human Expert vs The Expert System Problem
Solving
Long Term Memory Domain Knowledge
Case Facts
Reasoning
Conclusions
Working Memory Cases, Inferences, Conclusions
The Human Expert
User Interface (NL?)
Knowledge Base Domain Knowledge
Explanation Facility
Why? How?
User
Short Term Memory Cases,Inferences,Conclusions
Case Facts
Inference Engine
Conclusions
Working memory Cases, Inferences, Conclusions
The Expert System
4
Relationship between AI and Expert Systems
Other Areas
AI
ES
GPS
5
Why have an Expert System?

• Increased Availability
• Permanence
• Easy Duplication
• Reduced Cost
• Reduced Danger
• can be used in hostile environments
• Speed/Performance/Reliability
• variable with expert, consistent with ES
• used with large amounts of data
• Assist Expert
• Explanation Provided
• Multiple Expertise

6
Characteristics of ES

• Knowledge is separate from control.
• Expert system shell
• Reasons with symbols
• Uses heuristics
• Rules of thumb
• Allows inexact reasoning
• vague information
• Provides Explanations
• Problem is solvable by a human expert

7
Considerations

• Personnel
• Knowledge Engineer
• Programming
• Communications
• Domain Expert
• A real one that performs the desired task
  successfully
• Are the experts hard to come by?
• Is the expert always busy?
• Financial
• Is the problem worth solving?
• Will the value remain with time?
• Is there a need to preserve/duplicate the
  expertise?
• Technical
• Could an algorithm work?
• Does the problem require symbolic knowledge?
• Is the problem primarily cognitive?

8
Example of an ES

• XCON
• Originally developed by J. McDermott at CMU
• Funded by DEC
• Later developments done by DEC in-house
• System for configuring DECs VAX computers
• Ensure consistency of order
• Ensure completeness of order
• Gives a lay-out for the components
• XCON is used for all orders received by DEC.
• Achieves an accuracy of around 95.
• Solves 10,000 problems per annum

9
Typical Applications for ES

• Interpretation of data
• Diagnosis
• Scheduling and Planning
• Design
• Process Control

10
Issues in Expert Systems

• Knowledge Representation
• Uncertainty
• Explanation
• Knowledge Acquisition

11
Knowledge Representation

• The quality of the expert system relies on the
  quality of the knowledge coded up.
• How is this knowledge represented?
• Must be done formally
• Language must be unambiguous
• Language must have a well defined syntax and
  semantics

12
Knowledge Representation Languages

• Logic
• Production Rules
• Semantic Nets
• Frames and Scripts
• Etc.

13
Uncertainty

• Domain theory may be imprecisely defined (or
  wrong).
• Experts may use inexact methods
• exact methods may not be known
• exact methods may be impractical
• Probability is used to represent uncertainty.
• Data may be imprecise or unavailable.
• Records with fields not filled

14
Explanation

• Because of uncertainty, expert systems need to
  justify their conclusions.

15
Knowledge Acquisition (KA)

• Getting the knowledge from the knowledge source
  to the computer system.
• E.g. of sources experts, books, technical
  reports, databases, forms,etc.
• Knowledge Elicitation (KE)
• Getting the knowledge from the expert.

16
Difficulties in Acquiring Knowledge

• Practical Problems
• Unavailability of the expert
• Hostility of the expert
• Cognitive Problems
• Expert may be unaware of their own rules of thumb
• Humans are not conscious of their activities and
  mental processes.
• Expert may give textbook knowledge, rather than
  the rules of thumb they use.
• Expert may not be very articulate.
• N.B. Similar problems in the early stages in
  traditional Software Engineering.