1. CSNB234 ARTIFICIAL INTELLIGENCE
Chapter 7
Introduction to Expert Systems
(Chapter 8, Textbook)
(Chapter 3 & Chapter 6, Ref. #1)
Instructor: Alicia Tang Y. C.

2. EXPERT SYSTEM (ES) Definition
ES is a set of computer programs that can advise, consult, diagnose, explain, forecast, interpret, justify, learn, plan and many more tasks that require ‘intelligence’ to perform.

3. An “expert system” is defined as “A computerized clone of a human expert”
(Definition taken from Oxford Science Publication)

4. EXPERT SYSTEMS: CHARACTERISTICS
Perform at a level equivalent to that of a human expert.
Highly domain specific.
Adequate response time
Can explain its reasoning.
It can propagate uncertainties and provide alternate solutions through probabilistic reasoning or fuzzy rules .

5. AN EXPERT AND A SHELL ES SHELL EXPERT:
A special purpose tool designed for certain types of applications in which user supply only the knowledge base (e.g. EMYCIN)
It isolates knowledge-bases from reasoning engine
Hence software portability can be improved
EXPERT:
An expert in a particular field is a person who possess considerable knowledge of his area of expertise
Domain-specific

6. Shell Concept for Building Expert SystemsKB
e.g. rules
Consultation
Manager
KB Editors
& debugger
shell
Inference
Engine
Explanation
Program
KBMF

7. Comparison (I) Expert Systems Conventional Systems
knowledge base is separated from the processing (inference) mechanism
program may make mistake (we want it to make mistake!)
explanation is part of most ES
the system can operate with only a few rules (this is called fast prototyping)
changes in the rules are easy to accomplish
Conventional Systems
information & its processing are combined in one sequential program
programs do not make mistake (but programmers do make it)
the system operates only when it is completed
execution is done on a step-by-step basis (i.e. it is algorithmic)

8. Comparison (II) Expert Systems
Conventional Systems
changes in programs are tedious
do not usually explain why or how conclusions were drawn
need complete information to operate
E__________ is a major goal
easily deal with q_________ data
Expert Systems
can operate with incomplete or uncertain information
execution is done by using heuristic based on rules of logic
E___________ is the major goal
easily deal with q______ data

9. RIGHT TASKS FOR RIGHT SYSTEMS
Facts that are known
Expertise available but is expensive
Analyzing large/diverse data
E.g. Production scheduling & planning, diagnosing and troubleshooting, etc.

10. Generic Categories of Expert Systems (1)
Interpretation
inferring situation descriptions from observation
Prediction
inferring likely consequences of given situations
Diagnosis
inferring system malfunctions from observations

11. Generic Categories of Expert Systems (2)
Design
configuring objects under constraints
Planning
developing plans to achieve goals
Repair
executing a plan to administer a prescribed remedy
Other categories include: monitoring,
debugging, control, instruction

12. BENEFITS OF EXPERT SYSTEMS (I)
Expertise in a field is made available to many more people (even when human experts are not around in the company).
Top experts’ knowledge gets saved rather than being lost, when they retire or should they have resigned.
Facts are stored in a “Systematic” way.
& Easy to keep on adding new knowledge on it
Allows human experts to handle more complex problems rapidly and reliably.

13. Early EXPERT SYSTEMS (70s – mid 80s)
MYCIN
USES RULE-BASED SYSTEM
GOAL-DRIVEN
RULES INCORPORATED IN MYCIN REFLECTED UNCERTAINTY ASSOCIATED WITH KNOWLEDGE
CERTAINTY FACTOR WAS USED TO DERIVE CONCLUSION
DENTRAL
WAS DEVELOPED IN STANFORD UNIVERSITY TO ANALYZE AND INTERPRET CHEMICALS AND THEIR MOLECULAR STRUCTURES
DEVELOPERS INCLUDE JOSHUA LEDERBERG (NOBEL PROZE WINNER IN GENETICS)
EXPERT’S “KNOW-HOW” ARE EXPRESSED IN RULES; RULE-OF-THUMB TECHNIQUE IS USED
PROSPECTOR
KBS TO INTERPRET GEOLOGIC DATA FOR MINERALS EXPLORATION
INCOPORATED BAYES THEOREM (PROBABILISTIC REASONING APPROACH)

14. Early EXPERT SYSTEMS (70s – mid 80s)
XCON
RULE-BASED SYSTEM, DATA-DRIVEN
REVEAL
FUZZY LOGIC USED
CENTAUR
RULES AND FRAMES-BASED SYSTEM
HEARSAY I
– for speech recognition

15. Characteristics common to early ES
Could perform at a level equivalent to human experts
Large amount of domain specific knowledge
Rule-based systems: knowledge incorporated in the form of production rule

16. Popular Expert Systems
Application Domain
Electronics : helps in VLSI/ULSI design
Law : system serves as an auditor
Manufacturing : in production & process controls
Medicine : illness diagnosis
Chemistry : synthesis planning

17. EXPERT SYSTEMS: LIMITATIONS
SYSTEMS ARE TOO SUPERFICIAL
RAPID DEGRADATION OF PERFORMANCE
INTERFACES ARE STILL CRUDE
INABILITY TO ADAPT TO MORE THAN ONE TYPE OF REASONING
E.g. either forward or backward and not both

18. The Resistors Domain experts Non-experts
Other information technologists such as DBA, Network specialists
Users
Management
Troublemakers

19. STRUCTURE OF AN EXPERT SYSTEM
Consultation Environment
(Use)
Development Environment
(Knowledge Acquisition)
User
Expert
Facts of
the Case
Recommendation,
Explanation
User Interface
Explanation
Facility
Knowledge
Engineer
Inference Engine
Facts of
the
Case
Knowledge
Acquisition
Facility
Working Memory
Knowledge
Base
Domain Knowledge
(Elements of Knowledge Base)

20. Figure: Key components of an Expert Systems

21. Explanation Facility Why need it? Common types
It provides sound reasoning besides quality result.
Common types
“How” a conclusion was reached
“Why” a particular question was asked

22. Importance of Explanation
It can influence the ultimate acceptance of an Expert System.
Use as a debugging tool.
Use as a component of a tutoring system.
Who needs explanation?
Our clients : to be convinced to purchase.
Knowledge Engineer: to check if all specifications are met?

23. Approaches Used (1) Canned Text
prepared in advance all questions and answers as text
system finds explanation module and displays the corresponding answer
problem:
difficult to secure consistency
suitable for slow changing system only

24. Approaches Used (2) Paraphrase Tree Traverse to answer WHY
look up the tree
to answer HOW
look down the tree to see sub goals that were satisfied to achieve the goal

25. Most expert systems are rule-based.
Rule-based Systems
In expert system development, a tool is used to help us to make a task easier. The tool for machine thinking is the Inference Engine.
Most expert systems are rule-based.

26. FACTS AND RULES (revision)
A mammal is an animal
A bird is an animal
Arthur is a man
Ben drives a car
Catherine has blue eyes
RULES :
If a person has RM1,000,000 then he is a millionaire.
If an animal builds a nest and lays eggs then the animal is a bird.

27. Examples of rules: Rule 1: if you work hard and smart
then you will pass all examinations
Rule 2: if the food is good
then give tips to the waiter
Rule 3: if a person has US1,000,000
then he is a millionaire

28. Forward Chaining and Backward Chaining
These are methods for deducing conclusions. The former predicts the outcome (conclusion) from various factors (conditions) while the latter could be very useful in trying to determine the causes once something has occurred.
Detailed description and working examples of rule-based systems and their reasoning methods will be dealt separately in other chapters.

29. Chaining Systems Forward
it predicts the outcome from various factors (conditions)
Backward
it could be very useful in trying to determine the cause (reason) once something has occurred

30. Inference Strategies (I)
Conclusion
(Goals)
Input
Data
Few Items
(For Example, User
Specifications for
a Computer
System)
Many Possibilities
(For Example, a
Computer
Configuration)
(a) Forward Chaining: IF - Part Matches Shown

31. Inference Strategies (II)
Input
Data
Conclusion
(Goals)
Extensive;
Much of the Data
Obtained by the
System Querying
the User (For
Example,
Investor’s Profile)
Few Possibilities
(Known in Advance
((For Example,
Investment Options)
(b) Backward Chaining: THEN - Part Matches Shown

32. Exercise #1
You have seen what tasks are “just right” for ES and now you are required to answer the following question:
List a “Too hard” task for computers and explain briefly why they are considered ‘too difficult’.

33. Nice to know…

34. RULE-BASED VALIDATION
There are essentially 5 types of inconsistency that may be identified, these are:
Redundant rules
Conflicting rule
Subsumed
Unnecessary Premise(IF) Clauses
Circular rules

35. REDUNDANT RULES Rule 1 IFA = X AND B= Y THEN C = Z Rule 2
IF B=Y AND A=X THEN C=Z AND D=W
Rule 1 is made redundant by rule 2.

36. CONFLICTING RULES Rule 1 IF A = X AND B= Y THEN C = Z Rule 2
IF A=X AND B=Y THEN C=W
Rule 1 is subsumed by rule 2 thus becomes unnecessary.

37. SUBSUMED RULES Rule 1 if A = X AND B= Y THEN C = Z Rule 2
if A=X THEN C=Z
to be revised.

38. UNNECESSARY PREMISE (IF) CLAUSES
Rule 1
IF A = X AND B= Y THEN C = Z
Rule 2
IF A=X AND NOT B=Y THEN C=Z
Remove B=Y and NOT B=Y to have just one rule.

39. CIRCULAR RULES Rule 1 IF A = X THEN B = Y Rule 2
IF B=Y AND C=Z THEN DECISION=YES
Rule 3
IF DECISION=YES THEN A = X
Restructure these rules !