EXPERT SYSTEM

1
EXPERT SYSTEM
CIS 430 ( Expert System ) Supervised By Mr.
Ashraf Yaseen Student name Ziad N. Al-Aabed
Student 20032174018
2
Chapter 16 Designing For Explanation

• ? Overview of this chapter
• Reasons for the requirement that the expert
• system should be Transparent
• 1. Able to explain their reasoning .
• 2. Justify their conclusions in manner that is
• Intelligible to users .

3
16.1 RULE-BASED EXPLANATION

• It has been recognized that automatic explanation
  requires access to a domain model .
• In other words , as a program needs a modicum of
  domain knowledge in order to acquire more
  knowledge , so a program need access to a
  representation of deep knowledge about the domain
  in order to explain its own behavior .
• Such knowledge is essential to bridge the gap
  between the-lower level implementation and the
  higher-level strategy that the system was
  pursuing .

4
16.1 RULE-BASED EXPLANATION(CONT.)

• There is indeed an intimate connection between
  the problems of knowledge elicitation and
  explanation , in that the way that knowledge is
  acquired and compiled has effect upon the way in
  which it can subsequently be used to explain
  system output .

5
16.1.1 MYCINs EXPLANATION system

• The explanation module of MYCIN system was
  automatically invoked at the end of every
  consultation .
• To explain how the value of a particular medical
  parameters was established , the module retrieved
  the list of rules that were successfully applied
  and printed them , along with the conclusions
  drawn .
• It also allowed the user to interrogate the
  system about the consultation , and ask more
  general questions.
• All question-answering facilities were based upon
  the systems ability to
• 1. Display the rule .
• 2. Record rule invocations
• 3. Use rule indexing .

6
16.1.1 MYCINs EXPLANATION
system(CONT.)

• A consultation with a backward chaining expert
  system involves a search through a tree of goals
  .
• Consequently, inquiries during a consultation
  fall into two types
• 1. Those that ask WHY a particular question was
  put .
• 2. Those that ask HOW a particular conclusion was
  reached .
• To Answer a WHY Questions , one must look up the
  tree to see what the higher goals the system is
  trying to achieve .
• To answer a HOW question , one must look down the
  tree to see what subgoals were satisfied to
  achieve the goal.
• Note that the explanation process can be
  considered as kind of tree traversal .

7
16.1.1 MYCINs EXPLANATION
system(CONT.)

• IF1) The stain of the organism is gramneg, and
• 2) The Morphology of the organism is
  rod,and
• 3) the aerobicity of the organism is
  aerobic
• THENThere is strongly suggestive evidence(0.8)
• that the class of the organism is
  entrobacteriaceae .

ENTROBACTERIACEAE
HOW??
WHY??
STAIN
MORPHOLOGY
AEROBICITY
8
16.1.1 MYCINs EXPLANATION
system(CONT.)

• It can simply cite the production rule which
  states that gram negative staining, in
  conjunction with various other conditions, would
  suggest that the class of the organism was
  entrobacteriaceae , and that the current goal was
  to identify the organism.
• MYCIN also maintain a record of the decision it
  makes , and uses this record to explain and
  justify its decision in response to HOW questions
  .
• In reply, MYCIN cites the rules that it applied ,
  its degree of certainty in that decision , and
  the last question asked . General questions can
  also be asked , these reference the rules without
  considering the state of the dynamic database
  with respect to a particular patient.
• For example What do you prescribe for
  pseudomonas infection?

9
16.1.2 EXPLANATION in MYCIN
derivatives EMYCIN and NEOMYCIN

• It is well known that the problems associated
  with understanding monitoring and correcting the
  behavior of an expert system multiply as the
  knowledge base increases .
• For example , it becomes more difficult to ensure
  that new rules are consistent with old ones,
  and to understand the flow of control in
  situations where large numbers of applicable
  rules may be in competition for the attention of
  the interpreter.
• MYCIN developed and elaborate MYCINs facilities
  to some extent . Thus EXPLAIN , TEST and REVIEW
  commands were provided as debugging aids for
  knowledge engineer . As in MYCIN , EXPLAIN worked
  by printing each rule that contributed to the
  conclusion, together with

10
16.1.2 EXPLANATION in MYCIN
derivatives EMYCIN and NEOMYCIN
(cont.)

• As in MYCIN , EXPLAIN worked by printing each
  rule that contributed to the conclusion, together
  with
• 1. The certainty factor .
• 2. The tally value .
• 3. The last question asked by the system .
• The use of meta-rules in MYCIN and EMYCIN , which
  was intended to
• 1. Make some of the control choices explicit and
• 2. Open the door to reasoning about problem
  solving strategy .

11
16.1.2 EXPLANATION in MYCIN
derivatives EMYCIN and NEOMYCIN
(cont.)

• Two rational reconstructions of the early
  Stanford work were begun at the end of the 1970s
  . One was the NEOMYCIN system . Which represented
  an attempt to take a more abstract approach to
  MYCIN style medical problem solving , based on
  epistemological and physical consideration .
• Thus NEOMYCIN was much more concerned with the
  simulation of human problem solving than MYCIMN .
  
• NEOMYCIN had the following basic organization
• 1. Strategic knowledge was separate out from the
  medical knowledge and encoded in meta-rules.
• 2. Diseases were organized taxonomically .

12
16.1.2 EXPLANATION in MYCIN
derivatives EMYCIN and NEOMYCIN
(cont.)

• Note both of the above kinds of knowledge were
  kept separate from the rules .
• (The basic approach is still based on HURISTIC
  classification
• but representation structure and controls use
  domain rules . )
• The domain rules are differentiated into 4
  classes
• 1. Casual rules .
• 2. Trigger rules .
• 3. Data rules .
• 4. Screening rule .

13
16.1.2 EXPLANATION in MYCIN
derivatives EMYCIN and NEOMYCIN
(cont.)

• Thus , in addition we elicit particular data from
  the user , there is a NEOMYCIN meta-rule which
  guides the asking questions .
• Such rules can then be cited as an explanation of
  why a particular question was asked .