[mod.ai] Checking Rule-Based Expert Systems

gjoly@CS.UCL.AC.UK ("G. Joly", Birkbeck) (03/08/87)

We are at the start of a project which is examining the area
of validation and verification of rule-based expert systems.
CHECK [1] and ONCOCIN [2] are the two major systems of which
we are aware. Are there any others? How isomorphic are rule-based
systems; can these and other techniques be applied in general?
Are any other (e.g. database) techniques applicable?

Thanks in advance for any pointers and information,
Gordon Joly,
Dept. of Computer Science,
Birkbeck College,
University of London.

ARPA: gjoly@cs.ucl.ac.uk
BITNET: UBACW59%uk.ac.bbk.cu@AC.UK
UUCP: ...{seismo,decvax,ucbvax}!mcvax!ukc!uk.ac.bbk.cs!gordon

[1] T.A.Nguyen, W.A.Perkins, T.J.Laffey and D.Pecora, "Checking
    an Expert Systems Knowledge Base for Consistency and Completeness".,
    IJCAI 1985, pp 375-378.
[2] M.Suwa, C.Scott and E.H.Shortliffe, "An Approach to Verifying 
    Completeness and Consistency in a Rule-Based Expert System",
    The AI Magazine, Fall 1982, pp 16-21.

mcguire@AERO2.AERO.ORG.UUCP (03/13/87)

We have been working in this area for a while.  In addition to checking
for completeness and consistency we analyze a rule-base for the
"effectiveness" of its information. It is possible for rules or
distinctions to appear to have meaning, but through faulty interaction
they wind up never influencing the answers the system gives.  This sort
of interference is unbounded in scope. We have developed propagation
style algorithms for finding ineffective information in simple types of
rule bases.

A paper on this work is almost ready for release. I can mail out copies
then.

  Roderick McGuire
  The Aerospace Corporation
  Box 92957
  Los Angeles, CA 90009

  ARPA: mcguire@aerospace.aero.org