Functional requirements for a fuzzy expert system shell

ABSTRACT There are several general-purpose fuzzy expert system shells now in existence. However, experience with our system FLOPS indicates that several hurdles must be overcome before widespread successful use of the full power of such fuzzy reaoning systems can be confidently expected. Many of these hurdles confront non-fuzzy systems as well. Major developments needed include advances in appropriate fuzzy systems theory in several areas; abstract definitions of problem domain which transcend particular fields of application; standardization of blackboard architecture; standardized interfaces to blackboards and data bases? standardized interfaces to procedural languages; a sharp increase in production system computational efficiency; and software tools for development and debugging of rules. Theoretical developments badly needed include use of prior associations in fuzzy logic, i.e. dealing with the problems that have plagued Bayes' theorem ior over a hundred years; development of a theory of nonmonotonic fuzzy logic; further development of the theory of fuzzy numbers, especially in settheoretic operations and inequalities; narrowing the field of set-theoretic operations on fuzzy sets by tieing particular operators to their sphere of usefulness, and extending them to fuzzy sets of higher levels and types; creating theories of automated fuzzy reasoning, deductive and inductive; and refining a theory of fuzzy logic which takes account of the special role of expert system rules as replacement operators rather than logical propositions.