A methodology to incorporateformal methods in hybrid KBS verification

Abstract Thereis an increasing need for hybrid knowledge based systems(KBS) that accommodate more complex AI applications. Weconsider a hybrid KBS to be one that combinesobject-oriented, frame-based, and rule-based paradigms.Building hybrid KBSs requires formal methods forverification due to the possible interactions betweenobjects, inheritance, methods, and rules. The necessarycomplex verification for this type of KBS can be achievedwith formal specification and derivation because theyprovide unambiguous specifications and proof mechanisms.However, the complete use of formal methods is often tootime consuming and hence, impractical. Therefore, it isnecessary to find a balance between using formal methods andtraditional KBS techniques for development and verification.This paper presents a practical methodology to constructverifiable hybrid KBSs from formal specifications thatfollows traditional KBS development techniques. Themethodology involves prototyping a partial specificationexpressed in a formal language into an incrementallyverifiable KBS. The prototype is subjected to automatedanalysis and expert examination to suggest modifications andrefinements to the partial specification. The transformationof the partial specification to a prototype is consistentwith the refinement and transformation strategies found informal program derivation. Thus, the prototype isverifiable. The cycle of specification, prototype, andanalysis continues until the specification is believed to besufficiently complete to fully implement the KBS. Acombination of two formal specification languages isexamined within the methodology to allow for moreflexibility in the specification. We illustrate themethodology in an example hybrid KBS for oil reservoircharacterization.