Formal verification of AI software

Abstract

The application of formal verification techniques to Artificial Intelligence (AI) software, particularly expert systems, is investigated. Constraint satisfaction and model inversion are identified as two formal specification paradigms for different classes of expert systems. A formal definition of consistency is developed, and the notion of approximate semantics is introduced. Examples are given of how these ideas can be applied in both declarative and imperative forms.