An approach to enhancing the maintainability of expert systems

Abstract

The task of maintaining expert systems has become increasingly difficult as the size of their knowledge bases increases. To address this issue, a unified AI programming environment (CLASP) has been developed; this environment tightly integrates three AI programming schemes: the term subsumption languages in knowledge representation the production system architecture, and methods in object-oriented programming. The CLASP architecture separates the knowledge about when to trigger a task from the knowledge about how to accomplish a given task. It also extends the pattern matching capabilities of conventional rule-based systems by using the semantic information related to rule conditions. In addition, it uses a pattern classifier to compute a principled measure about the specificity of rules. Using a monkey-bananas problem, the authors demonstrate that an expert system built in CLASP is easier to maintain because the architecture facilitates the development of a consistent and homogeneous knowledge base, enhances the predictability of rules, and improves the organization and reusability of knowledge.<>