Building an expert system has traditionally been thought of as a knowledge-intensive activity, regardless of the approach taken to design the system. In a shallow-model approach, much time is spent by the knowledge engineers in interviewing experts to deduce expert response in a given situation and encoding rules that capture this compiled expertise. In a deep-model approach, that time is spent by the domain expert in understanding and explaining the causality of events to the knowledge engineer, who then has the task of devising a representation for the causal model. We present a declarative paradigm for designing an expert system in which compiled, heuristic knowledge and causal knowledge can be integrated in a system effectively. In the declarative style of building expert systems, there is a shift in emphasis placing domain knowledge in structured objects rather than in rules. Fewer rules are then needed to express the equivalent domain knowledge. We show that this restructuring of the domain knowledge enables knowledge acquisition and verification to be considerably simplified. Additional benefits of the declarative style are the increased clarity, maintainability and flexibility of the knowledge base. These principles are illustrated in a system used in a generic classification type problem solving shell. An interactive interface to the system has been built enabling a domain expert to change and extend the capabilities of the system without the assistance of a knowledge engineer.

Date Published: 29 March 1988
PDF: 6 pages
Proc. SPIE 0937, Applications of Artificial Intelligence VI, (29 March 1988); doi: 10.1117/12.947024

Published in SPIE Proceedings Vol. 0937:
Applications of Artificial Intelligence VI
Mohan M. Trivedi, Editor(s)