PRINCIPLES OF DESIGN FOR KNOWLEDGE ACQUISITION.

The authors explore the hypothesis that the design of a knowledge system architecture--the choice of representational primitives and their accompanying interpreter--is central to success at knowledge acquisition. Principles of 'design for acquisition' are presented, and it is shown how they are applied in the architecture of a medical expert system called MUM. One result is that design for acquisition can make it possible to acquire, directly from experts, types of knowledge traditionally programmed in by the knowledge engineer. A second result is that following the same design principles makes it possible to use conventional data entry technology (form-filling interfaces) to partially automate knowledge acquisition. A negative result is that good design alone is insufficient to facilitate knowledge acquisition. The limits of design for acquisition are illustrated with a case in which its principles conflict; it is difficult to represent general procedural knowledge without violating one or more of the design principles. For this kind of knowledge, an integration of form-filling approaches and induction form cases is proposed.