Modeling Human Expertise in Expert Systems
Nancy J. Cooke
The goal of cognitive psychologists interested in human expertise is to identify the cognitive structures and processes that are responsible for skilled performance within a domain. It is assumed that these structures and processes maintain some degree of generality regardless of the specific content of knowledge (i.e., the specific facts and rules making up the knowledge base). On the other hand, the goal of those interested in expert systems is often quite different. For knowledge engineers, the development of the expert system is the goal, and the explanation of expertise is typically secondary at most. Also, because knowledge, in the form of specific facts and rules, is assumed to be the power behind expert systems ( Minsky & Papert, 1974), knowledge engineers tend to be less concerned with general characteristics of expert knowledge than they are with the specific content of that knowledge. As a consequence, knowledge acquisition, the process of transferring knowledge from a source of expertise (either human or textual) to the expert system, is of paramount importance to the development of expert systems, but unfortunately it is also a major bottleneck in expert system design. Although the bulk of the research that has been done in cognitive psychology on expertise does not directly address the elicitation of specific facts and rules from human experts, methods that have been used in cognitive psychology to study memory organization and expertise can be applied to the knowledge elicitation problem. Furthermore, there are numerous arguments for considering the knowledge engineering implications of cognitive research related to the general structures and processes that underlie expertise.
The purpose of this chapter is to bridge the gap between cognitive psychology and knowledge engineering by pointing to some cognitive methods that have direct application to knowledge engineering, as well as to some research that is relevant to expert system design. The following section addresses methodologies used in cognitive psychology that can also be used to elicit expert knowledge. The remainder of the chapter reviews research on expertise. This work is organized according to the relevant cognitive faculties of pattern recognition, memory and knowledge organization, problem solving, and learning. The aim is to familiarize the reader with research in this area and to suggest some possible implications of this work. Whereas new knowledge acquisition methodologies are eagerly accepted by knowledge engineers, the implications of cognitive research are sometimes less apparent and are considered as a last resort, if at all. There are, however, several arguments for the application of cognitive research to knowledge engineering.
First, although the development of a successful expert system does not require the modeling of human expertise -- because there is no other form of expertise -- it is difficult to build such systems that are not inspired