Real Versus Artificial Expertise: The Development of Cognitive Models of Clinical Reasoning

Abstract

The purpose of this paper is to give an account of our approach to the study of clinical reasoning in medicine. This research has been in the domain of cognitive psychology rather than artificial intelligence and it is important to begin by stressing that they are two somewhat separate areas with their own paradigmatic approaches. However, there exists a well-established tradition of cross-fertilization of ideas between the two areas. There are, in general, two ways in which this cross-fertilization can take place. One is to develop a model that operates as both a psychological theory and an AI model, the most prominent recent example being SOAR (Newell, in press). The second, which is far more common, is to make use of the ideas and techniques in one area to develop a theory in the other, resulting in a complementary evolution of parallel areas rather than the development of a homogeneous theory. The predominance of this latter approach stems from the fact that the demands made upon theories are quite different in the two areas. The primary test of a psychological theory lies in its relationship to empirical data. AI does not suffer from this constraint. On the other hand, AI models need to satisfy a requirement of precision of definition that tends to be impossible to achieve in psychological models except within highly delimited domains.