Knowledge versus Knowledge Representation

The conception of knowledge by the engineers charged with building expert systems contrasts sharply with that of social scientists. However, the narrow focus of Diana Forsythe on knowledge engineers missed a respect of the effects of representing knowledge in the larger AI community [FOR2]. In “What Is a Knowledge Representation” [DAVIS], Davis et al. describe the five roles of a knowledge representation, observing that a knowledge representation requires that a set of ontological commitments as well as theories of reasoning be embodied, necessarily biasing the knowledge and resulting inferences. I argue that the failure of knowledge engineers to express their notion of knowledge is largely due to the restrictive apparatus of the rule-based expert systems in which they operate. Instead, I argue that a truly “expert” system will require numerous knowledge representations, just as a human uses, and that a true, automated expert will most likely need to be trained in much the same way that a human expert is: through life experiences as well as expert training. Expertise is not solely the formal knowledge in the expert domain, but the entire collection of social and cultural experiences.