Metaphors and models
Michael Genesereth · National Conference on Artificial Intelligence · 1980
Much of one’s knowledge of a task domain is in the form of simple facts and procedures. While these facts and procedures may vary from domain to domain, there is often substantial &nilarity in the “abstract structure” of the knowledge. For example, the notion of a hierarchy is ‘found in biological taxonomy, the geological classification of time, and the organization chart of a corporation. One advantage of recognizing such abstractions is that they can be used in selecting metaphors and models that are computationally very powerful and efficient. This power and efficiency can be used in evaluating plausible hypotheses about new domains and can thereby motivate the induction of abstractions even in the face of partial or inconsistent data. Furthermore, there is a seductive argument for how such information processing criteria can be used in characterizing “intuitive” thought and in explaining the cogency of causal arguments. The idea of large-scale, unified knowledge structures like abstractions is not a new one. The gestalt psychologists (e.g. [Kohler]) had the intuition decades ago, and recently Kuhn [Kuhn], Minsky [Minsky], and Schank [Schank & Abelson] have embodied similar intuitions in their notions of paradigms, frames, and scripts. (See also [Bobrow & Norman] and [Moore & Newell] for related ideas.) The novelty hcrc lies in the use of such structures to select g6od metaphoa and models and in the effects of the resulting power and efficiency on cognitive behavior.