Differing Methodological Perspectives in Artificial Intelligence Research

A variety of proposals for preferred methodological approaches has been advanced in the recent artificial intelligence (AI) literature Rather than advocating a particular approach, this article attempts to explain the apparent confusion of efforts in the field in terms of differences among underlying methodological perspectives held by practicing researchers The article presents a review of such perspectives discussed in the existing literature and then considers a descriptive and relatively specific typology of these differing research perspectives. It is argued that researchers should make their methodological orientations explicit when communicating research results, to increase both the quality of research reports and their comprehensibility for other participants in the field. For a reader of the AI literature, an understanding of the various methodological perspectives will be of immediate benefit, giving a framework for understanding and evaluating research reports In addition, explicit attention to methodological commitments might be a step towards providing a coherent intellectual structure that can be more easily assimilated by newcomers to the field. More than a quarter century after its beginnings, AI has yet to produce a commonly accepted statement of purpose or description of conventional research practices. Studies are reported in a wide range of publications. While some focus on the field (e.g., Artzficial Intelligence), others are concerned with different research areas (e.g., Behavzoral and Brain Sczences). What results is a profusion of literature that is difficult to encompass for stuThis research was supported in part by a gift from the Hughes AI Research Laboratory, a division of Hughes Aircraft. This article has been a focus for energetic discussion with a number of individuals, each of whom has influenced our thinking in a variety of ways, but none of whom should be held responsible for our conclusions We would like to thank Rob Kling, Bruce Porter, Doug Fisher and Dan Easterlin for providing a patient, but challenging audience In addition, we wish to acknowledge helpful editorial suggestions by Martin Ringle dents and practitioners alike. If the study of AI is to be considered, and conducted as a scientific endeavor rather than an amorphous enterprise whose subject matter is constantly shifting (or even disappearing as results are incorporated into other fields), one might profitably ask if distinct methodological perspectives can be identified by which to organize some of the current confusion of efforts. Perhaps, as others have pointed out, “there are undoubtedly some views of AI that are more fruitful than others . . . We ought to be guided by the most productive paradigms” (Nilsson, 1982). Concern for methodological issues in AI research is on the upswing (Ohlsson, 1983; Bundy, 1983a; Cercone and McCalla, 1983). However, this interest appears to be prescriptive focusing on what AI researchers should be rather than what they actually are doing. For example, Cercone and McCalla relegate the multitude of differing AI approaches to a rather constraining spatial metaphor, a “pie” composed of problem areas like vision, expert systems, or learning. They then specify “design objectives that any ideal AI system should meet” (p. 4, italics added). These objectives include development of a working system, external validation of system capabilities, and identification of generalized results. While these may well be desirable characteristics for some idealized view of AI research, they fall short of specifying the underlying assumptions or basic objectives of practicing researchers. This despite