Alternative Metaphors for Artificial Intelligence

reliance on internal representation. Further discussion showed that many of the hard prob-lems encountered by AI researchers derive from con-straints imposed by the perspective of attack and might be avoidable attributes of the mind-modeling problem. Similarly, most of the points attacked by critics of AI are more properly directed toward one or more of the mechanist, formalistic, or representation-al presuppositions rather than the “grand objective ” of AI itself. It seems desirable, therefore, to determine if alternative perspectives might yield alterna-tive methods of attack. One fruitful starting point in the quest to find alternative (or com-plementary) perspectives should be an inven-tory of metaphors that have been or are being Previous examina-tion of the computa-tional metaphor exposed behavior inconsistent with that expected of metaphors in general. Specifically, despite demonstrated dis-similarity in the ref-erents of brains (minds) and comput-ers, the metaphor persists, not dissolves. Seeking an explanation of this behavior led to the conclusion that the computational metaphor is not truly a metaphor at all. Instead, it is a kind of shorthand expression, a label, for a set of philosophical presupposi-tions. These presuppositions generate a par-ticular perspective from which the problem of how to model a mind has been approached —a perspective that is intrinsically formalis-tic, mechanistic, and dependent on the methodological dualism that results from