A Postmodern Critique of Artificial Intelligence

I argue that the "postmodern" understanding of language that has developed over the last few decades in Anglo-American philosophy provides the basis for a useful critique of artificial intelligence. This postmodern view corrects an error in the traditional Western conception of language that has led many researchers in AI and cognitive science into taking a rule-based or information-processing approach. Wittgenstein's view that language does not receive its meaning through definition, and Quine's view that neither words nor sentences but only discourse as a whole is the proper unit of meaning, argue against an attempt to formulate rules for understanding language, which is an essential part of "strong" AI. AI researchers are already beginning to correct this mistake, but an understanding of its true extent and depth can lead to the sort of radical rethinking that is necessary. The field of artificial intelligence (AI) aims to reproduce human intelligence by artificial means. The AI community generally concedes that achieving this goal has been harder than anticipated and awaits significant breakthroughs. But it insists that there are no barriers in principle. The obstacles to artificial intelligence are technical in nature, and a vigorous application of human intelligence can eventually overcome them. There are grounds to believe, however, that AI as customarily practiced rests on a fundamental mistake of principle. Hubert Dreyfus, one http://wpweb2.tepper.cmu.edu/jnh/ai.txt (2 of 19) [1/7/2010 1:37:12 PM] http://wpweb2.tepper.cmu.edu/jnh/ai.txt of the loudest of AI's critics, makes such a claim in his book, What Computers Can't Do (1979), and elaborates it in subsequent articles (1984; Dreyfus and Dreyfus, 1984, 1986, 1988). His critique actually has two prongs. He castigates the leading figures of AI for overestimating the prospects of their field and consistently failing to deliver. His shrill polemic along these lines has generated controversy that can obscure the second, more general prong of his critique. It is that AI rests on the mistaken assumption that human intelligence can (roughly speaking) be understood at the information processing level. A philosopher trained in the Continental tradition, he draws on the work of Heidegger to identify the error and show how it surfaces in AI. It is not my purpose to explain or defend Dreyfus' philosophical critique of AI. Rather I want to show that a related critique can be mounted on the "postmodern" understanding of language that has developed in the Anglo-American philosophical tradition, particularly in the work of Ludwig Wittgenstein and Willard Quine. I do not claim that AI is impossible. On the contrary, AI's philosophical underpinnings, as I will identify them, have a complex and indirect relationship to its potential for technical success. But if I am right, AI practitioners have typically operated on a set of assumptions that are not only incorrect but uninformed by Western thought's self-criticism of the last few decades. My critique also applies to the science of cognitive psychology, to the extent that it presupposes that human intelligence can be understood on a purely information processing level. In the five following sections I state my thesis, briefly paint the Western intellectual background that supplies AI with its present "modern" (as opposed to the "postmodern") worldview, explain why Wittgenstein and Quine think this worldview is mistaken, and then outline what I take to be the implications of this mistake for AI.