Originally written in 1988.

proposal generally regarded as one of the clearest statements of the presumption held by most artificial intelligence researchers. In their hypothesis, they stated that “A physical symbol system has the necessary and sufficient means for general intelligent action. ” They defined “general intelligence” as intelligence of the same scope as human intelligence. Their suggestion was that certain kinds of manipulation of explicit information can be construed as intelligent behavior. Long before 1976 the search had concentrated on the proper kinds of explicit representations and the proper kinds of manipulation of that information. (You can trace this back at least as far as Leibniz in the 17th century.) What has emerged from this premise has been a flurry of successful, exciting starts in various areas of research. What has followed has been sad realizations that what is needed to extend these successes is computational power of such magnitude that it is mind boggling. This, the symbol manipulation approach to simulations of intelligence, now seems best suited for severely restricted domains (that is, expert systems). Forceful explanations of the poverty of explicit symbol manipulation have been advanced. Apparently, it is the explicitness of both the symbols and their manipulations that is a central problem. Is it possible to devise artificial systems that harbor, in some fashion, implicit knowledg