LESSONS FROM PERCEPTION FOR CHESS-PLAYING PROGRAMS (AND VICE VERSA)

For nearly twenty years, artificial intelligence and cognitive psychology have maintained a close symbiotic relationship to each other. It has often been remarked that their cooperation stems from no logical necessity. That a human being and a computer are both able to perform a certain task implies nothing for the identity, or even similarity, of their respective performance processes. Each may have capabilities not shared by the other, and may build its performances on those peculiar capabilities rather than upon those they hold in common. in spite of this logical possibility of total irrelevance of the one field for the other, during the last two decades there has been massive borrowing in both directions. Artificial intelligence programs capable'of humanoid performance in particular task domains have provided valuable hypotheses about the processes that humans might use to perform these same tasks, and some of these hypotheses have subsequently been supported by evidence. Bobrow's &] STUDENT program, for example, which translated story problems into algebraic equations, provided a model, later tested by Paige & Simon ' 11 J, for some of the human syntactic processes in performing that task. Conversely, hypotheses and data about human performance have been important inputs to artificial intelligence efforts. The General Problem Solver, for example, received its early shape from analyses of human think ing-aloud protocols in a problem solving task I 8*.