Ograniczenia sztucznej inteligencji w ujęciu Huberta L. Dreyfusa (cz. I)

Abstract

Artificial intelligence (AI) is an attempt of simulating of human intelligent behaviour using programming techniques. Since the field first evolved in the mid-1950s, AI researchers believed that insights into the nature of the mind can be gained by studying the operation of computer programs and invented dozens of computer programs that support some sort of intelligent behaviour. In suggesting that a machine with vast intellectual capability was in the offing and that this new intelligence might quickly be attained, these advocates mada themselves hostage to critics, who, increasingly aware of the limitations of certain of the theories being advanced, insisted on their inadequacies. The critics resented the exaggeration implicit in the claims of those, who saw only the promise of quick results and the epistemological assumptions implicit in the name the field had given itself. Hubert L. Dreyfus in his What Computers Can’t Do stirred up a controversy among all those interested in the possibility of formal models of man by arguing that, despite a decade of impressive print-outs and dire predictions of superintelhgent robots, workers in artificial intelligence were (1967) facing serious difficulties which they tended to cover up with special-purpose solutions and rhetorical claims of generality. In the seventies experimental AI systems included generally four kinds of programs: games, problem solving, language translating and pattern recognition (classification of H. Dreyfus). This first part of my paper Limitations of artificial intelligence by Hubert L. Dreyfus is about games and problem solving. The basic problem facing workers attempting to use computers in the simulation of human intelligent behaviour should now be clear: all alternatives must be made explicit – in game playing there is the exponential growth of the tree of these alternative paths. Dreyfus set human fringe consciousness against heuristically guided programm search. His analyses reveal two kinds of games: computable or quasicomputable (e.g. tic-tac-toe) and incomputable (e.g. chess). In problem solving the issue is not only how to direct a selective search among the explicit alternatives, but how to structure the problem so as to begin direct a selective search process (human essential/unessential discrimination vs. trialand-error search of computer). Impossible for counting out are open-structured problems, computable – maze problems, combinatorial problems, complex combinatorial problems. In surveying the two fields of AI (game playing and problem solving) underlying the optimistic interpretation of results in AI Dreyfus observed a current pattern: in each case the assumption was taken to be self-evident – an axiom seldom articulated and never called into question. In fact, the assumption turned out to be only one alternative hypothesis and a questionable one at that. The biological assumption that the brain must function like a digital computer no longer fits the evidence. The others lead to conceptual difficulties. The psychological assumption that the mind must obey a heuristic program cannot be defended on empirical grounds and a priori arguments in its defense fail to introduce a coherent level of discourse between the physical and the phenomenological.