Computing Machinery, Intelligence and Undecidability

Abstract

In 1950, Alan Turing proposed a decision criterion for intelligence validation in a computer. Most simply, if a human judge was incapable of deciding which of two hidden witnesses was the computer and which was the human being, the machine would have become intelligent. Turing starts his seminal paper «Computing Machinery and Intelligence» by asking «can machines think» [1]. As he himself recognized, this is a problem that demands being equated in a more practical way. To do that Turing devised a testable situation, using the so called "game of imitation", where before a blind jury; two different gender witnesses are instructed to answer by writing to a list of questions. One of the witnesses will try to imitate the gender of the other, trying to lead the judge into believing that he is a woman when he is a man or otherwise. The judge's mission is naturally to provide the right gender identification, thus ending the game. Turing then asked what would happen if we were to substitute the role of one of the human witnesses by a computer program to imitate human behaviour. The question «can machines think» therefore becoming an empirical one. Namely, «can a human judge distinguish between another human and a machine». He was, of course, taking very seriously the hypothesis that human intelligence as a set of producible behaviours is a computable procedure, within our powers to simulate given an enough amount of memory, processing power and programming complexity. The Turing debate, concerning the meaning of the Turing test to Artificial Intelligence feasibility and to what human intelligence is or is not, has been one of the major philosophical research domains over the last decades. Quite significantly, Horn's efforts to produce an argumentation map on the debate revealed, at the time he mapped it, 800 major "moves" «carried on by 400 scholars, researchers, and scientists worldwide, from at least ten academic disciplines» [2]. The debate includes claims, rebuttals, and counterrebutals upon the Turing test and his further claim that «at the end of the century… one will be able to speak of machines thinking without expecting to be contradicted» [1]. Fifty or so years debating over the issue, and the imitation game involving machines has been performed again and again. Using Eliza like strategies [3], chatterbots have been programmed and tested before human judges, only to find that we are still far from seriously consider that machines do think. Floridi et al., evaluating the 2008 Loebner Contest, reported: