为什么计算型的弱人工智能不可能成功--彭罗斯论战罗素和诺维格

Why Is Weak,Let alone Strong,AI Impossible--R.Penrose vs.S.Russell&P.Norvig

所谓计算型弱人工智能指的是在各种智力行为的表现上都至少跟人类一样好的计算型人工产品。彭罗斯改进了卢卡斯的论证,但保留了该论证中的核心部分(哥德尔不完备性定理),并据此论证所得出的结论是计算型的弱人工智能不可能成功。但哲学家如塞尔、计算机科学家如罗素和诺维格都认为彭罗斯的论证并不可信,并对彭罗斯的论证提出了三点反击。我们将在这篇论文中论证罗素和诺维格的回击并不成功,并指出思想实验方法的局限性。

By“weak AI”,the authors mean a computational machine whose observable and measurable performances are at least as good as those of an average matured human being in every respect involving intelligence.R.Penrose argues that computational weak AI is impossible.Penrose’s argument improves and evolves from that of J.R.Lucas but retains the core part of the latter,i.e.,the appeal to Godel’s second incompleteness theorem.However,many philosophers such as J.Searle and computer scientists like S.Russell and P.Norvig still incline to think that weak AI is certainly possible.In this presentation,the authors will(1)reconstruct Penrose’s complex argument in a simple but sensible way,(2)point out the weakness and confusions in Russell&Norvig’s objections,and(3)therefore show the danger and limitation of the philosophical method of thought experiment.