Why the Turing Test Revised is Still the Turing Test

Can machines think? So Alan Turing begins his paper "Computing Machinery and Intelligence", discussing how one might assess whether an electronic computer can truly "think" (Turing, 1950, p. 29). It is here that Turing explains his famous Turing Test: can a computer win in an "imitation game" with a human? That is, can an electronic computer provide written answers to an interrogator which would fool the interrogator into believing that device is human? To answer yes, Turing argues, is to acknowledge the reality of a thinking machine (Turing, 1950, p. 30). Since 1950, technology has advanced impressively. There has been a computer that “passed” the Turing Test, fooling testers with a clever ability to redirect conversation when questions became too challenging (Marcus, 2014). Yet this strategy feels more like a cheap trick than the mark of an authentic thinker, casting doubt on the test’s validity. What could be a more reliable test for machine understanding? I argue for two simple additions to the Turing Test that would eliminate loopholes but leave the spirit of the test unchanged. That Turing's original imagination of the Turing Test involved a game of deception is distracting to its fundamental principle: we will say a machine “thinks” when its logical outputs are in no way noticeably inferior to that of a human