Replacing an unanswerable question
Turing opens by rejecting the direct route to "Can machines think?" Defining "machine" and "think" by ordinary use, he writes, reduces the answer to "a statistical survey such as a Gallup poll," which he calls absurd. He substitutes a question in unambiguous terms: the imitation game, where a man, a woman, and an interrogator decide which is which from written answers alone. [2]
Turing then asks what happens when a machine takes the man's part. "These questions replace our original, 'Can machines think?'" Section 5 restates it as one machine standing in for the man against a human witness, the form the test is remembered by. [2]
What the procedure actually rules out and in
The written-answers-only rule is deliberate: no engineer can build skin indistinguishable from a person's, so the test draws its line at intellectual capacities, voice included. Turing's specimen exchanges, a declined sonnet, a slow sum, a chess move, show the format can probe any field, and a machine need not be flawless, only convincing. [2]
Atlas interpretation: That is narrower than the phrase "Turing test" now carries: indistinguishable performance on a timed written exchange, nothing about internal experience. Turing sets the harder question aside rather than answer it: "The original question, 'Can machines think?' I believe to be too meaningless to deserve discussion." [2]
The prediction, and its precise terms
Turing's forecast, stated once, in section 6: "I believe that in about fifty years' time it will be possible to programme computers, with a storage capacity of about 10^9, to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning." A machine fooling the judge about 30 percent of the time, in five minutes, by roughly 2000, on a store of about a billion bits. [2]
He had a live sense of scale: he cites the storage of the machine at the University of Manchester as about 165,000 binary digits, six orders of magnitude below the 10^9 he thought the test would need, which is why he called fifty years, not five. [2]
A pass on paper, contested in practice
In June 2014, a chatbot called Eugene Goostman, posing as a 13-year-old Ukrainian boy, convinced 33 percent of judges it was human across five-minute written conversations. The organizer called this a pass, since 33 percent exceeds the 30 percent implied by "not more than 70 per cent chance." [3]
Atlas interpretation: Matching the number is not answering Turing's question as he meant it. The persona did the work: a young non-native speaker excuses gaps a judge would otherwise flag, closer to the gaming his specimen answers guard against than to general competence. A bot that talks its way around the target has not thereby been shown to think, the gap between evidence a pass supplies and proof. [3][4]
Sources
- I.—Computing Machinery and Intelligence
Mind · Oct 1950
- Computing Machinery and Intelligence
Mind · Sep 9, 2026
- Virtual Tween Passes Turing Test
IEEE Spectrum · Jun 10, 2014
- The Turing Test
Stanford Encyclopedia of Philosophy · Oct 4, 2021