he most cited paper in the philosophy of artificial intelligence was published in Mind, a philosophy journal, in October 1950. It is still in copyright, so this page summarises it and quotes one sentence.

the document

Turing, "Computing Machinery and Intelligence". Read it at the publisher's page. It begins:

“I propose to consider the question, ‘Can machines think?’”
— Turing 1950, section 1

what it says

Turing declines to define "machine" and "think", on the ground that ordinary usage would make the answer a matter for an opinion poll, and replaces the question with a game. In the original version an interrogator, communicating only in writing, tries to tell which of two hidden players is a man and which a woman, while the man tries to mislead; Turing then asks what happens when a machine takes the man's part, and whether the interrogator will be wrong as often [paraphrase, not a quotation]. Most of the paper is a reply to nine objections, from the theological to Lady Lovelace's (that a machine can only do what it is told), followed by a proposal to build a "child machine" and educate it. He predicted that within about fifty years machines would play well enough that an average interrogator would have no better than a seventy per cent chance of a correct identification after five minutes.

what's contested

Which test? Sterrett (2000) argues the paper contains two tests, the original gender-based imitation game and the simpler "is it a machine?" version later called the Turing test, and that the original is the better test, because it asks the machine to imitate a performance rather than simply to pass as human [paraphrase, not a quotation]. Copeland (2000) reconstructs Turing's intentions from his other writings and broadcasts. Proudfoot (2011) reads the game as Turing's deliberate response to anthropomorphism: intelligence as a concept that depends partly on the observer, tested under controlled conditions so the observer can be fooled only fairly.

Is passing enough? Block (1981) imagined a machine that passes by looking up every possible conversation in a vast table, and concluded that behavioural tests cannot be sufficient for intelligence, because intelligence depends on how the behaviour is produced. Searle (1980)'s Chinese room argues that running a program, however well it converses, is not understanding. French (1990) argued the reverse problem: the test is too hard and too human, because "subcognitive" questions expose a machine that is intelligent but not human-shaped. Harnad (1991) proposed a "total" test including sensorimotor capacity.

methodological critiques

As an experiment the game is underspecified: who the interrogators are, how long they have, what they are told, and what counts as passing all change the result. Shieber (1994) criticised the restricted public contests of the early 1990s on just these grounds. The most careful recent run, a randomised and preregistered three-party version by Jones & Bergen (2025), found a language model prompted to adopt a persona was judged human more often than the real humans it was paired with; whether that shows intelligence, skill at imitation, or a weakness in five-minute judgments is exactly the old argument [paraphrase, not a quotation].

later reception

The surveys at fifty years (Saygin, Cicekli & Akman, 2000) found the paper discussed more than any other in the field. Its most durable part may be the least discussed: the section on learning machines, which proposed educating a machine rather than programming it. See also anthropomorphism, and the Turing page on this site for his wartime work.

in the tour: the room called "Governors and feedback" has a placard called "The question he replaced" ···

in the tour: a closed room has a placard called "The room that answers in Chinese" ···

References

1. Turing (1950). Computing machinery and intelligence. Mind. doi:10.1093/mind/LIX.236.433
2. Sterrett (2000). Turing's two tests for intelligence. Minds and Machines. doi:10.1023/A:1011242120015
3. Copeland (2000). The Turing test. Minds and Machines. doi:10.1023/A:1011285919106
4. Proudfoot (2011). Anthropomorphism and AI: Turing's much misunderstood imitation game. Artificial Intelligence. doi:10.1016/j.artint.2011.01.006
5. Block (1981). Psychologism and Behaviorism. The Philosophical Review. doi:10.2307/2184371
6. Searle (1980). Minds, brains, and programs. Behavioral and Brain Sciences. doi:10.1017/S0140525X00005756
7. French (1990). Subcognition and the limits of the Turing test. Mind. doi:10.1093/mind/XCIX.393.53
8. Harnad (1991). Other bodies, other minds: A machine incarnation of an old philosophical problem. Minds and Machines. doi:10.1007/BF00360578
9. Shieber (1994). Lessons from a restricted Turing test. Communications of the ACM. doi:10.1145/175208.175217
10. Jones & Bergen (2025). Large Language Models Pass the Turing Test. arXiv:2503.23674. arxiv.org/abs/2503.23674
11. Saygin, Cicekli & Akman (2000). Turing Test: 50 Years Later. Minds and Machines. doi:10.1023/A:1011288000451

see also