27 September 2026 · History · about 18 minutes

The Trick Question: What Turing's 1950 Paper Actually Said

Everyone knows the Turing test. Almost nobody has read the paper that introduced it, which turns out to be something stranger and sharper than the legend.

The Question He Refused to Answer

The nine objections that Alan Turing answered in his 1950 paper "Computing Machinery and Intelligence" are still in circulation. The "heads in the sand" objection -- the consequences of machines thinking would be too dreadful, so let us hope they cannot -- is reproduced almost verbatim in every public debate about whether we should be developing artificial intelligence at all. Lady Lovelace's objection, that a machine can only do whatever we know how to order it to perform, was the chief philosophical complaint against large language models in 2023, made by people who rarely cited Lovelace. The mathematical objection, in the form Roger Penrose gave it in The Emperor's New Mind (1989), continues to generate serious dispute. Nearly every argument made about machine intelligence since 1950 was anticipated in this paper, which is one reason it is worth reading carefully rather than taking on trust.

In October 1950, a thirty-eight-year-old mathematician at the University of Manchester published the paper in Mind, a leading philosophy journal.¹ He was not, by training or temperament, a philosopher.² He was a mathematician who worked in mathematical logic, who had spent the war breaking German ciphers at Bletchley Park and who had, in 1936, written the paper that defined the theoretical foundations of computation.³ His friend and fellow mathematical logician Robin Gandy recalled later that Turing wrote this paper quickly and with enjoyment, unlike his mathematical papers, reading passages aloud "always with a smile, sometimes with a giggle."⁴ Gandy offered a candid verdict: the piece was intended not so much as a penetrating contribution to philosophy but as propaganda. Turing thought the time had come for philosophers and mathematicians and scientists to take seriously the fact that computers were not merely calculating engines but were capable of behaviour which must be accounted as intelligent; he sought to persuade people that this was so.⁵

The 1950 paper did not appear from nowhere. Two years earlier, Turing had written an unpublished report for the National Physical Laboratory titled "Intelligent Machinery," where he first developed the ideas of "unorganised machines" -- randomly connected networks of neuron-like elements, capable of being trained to perform tasks -- and proposed that a machine given the blank mind of an infant could be educated into something resembling adult intelligence. That report was not published until 1969, when it appeared in Machine Intelligence, vol. 5, edited by Meltzer and Michie at Edinburgh University Press.⁶ The 1950 paper's appearance was also, as the Stanford Encyclopedia records, stimulated by Turing's discussions at Manchester with Michael Polanyi, then professor of social studies at Manchester, and it reflects the general sympathy of Gilbert Ryle, then editor of Mind, with Turing's point of view.⁷ Recent scholarship, particularly Gonçalves's archival reconstruction of the controversy that led to the test, gives at least equal weight to Turing's exchanges with the neurologist and neurosurgeon Geoffrey Jefferson, whose Lister Oration of June 1949 proved decisive in a specific way: it was Polanyi's criticism that made chess an inadequate testbed, but Jefferson's demand that a thinking machine be able to write a sonnet that directed Turing toward conversation as the form the game would take.⁸

The paper opens with a piece of philosophical sleight of hand that is easy to miss. Turing says he will consider the question, "Can machines think?" He then immediately refuses to answer it. Definitions of "machine" and "think," he observes, might be framed to reflect the normal use of the words, but "this attitude is dangerous." If one goes by how people actually use the word "think," the answer to the question would have to be sought in a statistical survey, like a Gallup poll. "But this is absurd."⁹ So instead of answering the question, he replaces it.

The replacement is the imitation game, and it is worth describing carefully because the standard summary of it is wrong. The game begins with three players: a man (A), a woman (B), and an interrogator (C) of either sex. The interrogator is separated from the other two and must determine which is the man and which is the woman, using only written messages passed through some intermediary. The man tries to deceive the interrogator; the woman tries to help. Turing's sample exchange gives some of the flavour: the interrogator asks X to describe the length of his or her hair, and the man, playing X, replies that his hair is shingled and the longest strands are about nine inches long.¹⁰ Both players, in short, may lie. The key move comes next. Turing asks: what will happen when a machine takes the part of the man in this game? "Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman? These questions replace our original, 'Can machines think?'"¹¹

This is no longer the question of whether a machine can think in some deep metaphysical sense. It is whether there are imaginable digital computers that would do well in the imitation game. This is a much more tractable question, and Turing knew it was. He was not defining intelligence; he was proposing a criterion for discussing it without becoming tangled in definitions that philosophers had never managed to agree on. The game's opening scenario -- a man pretending to be a woman -- has generated a substantial scholarly debate. Sterrett (2000) argues that there are in fact two structurally distinct tests in the paper: the one in section 1 (where the machine competes against a man to pass as a woman, testing comparative resourcefulness in imitation) and the one restated in section 5 (where the machine competes simply to pass as human). The standard reading follows section 5; Sterrett's reading of section 1 is the neglected one, and on her account most philosophical criticism of the test has engaged with the wrong version of it.¹²

There is a further subtlety that commentators frequently overlook. Turing is explicit that he is not asking whether all digital computers, or even any existing digital computer, would do well. He is asking whether there are imaginable computers that would.¹³ It is a question about what is possible in principle, not a benchmark for present machines. When Turing states his own prediction, he is careful to mark it as such: "I believe that in about fifty years' time it will be possible, to programme computers, with a storage capacity of about 10⁹, 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."¹⁴ He then adds something less often quoted: "The original question, 'Can machines think?' I believe to be too meaningless to deserve discussion."¹⁵

This is the heart of Turing's philosophical move. He is not answering the question he appeared to pose. He is showing that the question, as normally understood, is malformed, and he is substituting a better one. The imitation game is not a test of whether machines "really" think. It is a way of giving the conversation somewhere to go.

The Architecture of Doubt

The paper's structure is unusual and this is where its real intelligence lies. Having established the game and stated his belief, Turing devotes the bulk of the paper not to arguing for his view but to considering nine objections to it, each of which he answers. He is candid about the reason: "The reader will have anticipated that I have no very convincing arguments of a positive nature to support my views. If I had I should not have taken such pains to point out the fallacies in contrary views."¹⁶ The method is recognizably Socratic: clear the field of bad arguments rather than prove your own position. One of the few books in the paper's short bibliography is Bertrand Russell's A History of Western Philosophy, which describes that method; the link is this essay's reading, not Turing's.¹⁷

The nine objections are, in Turing's own labels: the Theological Objection; the "Heads in the Sand" Objection; the Mathematical Objection; the Argument from Consciousness; Arguments from Various Disabilities; Lady Lovelace's Objection; the Argument from Continuity in the Nervous System; the Argument from Informality of Behaviour; and the Argument from Extrasensory Perception.¹⁸ The list is a portrait of what intelligent people actually believed about thinking machines in 1950, and Turing deserves credit for taking each objection seriously rather than dismissing it.

Nine Objections, Nine Answers

The Theological Objection holds that thinking is a function of the immortal soul, which God has granted to every man and woman but not to animals or machines; therefore no machine can think. Turing is almost amused. He cannot accept any part of the argument, but he tries to meet it in theological terms, noting that it "implies a serious restriction of the omnipotence of the Almighty." Should we not believe that God has freedom to confer a soul on an elephant if He sees fit? An argument of exactly the same form may be made for machines, and building one should be understood not as "irreverently usurping His power of creating souls" but as being "instruments of His will."¹⁹ He then abandons the terrain entirely: "I am not very impressed with theological arguments whatever they may be used to support," citing Galileo as evidence that scriptural refutations of scientific claims tend not to age well.²⁰

The "Heads in the Sand" Objection is perhaps the most psychologically honest item on the list: the consequences of machines thinking would be too dreadful, so let us hope they cannot. Turing quotes it almost verbatim: "The consequences of machines thinking would be too dreadful. Let us hope and believe that they cannot do so." He says the argument is "seldom expressed quite so openly" but affects most of us who think about the matter at all, adding that the theological objection's popularity is clearly connected with the same feeling. He declines to refute it formally: "I do not think that this argument is sufficiently substantial to require refutation. Consolation would be more appropriate: perhaps this should be sought in the transmigration of souls."²¹

The Mathematical Objection is the one Turing takes most seriously, for the obvious reason that it comes from his own discipline. Gödel's incompleteness theorem (1931), he notes, shows that in any sufficiently powerful logical system, statements can be formulated that can neither be proved nor disproved within the system. There are similar results due to Church, Kleene, Rosser, and Turing himself (1937), the last of which refers directly to machines and states that there are certain things a machine with an infinite store cannot do: if rigged to answer questions, there will be some to which it will either give a wrong answer or fail to give an answer at all.²² The objection then runs: there are things machines cannot do that humans can, so machines cannot think as humans do.

Turing's reply is precise and worth dwelling on. The short answer is that although limitations on any particular machine are established, "it has only been stated, without any sort of proof, that no such limitations apply to the human intellect." But he refuses to dismiss the objection so easily, noting the genuine feeling of superiority one has on such an occasion. His deeper point is that even if the limits are real, they do not settle anything: "There might be men cleverer than any given machine, but then again there might be other machines cleverer again, and so on."²³ This is not a full answer to the mathematical objection; it is an honest statement of the limits of what the objection shows.

The Argument from Consciousness is presented through a quotation from a specific antagonist: Geoffrey Jefferson, neurologist and neurosurgeon, who had delivered the Lister Oration at the Royal College of Surgeons on 9 June 1949, under the title "The Mind of Mechanical Man."²⁴ Jefferson had argued: "Not until a machine can write a sonnet or compose a concerto because of thoughts and emotions felt, and not by the chance fall of symbols, could we agree that machine equals brain -- that is, not only write it but know that it had written it."²⁵ The Jefferson offprint was in Turing's papers; the archive record shows it came to the King's College archive via Gandy as one of Turing's executors.²⁶ Jefferson shared a Manchester base with Turing and, on this essay's reading, very likely discussed these matters with him directly, though no documentary record of such a conversation survives.

Turing's response is to identify the argument from consciousness as a form of solipsism. If the only way to be certain a machine thinks is to be the machine and feel oneself thinking, then by the same logic the only way to know that any other human thinks is to be that particular human. "It is in fact the solipsist point of view. It may be the most logical view to hold but it makes communication of ideas difficult." Instead of this, we have the "polite convention that everyone thinks." Turing believes that most who hold the consciousness argument could be persuaded to abandon it rather than be forced into the solipsist position, and would then probably accept the imitation game as a test.²⁷

To make the point vivid, Turing imagines a viva voce examination of a notional machine that has written a sonnet. The interrogator asks whether "a spring day" would not do as well as "a summer's day" in the first line. The machine replies: "It wouldn't scan." The interrogator tries "a winter's day" -- that would scan. The machine: "Yes, but nobody wants to be compared to a winter's day." When the interrogator presses further by noting that Mr Pickwick reminded one of Christmas, and Christmas is a winter's day, the machine answers: "I don't think you're serious. By a winter's day one means a typical winter's day, rather than a special one like Christmas."²⁸ Turing asks what Jefferson would say if the machine answered like this throughout. Hodges observes that Turing "took on a second-rank target in countering the published views of the brain surgeon G. Jefferson" and that Wittgenstein's views on mind "would have made a more serious point of departure" -- and Turing had attended Wittgenstein's 1939 Cambridge lectures on the foundations of mathematics, where the two were documented as disagreeing.²⁹ The paper that emerged engaged Jefferson rather than Wittgenstein, which is either a strategic choice or a missed opportunity, depending on one's view.

The Arguments from Various Disabilities are a long list of things critics say machines will never be able to do: be kind, resourceful, beautiful, friendly, have initiative, have a sense of humour, tell right from wrong, make mistakes, fall in love, enjoy strawberries and cream, make someone fall in love with them, learn from experience, use words properly, do something really new.³⁰ Turing's analysis of why people hold these views is shrewd. A man has seen thousands of machines in his lifetime and drawn general conclusions: they are ugly, built for limited purposes, and varied in behaviour hardly at all. He concludes that these are necessary properties of machines in general, but this is "scientific induction" applied to an inappropriate domain. The inability to enjoy strawberries and cream he treats with characteristic deadpan: the disability may seem frivolous, and possibly a machine could be made to enjoy the dish, but "any attempt to make one do so would be idiotic."

Lady Lovelace's Objection is the most celebrated, and Turing handles it with the most care. Ada Lovelace, in the Notes she appended to her translation of Menabrea's article on Babbage's Analytical Engine, published in Scientific Memoirs in 1843 -- Turing follows Hartree (1949) in dating the memoir 1842, which is when Menabrea's original French article appeared -- had stated: "The Analytical Engine has no pretensions to originate anything. It can do whatever we know how to order it to perform."³¹ This is really two objections: that a machine can only do what we tell it to do, and that a machine can never take us by surprise. Turing answers the second point with a line that is almost endearing in its self-deprecation: "Machines take me by surprise with great frequency. This is largely because I do not do sufficient calculation to decide what to expect them to do, or rather because, although I do a calculation, I do it in a hurried, slipshod fashion, taking risks."³² On the deeper question of origination, he turns the objection on humans: "Who can be certain that 'original work' that he has done was not simply the growth of the seed planted in him by teaching, or the effect of following well-known general principles?"

The Argument from Continuity in the Nervous System holds that the brain is not a discrete-state machine but a continuous one; small errors in the information about a nervous impulse can make large differences to the outgoing signal; therefore a discrete digital machine cannot imitate it. Turing's reply is that if one adheres to the conditions of the imitation game, the interrogator will not be able to exploit this difference. A digital computer cannot predict exactly what a differential analyser would give as an answer to a problem, but it can give the right sort of answer, and for the purposes of the game that is enough.³³

The Argument from Informality of Behaviour turns on the observation that no complete set of rules could describe what a human should do in every conceivable circumstance: what if a traffic light shows both red and green at once? Since humans can handle such situations without explicit rules, the argument runs, humans cannot be machines. Turing calls the inference an undistributed middle: it does not follow from the absence of explicit rules of conduct that there are no laws of behaviour operating. The distinction between "rules of conduct" (precepts one can act on consciously) and "laws of behaviour" (natural laws as applied to a body) is being confused.³⁴ He is less than fully satisfied with his own answer here, but the point is well taken: the absence of an explicit rulebook does not show that behaviour is not governed by laws.

The ninth objection, the Argument from Extrasensory Perception, is the most peculiar section of a paper that never stops testing its reader's credulity. The Stanford Encyclopedia's commentary is characteristically diplomatic: "Perhaps it is intended to be tongue-in-cheek, though, if it is, this fact is poorly signposted by Turing."³⁵ Turing says flatly that the statistical evidence, at least for telepathy, is "overwhelming" and that the idea disturbs him. If a human interrogator is telepathic, the imitation game is corrupted: the interrogator could read the machine's "mind" directly. His solution is to put the competitors into a "telepathy-proof room." Whether this is deadpan or sincere, it is the one section of the paper where one wishes he had written more carefully.

The Positive Case and the Child Machine

Having knocked down the objections, Turing turns at last to his own constructive proposals, under the heading "Learning Machines." He opens the section by repeating his earlier admission: "The reader will have anticipated that I have no very convincing arguments of a positive nature to support my views." He then pivots to what he calls "such evidence as I have," and the constructive case that follows is the least-discussed and most prescient part of the paper.

Rather than trying to write a programme that simulates an adult human mind directly, Turing asks, why not try to simulate a child's? "The child brain is something like a notebook as one buys it from the stationer's. Rather little mechanism, and lots of blank sheets." If this child machine were then subjected to an appropriate course of education, the adult brain would result.³⁶ The idea had first appeared in the 1948 "Intelligent Machinery" report; the 1950 paper adds an explicit evolutionary analogy: the structure of the child machine corresponds to hereditary material, changes to the child machine correspond to mutation, and natural selection corresponds to the judgment of the experimenter. The experimenter has one advantage over nature: he is not restricted to random mutations and can trace a cause for some weakness and think of the mutation that will improve it.³⁷

The machine's education would involve punishments and rewards, but Turing is clear that these alone are insufficient. There must be what he calls "unemotional" channels of communication through which language and knowledge can be transmitted more efficiently than by trial and error alone. He also suggests that a random element should probably be included in a learning machine, since a degree of randomness is useful when searching for a solution, just as random mutation is useful in evolution.³⁸ The teacher, Turing notes, will often be "very largely ignorant of quite what is going on inside" the learning machine, though still able "to some extent to predict his pupil's behavior." The machine would surprise its designers. Its behaviour would constitute "a departure from the completely disciplined behaviour involved in computation, but a rather slight one, which does not give rise to random behaviour, or to pointless repetitive loops."³⁹ The paper closes with a sentence that is both modest and grand: "We can only see a short distance ahead, but we can see plenty there that needs to be done."

Legacy and Honesty

Digital computers were still rare enough in 1950 that most educated people had never seen one. The Manchester machine that Turing worked with had a store of about 165,000 binary digits, and he thought 10⁹ bits would probably be needed for the imitation game -- roughly 125 megabytes, a figure that seems quaint now.⁴⁰ His prediction of 70 per cent interrogator accuracy after five minutes of questioning was already more cautious than the legend suggests; he did not claim machines would be indistinguishable from humans, only that they would be right more often than chance. In his BBC radio talks of 1951 and 1952 -- the first titled "Can Digital Computers Think?" -- he elaborated and qualified the imitation game further, and in the 1952 discussion "Can Automatic Calculating Machines Be Said to Think?" he was joined by Jefferson, Newman, and Braithwaite, facing across a microphone the man whose Lister Oration had helped shape the very game under discussion.⁴¹

The argument from consciousness, which Turing sourced to Jefferson's 1949 question about whether a machine that writes a sonnet knows that it has written it, addresses the same territory that David Chalmers named "the hard problem of consciousness" in 1995 -- why physical processes give rise to subjective experience at all -- though Jefferson's formulation is not identical to Chalmers's and should not be read as anticipating it exactly.⁴² The mathematical objection, in the form Penrose later gave it, continues to generate serious dispute. Lady Lovelace's objection about origination was the chief philosophical complaint against large language models in 2023. The "heads in the sand" objection is reproduced almost verbatim in every public debate about whether we should be developing AI at all.

The paper is not a definition of intelligence; Turing says the original question is too meaningless to deserve discussion. It is not a proof that machines can think; Turing says so frankly at the start of section 7. It is not quite what later philosophers made of it: the Turing test as it came to be understood after Searle's Chinese Room argument treated behavioural indistinguishability as a logically sufficient condition for intelligence, which is a stronger claim than Turing makes. His target was not deception but the double standard: the way people applied a stringent test to machines that they would never apply to one another. Andrew Hodges notes in the Stanford Encyclopedia entry on Turing that he "paid scant attention to the questions of authenticity and deception implicit in his test, essentially because he wished to by-pass questions about the reality of consciousness" -- this essay reads that as confirming the interpretation above rather than undermining it, though the point is contested.⁴³

What the paper does not contain is also instructive. There is no claim that intelligence is reducible to computation. There is no suggestion that passing the test is the only criterion for mind. There is no triumphalism. What there is, instead, is a remarkably honest account of the state of the argument: a man who believed machines would eventually think, who admitted he could not prove it, and who decided the most useful thing he could do was to clear away the rubbish that was blocking the view. Hodges observes that Wittgenstein's views on mind "would have made a more serious point of departure" than Jefferson's -- and this essay takes the observation seriously. Turing chose a winnable opponent. Whether the paper would have been better or worse for engaging Wittgenstein instead is a question it cannot answer for itself.

Turing answered none of the nine objections conclusively. Neither has anyone else.