Chapter 10: The Imitation Game

In one paragraph

Between 1947 and 1952, Alan Turing turned machine intelligence from a definition hunt into a controlled comparison of performances. Turing did not found AI as a research field — that came in 1956. He supplied a testable setup, a way to bracket bodily cues, and a provocation later readers would simplify under the name “Turing Test.”

Cast of characters

Name	Lifespan	Role
Alan Turing	1912–1954	Author of the 1947 LMS lecture, the 1948 NPL report Intelligent Machinery, and the 1950 Mind paper “Computing Machinery and Intelligence.” Reader in Mathematics at the University of Manchester from 1948.
Max H. A. Newman	1897–1984	Fielden Professor of Mathematics at Manchester; head of Turing’s department; built the Manchester Royal Society Computing Laboratory after the war. Participant in the January 1952 BBC broadcast.
Sir Geoffrey Jefferson	1886–1961	Professor of Neurosurgery at Manchester. His 1949 Lister Oration “The Mind of Mechanical Man” attacked machine-intelligence claims on biological grounds; in the 1952 broadcast he carried the same objection forward against Turing.
R. B. Braithwaite	1900–1990	Knightbridge Professor of Moral Philosophy at Cambridge; the philosophy-of-mind voice in the 1952 broadcast.
Ada Lovelace	1815–1852	Author of the 1842 Sketch of the Analytical Engine. Cited by name in Mind §6.6 (“Lady Lovelace’s Objection”): the Analytical Engine “has no pretensions to originate anything.”
Helen Keller	1880–1968	Cited by Turing in Mind §7 to argue that education does not require conventional sensory channels — communication in both directions suffices. Used to defend the child-machine programme.

Timeline (1947–1952)

timeline
    title Turing's Imitation-Game Pivot, 1947–1952
    20 February 1947 : Lecture to the London Mathematical Society on the ACE : "Fair play for the machine"; brain-memory estimate of 10¹⁰ binary digits; intelligence as learning from experience
    1948 : NPL report Intelligent Machinery (unpublished until 1968) : Unorganized A-type machines; cortex-as-unorganized-machine analogy; education by rewards and punishments; two-room chess proto-imitation experiment
    October 1950 : "Computing Machinery and Intelligence" published in Mind 59 (236), pp. 433–460 : Imitation Game definition; teleprinter as ideal interface; nine objections including Lady Lovelace; 10⁹-bit / 70% / five-minute / end-of-century prediction; child-machine programme
    10 January 1952 : BBC Third Programme records "Can Automatic Calculating Machines Be Said To Think?" : Four-way conversation: Turing, Newman, Jefferson, Braithwaite. Broadcast 14 January.

Plain-words glossary

• Imitation Game — Turing’s 1950 protocol for testing machine intelligence: a typed exchange in which an interrogator, hidden from the other parties, tries to identify which of two unseen respondents is the human and which is the machine. Turing’s term; the rebrand to “Turing Test” came later.
• Teleprinter — An electromechanical device that sent typed text over a wire. In Turing’s game it made the exchange text-only, so the interrogator judged linguistic behaviour rather than bodily cues.
• Discrete-state machine — In Turing’s 1950 framing, a system whose future behaviour is determined by its current state and input, moving through a finite (if enormous) space of configurations. The class to which Turing argues a digital computer belongs.
• Universality argument — Turing’s claim that a single discrete-state machine, given adequate storage, speed, and the right programme, can imitate any other discrete-state machine. The argumentative bridge from the parlour game to actual digital hardware.
• Unorganized machine — Turing’s 1948 term for a discrete-state random network of simple logical units (A-type machines used NAND-like elements). Its initial state is deliberately unformed; he proposed the human infant cortex as a biological analogue.
• Paper interference vs. screwdriver interference — Turing’s 1948 distinction between modifying a machine by physically rewiring it (screwdriver) and by communicating instructions to it (paper). Education, he argued, is largely paper interference.
• Child machine — Turing’s 1950 proposal for how to build the imitation player without sixty workers spending fifty years hand-coding it: programme an infant mind rather than an adult one, and educate it through rewards and punishments.

The historical importance of Alan Turing’s 1950 paper in Mind is not its prediction of a machine fooling an interrogator by the year 2000. It is an epistemic move. Between 1947 and 1952, Turing dismantled the definitional debate over the word “intelligence” and replaced it with a behavioural protocol. By routing interaction through a teleprinter and stripping away the physical body, Turing established an empirical baseline against which machine intelligence could be tested.

That is a narrower claim than the one later mythology often makes for the paper. Turing did not found a research field called artificial intelligence in 1950, and he did not offer a working demonstration. He supplied a protocol, a set of constraints, and a way of refusing an argument about essences. The field’s name, institutions, funding channels, and public programmes would come later. Here, the load-bearing move is the replacement of “Can machines think?” with a situation in which an interrogator has to decide.

The Precursors

Turing’s pivot toward an empirical test of machine intelligence was not an abrupt arrival in 1950. The philosophical groundwork and the operational shape of the test were laid out in the preceding three years, while he was thinking about machines that mostly existed as designs, reports, and paper simulations rather than as usable interlocutors.

On 20 February 1947, Turing addressed the London Mathematical Society. Speaking on the design of the Automatic Computing Engine (ACE) at the National Physical Laboratory (NPL), Turing confronted the assumption that a machine must be flawless to be considered intelligent. In his view, “if a machine is expected to be infallible, it cannot also be intelligent.” The demand for absolute precision was a trap, holding the machine to a standard that human intelligence did not meet. He concluded that “fair play must be given to the machine,” proposing that true intelligence resided in the learning loop, and that “what we want is a machine that can learn from experience.”

The phrase “fair play” did real argumentative work. A calculating machine could be dismissed for making a mistake, yet human intelligence was often recognized precisely in the ability to revise, generalize, guess, and learn from failure. Turing was not asking readers to forgive bad arithmetic. He was shifting the standard from mechanical infallibility to adaptive performance. If learning from experience was part of intelligence in people, then a machine built to alter its behaviour after experience had to be judged by a comparable rule.

During the same lecture, Turing estimated the memory capacity of the human brain to be “about ten thousand million binary digits”—roughly $10^{10}$ bits. By contrast, the total storage capacity of the ACE’s mercury delay lines was “about 200,000 binary digits,” which Turing wryly noted was “probably comparable with the memory capacity of a minnow.” Each five-foot mercury tube stored 1024 binary digits. The machine he was describing was therefore not a hidden thinking device waiting to be unmasked. It was an ambitious stored-program design whose memory was tiny beside the human comparison Turing himself introduced. Nevertheless, he argued that a more modest storage of “a few million digits” might suffice for limited demonstrations of intelligence, such as playing a game of chess.

In 1948, Turing drafted an unpublished NPL report titled Intelligent Machinery. The opening pages systematically listed five common objections to machine intelligence—human pride, Promethean religious feeling, the limitations of recent machinery, Gödel-style mathematical theorems, and the idea that intelligence is merely a “reflection” of the creator—and provided a “Refutation of Some Objections” for each. The report’s structure matters because it shows Turing already treating machine intelligence as a dispute about standards of evidence. The sceptical positions were not ignored; they were turned into test conditions.

One of Turing’s examples in the report came from the familiar schoolroom story of Gauss adding an arithmetic progression by violating the expected procedure. The point was not whether the anecdote was good biography. Turing used it to separate intelligence from obedience. A child who simply follows every instruction may get the expected marks, but intelligence sometimes appears when the subject sees a more general pattern than the examiner requested. That example prepared the ground for the later objection that a machine can never do more than it was told. Turing’s answer was already taking shape: the relevant question is not whether the machine stayed inside a human’s visible instructions, but whether its performance forces us to revise what those instructions made possible.

More critically, the 1948 report introduced the concept of “unorganized machines.” These were discrete-state random networks constructed from simple logical units, with A-type machines built from randomly connected NAND-like elements. They were not programmed in the later sense of a neatly written symbolic routine. Their initial state was deliberately unformed. Turing proposed the human infant cortex as a biological analogue to such an unorganized network: not a finished adult mind, but a substrate capable of being organized by training.

The challenge was how to organize it. Turing framed the education of machinery through a system of rewards and punishments: “events which shortly preceded the occurrence of a punishment signal are unlikely to be repeated, whereas a reward signal increased the probability of repetition of the events which led up to it.” He distinguished between “screwdriver interference”—physically rewiring the machine—and “paper interference,” the communication of instructions. Education, he noted, was largely a matter of paper interference. This distinction is one of the technical bridges from the 1948 report to the 1950 paper. A machine could be modified by altering its material wiring, but it could also be shaped by a symbolic traffic of orders, marks, and feedback. The future Imitation Game would depend on that second kind of traffic.

At the very end of the Intelligent Machinery report, in a section titled “Intelligence as an Emotional Concept,” Turing outlined a two-room chess experiment that served as a proto-imitation game. He imagined three participants: a poor chess player (A), a mathematician operating a paper machine (B), and another poor chess player acting as an observer (C). The observer plays one game against A and one against the paper machine, with moves communicated between two separate rooms. “C may find it quite difficult to tell which he is playing,” Turing wrote. He added a revealing note about this setup: “This is a rather idealized form of an experiment I have actually done.”

The claim should not be inflated into a priority slogan. The 1948 passage does not give the number of games, the duration of play, or a formal success criterion. It is narrower and more useful than that. It shows the operational shape of the later test already present: separated rooms, a communication channel, a hidden machine procedure, and a human judge trying to infer which source produced the behaviour. The medium is still chess rather than open conversation, and the “machine” may be a paper procedure executed by a mathematician, but the epistemic move is recognizable. Intelligence is approached through a blinded comparison of performances.

The Parlour Game

In October 1950, Turing published “Computing Machinery and Intelligence” in the philosophy journal Mind, volume 59. The paper begins with a deliberate subversion: “I propose to consider the question, ‘Can machines think?’”

Having posed the question, Turing immediately rejects the method of finding an answer by surveying the common usage of the words “machine” and “think.” He dismisses the definitional debate as an “absurd” exercise akin to a Gallup poll. Instead, he writes, “such a definition I shall replace the question by another, which is closely related to it and is expressed in relatively unambiguous words.”

This new question is the Imitation Game. In its original form, the game is a Victorian parlour amusement played with three people: a man (A), a woman (B), and an interrogator (C) who may be of either sex. The interrogator stays in a separate room from A and B, knowing them only by the labels X and Y. The interrogator’s task is to determine which is the man and which is the woman. The game is asymmetric: A’s object is to make the interrogator identify wrongly, while B’s object is to help the interrogator decide correctly.

That original arrangement is easy to lose because the later phrase “Turing Test” invites a simpler picture: a human on one side, a machine on the other, and a judge trying to tell them apart. Turing’s text starts somewhere more particular. The human baseline is already a game of imitation, deception, and social inference before the machine enters it. The machine is not first compared to an abstract human essence. It is inserted into a game where one human party is already trying to mislead the interrogator under controlled communicative conditions.

To ensure the game tests only intellectual capacity, the interface must be severely constrained. “In order that tones of voice may not help the interrogator,” Turing specifies, “the answers should be written, or better still, typewritten. The ideal arrangement is to have a teleprinter communicating between the two rooms.” Alternatively, an intermediary may relay questions and answers.

The teleprinter is not decorative period machinery. It is the apparatus that makes the protocol possible. No voice, no face, no gait, no skin, no nervous gesture, no visible body can decide the verdict. The interrogator receives only typed linguistic behaviour, under labels that hide the source. Turing’s stated reason is operational: tones of voice must not help the interrogator. The broader effect is to draw a boundary around the evidence that counts. The game does not ask whether a machine can grow a human body or possess a human throat. It asks whether, inside a typed exchange, its answers alter the interrogator’s chance of making the right identification.

Having established this gender-disambiguation game as the baseline of human deception and detection, Turing makes his crucial pivot: “We now ask the question, ‘What will happen when a machine takes the part of A 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?’”

The gender structure of the 1950 setup establishes the control group. The machine does not have to achieve absolute deception. It only has to fool the interrogator as often as a man playing the same game would fool the interrogator. The human baseline of deception is the metric against which the machine’s performance is substituted and judged.

This is why the opening move is more radical than a clever anecdote. Turing does not define thinking, then test whether the machine satisfies the definition. He specifies a situation in which a human judge can be wrong, specifies the channel through which the evidence may pass, and asks whether replacing one participant with a machine changes the pattern of wrong judgments. The problem has become empirical without becoming simple.

The Argumentative Spine

The middle sections of the Mind paper form an argumentative spine designed to defend the Imitation Game from theoretical attacks. In Section 2, Turing provides a “Critique of the New Problem,” defending the decision to decouple intelligence from physical embodiment. “No engineer or chemist claims to be able to produce a material which is indistinguishable from the human skin,” he observes. And even if such a synthetic skin could be invented, “we should feel there was little point in trying to make a ‘thinking machine’ more human by dressing it up in such artificial flesh.” The teleprinter constraint operationalises this decoupling: it prevents the interrogator from seeing, touching, or hearing the competitors, ensuring the test measures symbolic and linguistic behaviour rather than anatomical mimicry.

The point is not that bodies are unimportant to human life. Turing’s claim is more limited and more procedural. If the question is whether a machine can perform intellectually, then a test that can be won by artificial skin, a convincing voice, or visual impersonation is badly designed. The new problem draws “a fairly sharp line” between physical and intellectual capacities. That line is not metaphysical; it is experimental. It says what the interrogator is allowed to use as evidence.

Sections 3 through 5 reframe the problem around digital computers. Turing narrows “machine” to the digital computer, then treats the relevant class as discrete-state machines: systems whose future behaviour is determined by their current state and input, moving through a finite, if enormous, space of possible configurations. From there he advances a universality argument. A single discrete-state machine equipped with adequate storage and speed can imitate the behaviour of any other discrete-state machine, provided it is given the right programme.

This reframe is what connects the parlour game to actual postwar hardware. Turing was not claiming that every physical detail of a nervous system could be reproduced in a machine room. He was arguing that the behaviour relevant to the test could be generated by a digital computer if storage, speed, and programming were sufficient. To ground the theoretical point in contemporary reality, he references the storage parameters of the Manchester machine, noting its 64 magnetic tracks each with a capacity of 2560, and eight electronic tubes with a capacity of 1280. The Manchester example does not prove that a 1950 machine could play the game. It shows that the argument is no longer floating entirely above hardware. Storage counts, instruction tables count, and real machines are close enough to the discussion to serve as reference points.

Section 6, “Contrary Views on the Main Question,” catalogues nine distinct objections to the idea of thinking machines: theological resistance, “heads in the sand” refusal, mathematical limitation, consciousness, lists of alleged disabilities, Lady Lovelace’s objection, continuity in the nervous system, informality of behaviour, and extra-sensory perception. The list is odd by design. It gathers high theology, professional mathematics, common prejudice, neurology, philosophy of mind, and even ESP into one argumentative space. Turing’s aim is not to prove that every objection is foolish. It is to show that the Imitation Game has been built to survive them.

One of the rebuttals Turing engages most fully is the objection attributed to Lady Ada Lovelace, who wrote that the Analytical Engine “has no pretensions to originate anything.” The core of the Lovelace objection, as Turing uses it, is that machines can only do whatever we know how to order them to perform; they cannot surprise us. Turing counters this by observing that machines take him by surprise frequently. Surprise alone is not proof of intelligence, since a programmer can be surprised by a bug, an unnoticed implication, or a badly understood mechanism. But it punctures the too-simple version of the objection. Knowing the rules that govern a system is not the same as foreseeing every consequence of those rules.

Turing then pushes the issue from surprise to generativity. He asks whether a mind can be “supercritical” in the sense of an atomic pile: “An idea presented to such a mind that may give rise to a whole ‘theory’ consisting of secondary, tertiary and more remote ideas.” The analogy is carefully chosen. A subcritical system absorbs an input without producing a chain reaction. A supercritical system can turn one initiating event into a cascade. In the context of learning machines, the question becomes whether a trained machine might be organized so that an input does not merely trigger a stored answer, but provokes further structures of response. The answer is left open, but the Lovelace objection has been weakened. It no longer suffices to say that the machine began as human design.

Turing also addresses the “Argument from Consciousness,” which asserts that a machine must feel emotions and be aware of its own thoughts to be considered intelligent. The biological form of that objection was associated with Geoffrey Jefferson’s 1949 Lister Oration, “The Mind of Mechanical Man,” though the chapter does not need Jefferson’s full text to see how Turing handles the demand. Turing’s reply, in §6.4, turns the demand back on itself: the strict insistence on first-person verification would refuse the same evidence we routinely accept for other people, who are credited with thought on the basis of outward behaviour rather than direct inspection. By the same logic, a machine that satisfies the imitation test should not be excluded by a standard no human interlocutor could satisfy either.

The adjacent “various disabilities” objection lets Turing attack a more everyday version of the same mistake. The sceptic offers a list of supposed limits — kindness, originality, error-making, sensory pleasure, emotional response, language use — and treats each as a definitive impossibility. The list looks powerful because it is concrete. Turing’s response is to refuse the list as a settled inventory of impossibilities. Some items confuse absence of present engineering with impossibility in principle; others smuggle consciousness back into the test; still others are exactly the sort of behavioural capacities the Imitation Game can probe. The pattern is the same as in 1947. The machine must not be judged by a rule invented to make its defeat automatic.

The Prediction and the Child Machine

The most frequently quoted passage of the Mind paper occurs at the opening of Section 6, where Turing offers a specific, quantitative forecast.

“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.”

This is not a binary finish line for artificial intelligence. It is a frequency claim about a population of average interrogators, bounded by a five-minute time limit, set against an estimated storage capacity. The number $10^9$ means one billion binary digits, about 125 megabytes if converted arithmetically into modern byte units, but that conversion should not be made to carry more meaning than Turing gave it. He was not predicting the internet or a consumer computer. He was estimating the storage needed for a machine to perform well enough in a particular typed interrogation game.

Immediately following this, Turing makes a companion claim about cultural reception rather than engineering: “I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.” The two predictions are related but not identical. One concerns machine performance under a specified protocol; the other concerns how educated speakers will use the word “thinking” after decades of contact with such machines. Turing’s epistemic wager is that usage will follow practice. If machines can operate inside the protocol well enough, the linguistic taboo around “thinking” will weaken.

Turing was well aware that no 1950-vintage machine could play the game. In Section 7 he estimates human-brain storage as lying somewhere from $10^{10}$ to $10^{15}$ binary digits, while adding that he inclines toward the lower values and that much of the store is probably used for visual impressions. He also notes that a storage capacity of $10^7$ would be practicable by contemporary techniques, and that the Encyclopaedia Britannica, eleventh edition, would occupy about $2 \times 10^9$ binary digits. These comparisons give the prediction scale. The imitation player is not imagined as magic; it is imagined as a difficult storage-and-programming problem.

He estimated that a human programmer could write about a thousand digits of code a day, meaning it would take sixty workers fifty years of flawless work to programme the imitation player by hand, if nothing went into the wastepaper basket. “Some more expeditious method seems desirable,” he noted. That dry sentence is the hinge of the paper. If hand-programming the adult mind is absurdly laborious, the solution is not to abandon the test. It is to change the way the machine is produced.

This leads to Section 7, “Learning Machines,” which reorients the entire engineering problem. Rather than attempting to programme an adult mind from scratch, Turing proposes programming a child mind and subjecting it to “an appropriate course of education.” The idea reaches back to the 1948 unorganized-machine report. Start with something closer to the infant cortex than to an adult expert; then organize it through training.

Note

“Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s?”

The hinge of the Mind paper. Turing turns intelligence from a hand-coded artefact into a staged engineering problem in which education does work that direct instruction cannot.

Turing’s notebook metaphor makes the proposal concrete. The child brain, he suggests, is something like a notebook bought from a stationer’s: little mechanism and many blank sheets. The metaphor can be misleading if read too softly. He is not saying the child mind is empty in a philosophical sense. He is separating initial structure from education and later experience so that each can become an engineering variable. A machine designer can alter the starting configuration, the teaching method, or the non-educational experiences to which the machine is exposed.

Turing divides the mind into three components: “(a) The initial state of the mind, say at birth, (b) The education to which it has been subjected, (c) Other experience, not to be described as education, to which it has been subjected.” He returns to the system of rewards and punishments he had outlined in 1948, remarking modestly that he had done “some experiments with one such child machine, and succeeded in teaching it a few things, but the teaching method was too unorthodox for the experiment to be considered really successful.” The modesty is important. Turing is not reporting a triumphant prototype. He is marking a line of attack that is still crude, experimental, and in need of better teaching methods.

To demonstrate that education does not strictly require conventional sensory embodiment, he cites the example of Helen Keller, noting that communication in both directions is sufficient. This argument matters because it prevents the child-machine programme from collapsing back into a demand for humanoid bodies. A machine may need channels for instruction and response; it need not begin with all the sense organs through which humans ordinarily learn.

Turing draws an explicit analogy between this educational design loop and biological evolution. The structure of the child machine corresponds to hereditary material; changes made to the child machine are mutations; and the judgment of the experimenter takes the role of natural selection. The analogy turns the engineering problem into an iterative search. Build a child machine, educate it, measure its behaviour, alter the design, and repeat. In 1950 this was not yet an industrial research programme, but it was a recognizably modern way to think about machine intelligence: not as a single hand-coded adult intellect, but as a system whose performance emerges through repeated adjustment.

As the paper concludes, Turing identifies two possible paths forward. One approach would focus on highly abstract activities, such as chess. This path links back to the 1947 lecture and the 1948 paper-machine experiment: chess offered a bounded domain where moves could be represented symbolically and judged by play. The alternative would be to provide the machine “with the best sense organs that money can buy,” and teach it to understand and speak English. That path aims toward the richer conversational behaviour the Imitation Game requires. Refusing to declare a winner between the symbolic and the embodied approaches, Turing writes: “Again I do not know what the right answer is, but I think both approaches should be tried.”

The final sentence of the Mind paper reads: “We can only see a short distance ahead, but we can see plenty there that needs to be done.”

Aftermath: BBC and Reframing

Turing returned to the question in public after the paper’s publication. On 14 January 1952, the BBC Third Programme broadcast a four-way conversation titled “Can Automatic Calculating Machines Be Said To Think?” The panel included Turing, his Manchester colleague Max Newman, the Cambridge philosopher R. B. Braithwaite, and Geoffrey Jefferson.

The broadcast brought the theoretical debates of the Mind paper into a public conversational setting. The source record available for this chapter supports a cautious version of the scene: Jefferson’s role belongs to the biological-objection line already visible in the 1949 Lister Oration and in Turing’s “Argument from Consciousness” discussion, while Turing’s position remains the behavioural-test framing of the Mind paper. Without quoting the broadcast transcript, the important point is the cast and the setting. Within two years, the question had become a conversation among mathematics, neurosurgery, philosophy, and the author of the test.

Over the next fifty years, the reception of Turing’s paper shifted through various disciplines, as surveyed by Ayse Pinar Saygin, Ilyas Cicekli, and Varol Akman in 2000. Their survey tracks philosophical responses in the 1960s and 1970s, computational responses in the 1980s, and later arguments over anthropomorphism, gender, and the mechanics of the game in the 1990s. That arc should not be mistaken for steady clarification. It also records drift.

During this half-century, the “Imitation Game” underwent a gradual rebrand. The term “Turing Test” became the dominant label. Along with the name change came a subtle structural shift. The original 1950 setup—where the machine takes the part of A in a man-woman imitation game—was frequently simplified into a straightforward human-versus-machine test.

Whether this abstraction was faithful to Turing’s intent remains disputed. Diane Proudfoot has argued that subsequent commentary stripped out the gender-disambiguation structure upon which Turing built the test. In that reading, the “abstracted” Turing Test, where the interrogator merely tries to distinguish human from machine, is a different test from what Turing proposed. Other commentators have treated the gender axis as a vestigial feature that was meant to be abstracted away. The responsible historical point is not to settle that disagreement here, but to keep the original structure visible.

Regardless of whether the gender structure was load-bearing, Turing’s core epistemic move survived. By substituting an empirical, teleprinter-bound imitation game for the unanswerable question of whether machines can think, Turing set an operational baseline. The rest of the field would repeatedly argue about whether the baseline was sufficient, fair, too behavioural, too linguistic, too anthropomorphic, or too easy to game. But those arguments took place on the ground Turing had cleared: do not begin with a definition of thinking; begin with a test of what a machine can make a competent interrogator believe.

Why this still matters today

Every benchmark that asks whether a model can hold a conversation indistinguishable from a person’s descends from Turing’s move. So does every “AI evaluation” panel that compares a model’s typed output against a human’s, every chatbot transcript a tester scores, every adversarial probe designed to catch a machine in a tell. The framing is so naturalised that it is easy to forget it had to be invented — and that the invention was an act of refusal. The adequacy of the setup remains disputed; its durability lies in forcing the dispute into observable behaviour.