A. M. Turing (1950), Mind LIX (N.S. No. 236), October 1950, pp. 433–460 — DOI-10.1093/mind/LIX.236.433. A scan of Turing’s off-print is open in the Turing Digital Archive, King’s College Cambridge, item AMT/B/19 (https://turingarchive.kings.cam.ac.uk/publications-lectures-and-talks-amtb/amt-b-19), reproduced with Oxford University Press’s permission; his typescript is AMT/B/9. Reprinted as selection 4 of The Mind’s I (LIT-413).
Key takeaways
- The replacement, not a definition. “Can machines think?” is “too meaningless to deserve discussion” (§6); deciding it by the ordinary use of the words would make it a Gallup poll (§1). It is replaced by the imitation game: can a machine take A’s part, in a game where an interrogator questioning A and B by teleprinter tries to tell which is the man, as well as a man does?
- Machines are digital computers, and universality makes one enough. Only digital computers are admitted (§3). Since a digital computer with enough store and speed can mimic any discrete-state machine (§5), the question becomes whether one particular computer, suitably programmed, can play the part.
- A dated prediction. “In about fifty years’ time” a computer with a storage capacity of about 10^9 will play “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” (§6).
- Nine objections answered. Theological, “heads in the sand”, mathematical (Gödel, Turing), consciousness, various disabilities, Lady Lovelace, continuity of the nervous system, informality of behaviour, and extra-sensory perception, which Turing calls “quite a strong one”.
- Consciousness. Jefferson’s demand that a machine feel what it writes leads, at its extreme, to solipsism; the “polite convention that everyone thinks” should be extended. Turing grants “something of a paradox” about localising consciousness, but says it need not be solved first.
- Learning machines (§7). Programme a child-machine and educate it, with rewards and punishments plus “unemotional” channels, a built-in logical system governed by imperatives, and a random element. The teacher will be “very largely ignorant of quite what is going on inside”.
Standing in the record
Filed on 2026-10-02 for The Mind’s I (LIT-413), which reprints it, in the batch the owner asked for on Daniel Dennett. It is the root of the record’s machine-minds thread, and until now the record held it only by name. NOTE-136 reports Hofstadter’s later claim that a demanding Turing test suffices to attribute thinking (his essay is LIT-416, unread). NOTE-106 reads Angius et al. (LIT-153) as asserting that “the passing of Turing’s Imitation Game can now be considered empirically established”. That is a claim about Turing’s §6 test, and this reading shows what the test was: five minutes, an average interrogator, a 70 per cent threshold, a man and a machine competing.
Bearing on THEORY-023. That account holds that behavioural evidence cannot settle whether an AI system is conscious, because mimicry undercuts it. Turing’s paper is the case for treating behaviour as settling the different question of thinking, and it is careful about the difference. His answer to Jefferson does not say a passing machine is conscious. It says that demanding inner evidence before ascribing thought proves too much, because it would leave us unable to ascribe thought to one another. Birch’s gaming problem (LIT-111) and Schwitzgebel’s mimicry argument (LIT-191) are objections Turing did not consider: that a machine trained on human output might pass by imitation without the capacity the test was meant to detect. His own §2 remark that the machine’s best strategy is “to try to provide answers that would naturally be given by a man” is the assumption those arguments attack. This supports THEORY-023’s diagnosis that the dispute is over what counts as evidence, and adds nothing to either side of it.
Bearing on the minds-of-parts thread. The “skin of an onion” passage (§7) asks whether, stripping off the mechanically explicable operations of mind, we reach a “real” mind or a skin with nothing in it, “in which case the whole mind is mechanical”. That is the downward decomposition THEORY-043 attributes to Minsky and Selfridge, stated a decade before Pandemonium (LIT-427), and offered by Turing only as a “recitation tending to produce belief”.
Searle’s Chinese room (LIT-400), the best-known argument that passing this test would not show understanding, is filed beside it. Chalmers’s work on what it is for a physical system to implement a computation, being filed now by other readers, bears on the reduction in §5, which treats running the right programme as all that matters. Dennett’s own writing on the test, if it is filed with his other work in this batch, belongs beside this paper.
Boundary. The anthology does not hold it. §7 is a sketch of machine
learning (a child-machine, reward and punishment signals, random search,
a teacher who cannot see inside), and BIG-bench in the anthology (ANTH-LIT-077) takes its
name from the game. Under ADR-013 the anthology could hold the paper for
§7 as the earliest statement of learning-instead-of-programming. That is
for the anthology to decide, so anthology-candidate is not set: the
paper’s subject is whether machines think, which is this record’s question.