Imitation Game
The imitation game is the party game Alan Turing described at the start of "Computing Machinery and Intelligence" (1950) as a stand-in for the question "Can machines think?" In its original form it has three players: a man (A), a woman (B) and an interrogator (C) of either sex who sits in a separate room. The interrogator knows the other two only as labels X and Y and must decide, by written questioning alone, which is the man and which is the woman. The setup is the subject of this entry; what the game later became is the business of the turing-test entry, and whether it settles anything is the business of can-machines-think.
The two hidden players are given opposed goals, and that asymmetry is the heart of the design. The man's task is to cause the interrogator to make the wrong identification, so he must pretend to be a woman; the woman's task, in Turing's words, is to help the interrogator, so she answers truthfully and may add remarks of her own. Turing suggested that the answers be written, or better still typed or relayed through a teleprinter, so that tone of voice and handwriting cannot give the game away. The questioner is thus playing detective with language as the only evidence, and the deceiver is playing a character, which means the game rewards flexible, contextual, adversarial skill and not a fixed list of knowledge.
The substitution step is stated as a comparison of success rates, not as a pass mark. After describing the game, Turing asks what will happen when a machine takes the part of A, and whether the interrogator will then decide wrongly as often as when the game is played between a man and a woman. Read literally, the benchmark is the rate at which the interrogator is fooled in the human version, so the machine is being compared to a man impersonating a woman, which is itself a performance of imperfect success. Later in the paper Turing restates the target more loosely, as a computer playing so well that an average interrogator has no more than a seventy percent chance of a correct identification after five minutes, and he moves on to a version without the gender framing.
Why the game begins with gender is genuinely unsettled. Turing did not explain the choice. Commentators have suggested several readings: that it gave a familiar, concrete task in which one player must imitate a category they do not belong to; that it separated sincere from deceptive answerers; or that it made the machine's task a comparison against a human baseline. These are interpretations, not documented intentions, and a careful entry should say so. What is documented is that Susan Sterrett, in a 2000 article in the journal Minds and Machines, argued that Turing offered two distinct tests, the gender-imitation comparison and a machine-versus-human test, and that the standard textbook version quietly merges them.
The game has an earlier relative in Turing's own work. In his 1948 report "Intelligent Machinery," written for the National Physical Laboratory, he sketched a party game involving a chess-playing procedure carried out by hand, in which a player facing a chess opponent through an intermediary tries to tell whether it is a human or the paper machine. That precursor shows the idea was an engineering-minded one: judge a system by how it performs in an interaction, not by what it is made of. The conversational examples in the 1950 paper, including a poetry exchange about a sonnet and an arithmetic problem in which the machine would be wise to make a mistake on purpose, show the same preference for questions that test flexibility and plausibility at once.
Three misconceptions surround the setup. First, the game is not a certificate of consciousness, because the interrogator sees only text and Turing explicitly set aside questions of the machine's inner life. Second, the machine is not supposed to answer perfectly; deliberate imperfection is part of imitation, which is why a pure calculator would give itself away. Third, the single-judge, five-minute framing is not a standardized experiment: Turing's remark was a rough forecast, and the later contests that adopted similar rules, often fooling judges through evasion and chat tricks, tell more about the judges than about the machines. Modern chatbots make the setup newly relevant, because text-only conversation with an unseen partner is now an everyday experience, and the question of whether such a conversation can be trusted as evidence of understanding is live again.
Read the imitation game as a 1950 experimental protocol: hide the bodies, keep the language, give one contestant a motive to deceive, and ask whether a machine can occupy a role previously held by a person.