Computing Machinery and Intelligence: Turing's imitation game (1950)

In October 1950, the British mathematician Alan Turing published a paper in the philosophical journal Mind that would become one of the most cited texts in the history of artificial intelligence. Its title, Computing Machinery and Intelligence, was modest. Its opening question was not: “Can machines think?”

Turing immediately dismissed that question as too vague to be useful. Instead, he proposed replacing it with something observable: a game.

The imitation game

Turing described a setup with three participants: a human interrogator (C), a human respondent (B), and a machine (A), each in separate rooms, communicating only through typed messages. The interrogator’s task: figure out which respondent is the machine and which is the human.

If the machine can fool the interrogator as often as a real human would, Turing argued, then the question “Can machines think?” has been answered, or, more precisely, replaced by a question that can actually be tested.

He called this the imitation game. It was later popularised as the Turing test.

Nine objections, and Turing’s replies

The bulk of the paper is Turing responding to anticipated objections, each with characteristic clarity:

1. The theological objection, thinking is a function of the soul, which only humans have. Turing noted that similar arguments had been wrong before (e.g., opposing Galileo).
2. “Heads in the sand”, the consequences of machines thinking would be too dreadful, so we prefer not to believe it. Turing dismissed this as emotional, not logical.
3. The mathematical objection, Gödel’s incompleteness theorems show limits to formal systems. Turing replied that humans are also subject to limitations and make errors.
4. The argument from consciousness, unless a machine genuinely feels, it cannot truly think. Turing considered this solipsistic: we don’t demand proof of consciousness from other humans either.
5. Arguments from various disabilities, “a machine can never enjoy strawberries” or “fall in love.” Turing viewed these as an argument from lack of imagination, not from principle.
6. Lady Lovelace’s objection, machines can only do what they are told. Turing disagreed, arguing that machines could surprise their creators.
7. The argument from continuity in the nervous system, the brain is not digital. Turing argued that a digital simulation could approximate any continuous system closely enough.
8. The argument from informality of behaviour, humans don’t follow fixed rules. Turing questioned whether that was truly the case.
9. Extra-sensory perception, Turing treated this semi-seriously, suggesting telepathy-proof conditions for the test.

A prediction

Turing closed with a specific forecast (recorded in our predictions collection): by the year 2000, computers with about 10⁹ bits of storage would be able to play the imitation game well enough that “an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning.” We evaluate that claim against 2000 as partially true, storage scale yes, the behavioural bar in a fair blind test, no.

That prediction was optimistic, but the framework endured. When today’s large language models pass or fail conversational benchmarks, they are being measured, consciously or not, against the standard Turing proposed in 1950.

Why it matters

Turing did not answer the question of machine intelligence. He reframed it, from an unanswerable philosophical puzzle to a testable behavioural criterion. That move influenced decades of AI research, and it remains at the centre of debates about what intelligence means when the system producing it is not biological.