Computing Machinery and Intelligence

Turing's paper "Computing Machinery and Intelligence" appeared in October 1950 in Mind, a philosophy journal, in volume 59, number 236, pages 433 to 460.^[1] Its subject was whether machines could be intelligent, a question people had argued about for a long time without reaching agreement.

The problem

The paper opens with a direct sentence: "I propose to consider the question, 'Can machines think?'" Turing judged that question too vague to be useful as it stood, so he replaced it with a more concrete test.^[2]

The method: the imitation game

In the original game, an interrogator communicates only in writing, ideally through a teleprinter. The interrogator talks with two hidden players, a man (A) and a woman (B), and must work out which is which. Turing then asks what happens when a machine takes the part of A. The arrangement is now known as the Turing test.^[2]

Here is how it might look in practice. The interrogator could ask each hidden player to write a short poem about the sea, or to solve a sum and explain the steps. The only evidence is the text on the screen. If the machine's answers are good enough that the interrogator takes it for a person, the machine has succeeded in the game.

Turing also made a concrete prediction. He suggested that in about fifty years, computers with a storage capacity of about 10^9 could be programmed to play the game so well that an average interrogator would have no more than a 70 per cent chance of making the right identification after five minutes of questioning.^[2]

The arguments

Turing describes digital computers as having three parts: a store, an executive unit and a control. He recalls Charles Babbage's Analytical Engine, which was never completed, and argues in a section on the universality of digital computers that one such machine can mimic any discrete-state machine. This universality is one reason he thought the question was worth taking seriously.^[2]

The paper answers nine objections to machine intelligence. They include the Theological Objection, the "Heads in the Sand" Objection, the Mathematical Objection, which draws on Gödel's theorem and on results by Church, Kleene, Rosser and Turing, the Argument from Consciousness, and Lady Lovelace's Objection, that a machine can only do what it is told to do.^[2]

The final section, "Learning Machines," proposes a different route. Instead of programming an adult mind directly, the programmer would build a "child machine" and educate it, using rewards and punishments, and the paper compares this process to evolution.^[2]

Turing closes by suggesting that work start with abstract activities such as chess, or with machines that have good sense organs and are taught to understand and speak English. His last line is: "We can only see a short distance ahead, but we can see plenty there that needs to be done."^[2]

Why it mattered

The paper's contribution is to turn a vague question into a test that can be argued about and checked. Instead of asking what a machine is "really" doing inside, it asks whether a judge, reading a conversation, can tell the machine from a person. Its answers to objections also gave later discussions a clear list of points to address.^[2]

The paper also connects two of Turing's major pieces of work. Its discussion of the Mathematical Objection cites Turing's own 1937 computability result, which links the 1950 argument back to his 1936 work on computable numbers.^[2]

What followed

The imitation game became known as the Turing test, a name still in use today. It remains a common reference point in discussions of machine intelligence, and the paper's list of objections continues to be a useful checklist for anyone assessing claims about machines that seem to think.