Can Machines Think? Turing and the Origin of a Question

Long before computers could do much more than arithmetic, a few people were already asking whether a machine could ever be said to think. The most influential of them was the British mathematician Alan Turing.

A machine on paper

In 1936 Turing submitted a paper “On Computable Numbers, with an Application to the Entscheidungsproblem” to the London Mathematical Society. To answer a question in mathematical logic, he described an imaginary device that reads and writes symbols on an unlimited tape, following a finite table of rules. The device was never meant to be built. Its purpose was to make precise what it means to carry out a procedure step by step.

The paper also showed that a single machine of this kind could imitate any other, provided it was given a description of that other machine as input. This “universal” machine is the abstract model behind every general-purpose computer: one device, many possible programs. Once that was clear, it was natural to ask what such programs might eventually be able to do.

The imitation game

In 1950 Turing published “Computing Machinery and Intelligence” in the philosophy journal Mind. The essay opens with the question “Can machines think?” and then sets it aside as too vague to answer directly. In its place Turing proposed a game. An interrogator exchanges typed messages with two hidden participants, one human and one machine, and tries to tell which is which. If the interrogator cannot reliably do so, Turing argued, we should have little reason to deny the machine some form of thinking.

The proposal replaced an argument about inner experience with something that could, at least in principle, be observed. The same essay also works through a series of objections — among them theological arguments, the view that machines can only do what they are told, and the argument from consciousness — and Turing’s replies remain a readable introduction to the debate.

What the test does and does not measure

The imitation game is often treated as a finish line for artificial intelligence, but it measures something narrow: whether conversational behaviour is convincing. It says nothing direct about understanding, about acting in the physical world, or about whether a system is reliable when nobody is checking.

People are also easily impressed by fluent language. In 1966 Joseph Weizenbaum of MIT described ELIZA, a program that produced replies by matching key words in the user’s sentence and reassembling them according to simple rules. Its best-known script imitated a certain style of psychotherapist. In the same paper Weizenbaum discussed the psychological questions raised by how people responded to such a simple program.

Why the question still matters

Modern language systems can hold long, fluent conversations, so the imitation game is no longer a distant hypothetical. Yet the original question has not gone away. Researchers increasingly evaluate systems on specific abilities — solving problems, following instructions, stating uncertainty — rather than on whether they can pass as human.

That shift is, in a sense, faithful to Turing. He did not claim to have settled what thinking is. He offered a practical way to make progress on a question that otherwise invites endless argument.

Sources

1. Turing, A. M. “On Computable Numbers, with an Application to the Entscheidungsproblem”. Proceedings of the London Mathematical Society, s2-42(1), 230–265. doi:10.1112/plms/s2-42.1.230
2. Turing, A. M. “Computing Machinery and Intelligence”. Mind, LIX(236), 433–460, 1950. doi:10.1093/mind/LIX.236.433
3. Weizenbaum, J. “ELIZA — A Computer Program for the Study of Natural Language Communication Between Man and Machine”. Communications of the ACM, 9(1), 36–45, 1966. doi:10.1145/365153.365168