the Turing test

programmingops culture

Alan Turing's 1950 proposal to judge machine intelligence by whether it can converse indistinguishably from a human.

Rather than ask 'can machines think', Turing reframed it as an imitation game: if a human judge cannot reliably tell the machine from a person by text alone, the question is answered pragmatically. It set the terms for decades of debate about artificial intelligence.

Turing's 1950 paper proposed replacing the unanswerable question of whether machines can think with an operational one: can a machine's responses be distinguished from a human's by an interrogator communicating through text. He called it the imitation game, and framed it explicitly as a way to sidestep definitions nobody could agree on.

The test has been criticized continuously and productively. Searle's Chinese Room argues that behaviour indistinguishable from understanding is not understanding, and the practical objection is that the test rewards deception rather than intelligence, since a system optimized to pass is optimized to seem human, which includes pretending to be worse at arithmetic than it is.

What large language models did was collapse the test's usefulness rather than settle it. Systems now produce text that convinces many people in casual conversation, and the reaction has largely been to conclude that the test was measuring the wrong thing rather than to accept the result. Turing anticipated some of this and would probably have found it interesting: the paper is more careful and more provisional than its reputation, and it was always about what we can observe rather than about what is really happening inside.

Also known as: imitation game, Turing test

← All glossary entries