What is Turing Test?
The Turing test is a proposed criterion for machine intelligence, introduced by Alan Turing in 1950 as the imitation game, in which a human judge holds text conversations with a person and a machine and tries to tell them apart. If the judge cannot reliably do so, the machine is said to pass. Its status as a measure of intelligence is contested.
Turing's own framing was partly a move to replace an unanswerable question. Rather than asking whether machines can think, which he considered too poorly defined to settle, he substituted an operational question about observable conversational behavior. The original paper describes variations of the setup and anticipates several objections, including appeals to consciousness and to the supposed limits of machine originality.
Criticism falls into several lines. The test measures the ability to imitate human conversation, which is narrower than and partly separate from general capability, so a system might pass through mimicry or by exploiting a judge's expectations. The Chinese room argument holds that symbol manipulation adequate to pass would still not constitute understanding. Others note that the setup rewards deception, including feigned ignorance.
Practical results have not settled anything. Early conversational programs elicited belief from users despite trivial mechanisms, and later chat systems have been reported to fool judges in restricted settings. Such reports depend heavily on the judges, the time allowed, the topic, and the instructions given, so there is no single accepted protocol and no consensus that any system has passed in a meaningful sense.
The test now retains historical and conceptual importance rather than practical use. Contemporary evaluation relies on task benchmarks, human preference ratings, and domain specific measures, none of which claim to detect intelligence in general. Discussion of the Turing test functions largely as a way to sharpen what is actually being claimed when a system is called intelligent.
Key points
- Proposed by Alan Turing in 1950 as the imitation game.
- A judge tries to distinguish machine from human by conversation.
- Measures imitation of dialogue, not capability in general.
- Contested, with no accepted protocol and no consensus pass.
- Now mostly of historical and conceptual interest.
In practice
In a typical setup a judge exchanges typed messages with two hidden participants for a fixed period, then names which one is the machine. A program can improve its odds by claiming to be a distracted teenager, deflecting hard questions, and making typing errors. Success of that kind demonstrates imitation and misdirection, which is why the result is disputed.