What is the Turing test? Can a machine pass as human?

The Turing test is a way of judging a machine's intelligence by its conversation alone: a human judge exchanges typed messages with a hidden person and a hidden machine, and the machine passes if the judge can't reliably tell which is which. Alan Turing proposed it in 1950, as "the imitation game", to replace the question "Can machines think?" with one that could be tested.

In short

• Turing swapped a question nobody could define, "can machines think?", for one anyone could run: can a machine pass for human in conversation?
• It tests behavior only, which is both its strength and the root of every major criticism of it.
• In a 2025 study, a chatbot told to play a persona was picked as the human 73% of the time in five-minute chats, which shows how far imitation has come, not that the debate is over.

How did Turing set up the imitation game?

Turing began his paper "Computing Machinery and Intelligence", published in the journal Mind in 1950, with a party game. A man (A) and a woman (B) sit in another room, and an interrogator (C) puts written questions to both. The man tries to make the interrogator guess wrongly which is the woman; the woman tries to help. Answers are typed, ideally by teleprinter, so voices and handwriting give nothing away.

Then he changed one player. "What will happen when a machine takes the part of A in this game?" If the interrogator guesses wrongly as often with a machine in the room as with a man, Turing suggested, the original question has been answered as well as it can be. He called "Can machines think?" itself "too meaningless to deserve discussion."

He also made a prediction. In about fifty years, he wrote, computers with a storage capacity of about 10⁹ (roughly 125 megabytes) would play the game so well that "an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning." That is a modest bar: the machine needs to fool the judge only 30% of the time, and only for five minutes.

Why was it a clever idea?

Because it sidesteps the hardest part of the question. Nobody agrees on what thinking is, or how to look for it inside a brain or a machine. Turing proposed judging machines the way people already judge each other: by what they say and do. Nobody can see into another person's mind either; Turing noted that people get along by "the polite convention that everyone thinks."

The test is also fair in a particular way. The judge can't see the players, so the machine isn't marked down for having no body, no face or a metal voice. And the judge can ask anything, from arithmetic to poetry to "what did you have for breakfast?", so the test probes general ability, not one narrow skill.

What objections did Turing answer himself?

Turing listed nine objections to the idea that machines could think, and replied to each. His list still covers most of the debate.

Objection	The claim	Turing's reply, in brief
Theological	Thinking needs a soul, and machines have none	He rejects it, and notes it would limit what God could do
"Heads in the sand"	The consequences would be too dreadful	A hope, not an argument
Mathematical	Gödel's theorem shows limits on machines	Humans have limits too, and no one has shown we escape them
Consciousness	A machine can't feel what it writes	Then we couldn't know that other people think either
Various disabilities	A machine can never be kind, fall in love, enjoy strawberries	Based on the limited machines people had seen
Lady Lovelace	A machine can only do what it's told	Machines surprise him often
Nervous system	The brain is continuous, not digital	A digital machine can imitate it well enough for the game
Informality	No rules could cover all human behavior	Lacking complete rules isn't the same as lacking laws of behavior
Extrasensory perception	Telepathy would give humans away	He took it seriously, and proposed a "telepathy-proof room"

What are the main criticisms?

The deepest is that passing proves only imitation. In 1980, the philosopher John Searle described the Chinese room: a person following a rulebook can produce fluent Chinese replies without understanding a word. If that's possible, a machine could pass the test with no understanding at all. The philosopher Ned Block made the point in 1981 with a thought experiment of his own: a machine that stored a canned reply to every possible conversation would pass, yet nobody would call a lookup table intelligent.

A second criticism is that the test rewards deception and human quirks. Early chatbots did best by dodging questions, making typos, or claiming a reason for their ignorance. In 2014, a program called Eugene Goostman convinced 33% of judges at an event in London, in part by posing as a 13-year-old Ukrainian boy whose gaps could be excused. Critics said that showed how to game the test, not how to think.

A third is that it is too human-centered and too narrow. A system might be very intelligent in ways that don't resemble human chat, or fail the test only because it is too fast or knows too much. Many researchers now say the test is a poor measure of progress in artificial intelligence, and toward artificial general intelligence in particular, because conversation is only one of many abilities.

Has any machine passed the Turing test?

In the strict three-player form, a 2025 study is the strongest claim yet. Cameron Jones and Benjamin Bergen of the University of California San Diego ran randomized, preregistered tests in which participants chatted for five minutes with a human and a machine at once, then chose which was human. When told to adopt a humanlike persona, GPT-4.5 was judged to be the human 73% of the time, more often than the real humans. LLaMa-3.1-405B with the same prompt was picked 56% of the time; ELIZA, a 1960s chatbot, 23%, and GPT-4o without the persona prompt, 21%.

That meets the bar Turing set for the year 2000. It doesn't settle what he wanted to know. The conversations were short, the judges were ordinary participants rather than experts, and the result depended heavily on the persona prompt. The language models behind modern chatbots are trained to predict human text, which makes sounding human their strongest skill; how ChatGPT works explains why.

Still argued

Whether passing the Turing test shows intelligence is as contested as it was in 1950. One side holds, with Turing, that behavior is the only evidence we ever have for any mind, so a machine that converses like a person has earned the same courtesy. The other side, Searle's among them, holds that convincing talk can come from pattern-matching with no understanding behind it. The 2025 results sharpen the question rather than answer it.

Check yourself

• Why did Turing replace "Can machines think?" with the imitation game?
• What exactly did Turing predict for the year 2000, and how does the 2025 study compare?
• What is the strongest reason to doubt that passing the test proves a machine understands?

Sources

• Computing Machinery and Intelligence — A. M. Turing, Mind, 1950
• The Turing Test — Graham Oppy and David Dowe, Stanford Encyclopedia of Philosophy, revised 2021
• Large Language Models Pass the Turing Test — Cameron R. Jones and Benjamin K. Bergen, 2025
• Turing test success marks milestone in computing history — University of Reading, via ScienceDaily, 2014