In plain words
In 1950 the British mathematician Alan Turing asked: can machines think? He felt the question was too vague to answer, so he proposed a game instead.
A judge types messages to two hidden partners. One is a person, one is a machine. Both try to convince the judge they are the human. If the judge cannot reliably pick out the machine, the machine passes.
Turing called it the imitation game. Its appeal was that it needed no definition of thinking, only behaviour that anyone could test.
For decades no machine came close. Today's chatbots changed that. In a 2025 study, judges picked a language model told to act human as the person more often than the real person. That says a lot about how well AI imitates conversation, and much less about whether it thinks.
Everyday example
Who is really texting?
Your friend's younger brother grabs her phone and keeps chatting with you, pretending to be her. Can you tell? You might ask something only she would know, or notice he types differently. If he fools you for five minutes, he has passed your own little Turing test. Turing's test asks the same of a machine.
Where the example breaks: Fooling you proves the brother can imitate his sister, not that he is her. In the same way, passing the test shows a machine can imitate a person in text, not that it thinks or feels.
How it actually works
Turing's 1950 paper set out the game, a prediction and the objections he expected.
The original game
Turing began with a parlour game: a judge tries to tell a man from a woman through typed answers while the man tries to mislead. He then asked what would happen if a machine took the man's place.
Turing's prediction
He guessed that in about fifty years a computer could play so well that an average judge would have no more than a 70 per cent chance of identifying it correctly after five minutes of questioning.
Early tricks
ELIZA, a rule-based chatbot from the 1960s, fooled some people with simple pattern matching. It showed how readily people read intelligence into text, and made many researchers treat the test as a weak measure of ability.
Passing it in 2025
In a pre-registered three-party study, Cameron Jones and Benjamin Bergen found GPT-4.5, prompted to play an introverted young person who uses slang, was judged to be the human 73% of the time in five-minute chats. ELIZA scored 23%.
What it does not measure
The test checks imitation, not truth, reasoning or understanding. Philosopher John Searle's 1980 Chinese Room argument claims a system could pass while understanding nothing. That is why researchers rely on many benchmarks instead.
- Proposed
- October 1950, in "Computing Machinery and Intelligence", in the journal Mind
- By
- Alan Turing, British mathematician and computing pioneer
- Turing's name for it
- The imitation game
- 2025 result
- GPT-4.5 with a persona prompt was judged human in 73% of games (Jones and Bergen, 2025 preprint)
Questions people ask
Has any AI passed the Turing test?
A 2025 study by Cameron Jones and Benjamin Bergen reported that GPT-4.5, told to adopt a humanlike persona, was picked as the human 73% of the time in five-minute, three-party tests. The authors call this the first evidence of a system passing a standard Turing test; many note it measures imitation, not intelligence.
What is the point of the Turing test?
Turing wanted to replace the vague question "Can machines think?" with something testable. If you cannot tell a machine from a person by conversation alone, he argued, it would be odd to refuse to call it intelligent.
Why is the Turing test criticised?
It rewards fooling people, not being correct or useful. Tricks like deliberate typos and dodging questions can help a program pass. And imitating human talk is different from understanding it, as John Searle's Chinese Room argument claims.
Is a CAPTCHA a Turing test?
A kind of one. CAPTCHA stands for Completely Automated Public Turing test to tell Computers and Humans Apart. The difference is that a computer is the judge, setting puzzles that were once easy for people and hard for machines.