In five minutes
The Turing test is a way of asking whether a machine can talk like a person. A human judge chats by text with two hidden partners, one a person and one a machine. If the judge cannot reliably say which is which, the machine has passed. The English mathematician Alan Turing described the idea in 1950, and it remains the best-known test in the history of artificial intelligence.
It is also one of the great ideas of science fiction. Writers have used it to ask what makes a mind, how we would recognise a new one, and what we owe to it. This page is part of our guide to the History of Science Fiction. Below you will find how the test works, what it does and does not prove, and where stories use it.
How the Turing test works
Turing did not call it a test. He called it the imitation game. In his 1950 paper "Computing Machinery and Intelligence", published in the journal Mind, he started from a question he thought too vague to answer: "Can machines think?" He proposed swapping it for a question that could be checked in practice.
His original game has three players. A man and a woman sit in separate rooms. A judge, the interrogator, sits in a third room and sends them written questions. The judge's task is to work out which one is the woman, while the man tries to fool the judge. Turing then asks what happens if a machine takes the part of the man. Will the judge be wrong as often as before?
Most people now describe a simpler version:
- A judge types questions to two hidden partners.
- One partner is a human, the other is a computer program.
- Both try to seem human, or at least the program does.
- After a set time, the judge says which is the machine.
- If the judge guesses wrong often enough, the machine passes.
Everything happens through text on purpose. Turing wanted to remove voice, face and body, so that the judge could only weigh what was said. The test measures behaviour in conversation, not what happens inside the machine.
What Turing predicted
Turing made a modest forecast. He thought that in about fifty years, so around the year 2000, computers would be good enough that an average judge would have no more than a 70 percent chance of picking the machine after five minutes of questioning. In other words, the machine would fool the judge about three times in ten. He did not say that such a machine would be conscious. He argued that if its answers were as good as a person's, arguing about whether it "really" thinks would not get us far.
Where the idea came from
Turing was already well known among mathematicians. In 1936 he described an abstract device, now called the Turing machine, that became the basis for the theory of computing. During the Second World War he worked at Bletchley Park on breaking German codes. After the war he turned to the practical question of building machines that could do more than arithmetic.
The 1950 paper is playful and careful at once. Turing listed objections he expected to hear and answered each one. Some were religious, some mathematical, and one was that machines could never be creative or surprise us. His reply to the last was simple: machines surprised him all the time. He also suggested that the better route to a thinking machine might be to build a childlike program and then teach it, not to write an adult mind from scratch.
The real science
The Turing test is a thought experiment that later people tried to turn into a contest. Here is what is established, and what is still argued about.
Early programs that fooled people
In 1966 the computer scientist Joseph Weizenbaum at MIT wrote ELIZA, a short program that answered users by turning their own sentences into questions, in the style of a therapist. Weizenbaum was surprised that some users treated it as if it understood them. ELIZA used simple pattern matching and understood nothing. It showed that people are quick to read a mind into a few well-chosen words.
In 1972 the psychiatrist Kenneth Colby built PARRY, which imitated a person with paranoia. In a test with psychiatrists, they could not reliably tell its transcripts from those of real patients. This was not the Turing test as Turing described it, but it pointed the same way: a narrow character is easier to imitate than a general one.
Contests and claims
The Loebner Prize, first held in 1991, offered a yearly award for the most human-seeming chatbot. For years the winners were clever tricks: typing errors, jokes and dodges. In 2014 organisers at the Royal Society in London announced that a program called Eugene Goostman, which posed as a 13-year-old Ukrainian boy, had passed the test by convincing about a third of its judges. Many researchers rejected the claim. The program leaned on its imagined age and on changing the subject, and the pass mark was Turing's rough forecast, not a rule he set.
In 2025 researchers at the University of California San Diego ran a three-party test in which judges chatted with a person and an AI at the same time for five minutes. A large language model, GPT-4.5, was judged to be the human about 73 percent of the time when it was given a persona to play. That was more often than the real people were. It is the clearest sign yet that chatting convincingly is no longer a hard target for modern language models.
What the test does not show
The philosopher John Searle argued in 1980 against the idea that passing proves understanding. His Chinese Room thought experiment imagines a person who does not know Chinese, following a rulebook to answer Chinese messages. From outside, the answers look fluent. Inside, nobody understands anything. Searle said that a computer running a program is like that person.
Others reply that the room as a whole, rulebook included, might understand, and the argument is still debated. Most researchers today agree on a narrower point: passing a conversation test shows skill at imitating human talk. It does not settle consciousness, feeling or understanding. Because of this, AI research has largely moved to many specific benchmarks instead of one single test.
A close cousin is the CAPTCHA, whose name stands for Completely Automated Public Turing test to tell Computers and Humans Apart. There the roles are reversed: a machine judges whether the visitor is a person.
Where it shows up in fiction
Science fiction used the idea of testing a mind long before chatbots existed. In Philip K. Dick's novel Do Androids Dream of Electric Sheep? (1968), bounty hunters use the Voight-Kampff test, which measures emotional responses to questions to expose androids. It is not a Turing test, since it reads bodies and not words, but it asks the same question: how can you tell? The film Blade Runner (1982) made the scene famous.
The film Ex Machina (2015) turns the test upside down. The programmer Caleb is brought to a remote house to talk with an android, Ava. He already knows she is a machine. The question is whether she can still persuade him that she has a mind, and who is really being tested. Stories like this treat the Turing test less as a pass-or-fail exam and more as a mirror for human weaknesses, such as loneliness, trust and the wish to be understood.
The test also sits close to two other topics. The uncanny valley describes how a nearly human copy can feel unsettling, while the technological singularity imagines machines that go far beyond human ability. A program that passes for human is only the first step in many stories about both.
Common misconceptions
- "Passing means the machine is conscious." Turing's test looks only at conversation. It says nothing about inner experience, and Turing himself did not claim it did.
- "Eugene Goostman proved computers can think." The 2014 claim was widely criticised, and the program relied on tricks, not understanding.
- "There is one official Turing test." There is no standard version. Time limits, number of judges and the pass mark all vary between experiments.
- "Turing wanted a machine that could fool people." He wanted a practical way to discuss machine intelligence without getting stuck on the word "think".
- "The test is a pure computer-science tool." It began as a philosophy question in a paper, and it works best today as a prompt for debate, in classrooms and in stories.
Read next
- History of Science Fiction: the wider story of how fiction imagined thinking machines.
- Uncanny valley: why almost-human copies can feel eerie.
- Technological singularity: what happens when machines outthink us, in theory and in fiction.
- Philip K. Dick: the author who asked how to tell a human from an android.
