Definition
The Turing Test, proposed by Alan Turing in 1950 as the imitation game, asks whether a machine can converse so well that a human judge can't reliably tell it from a person. The Chinese Room, proposed by philosopher John Searle in 1980, argues that passing such a test wouldn't prove real understanding.
In simple words
You chat by text with two others, one a person and one a machine. If you can't tell which is which, the machine passes. Searle's reply: imagine someone in a room following a rule book to answer Chinese questions, without knowing any Chinese. The answers look perfect, yet no understanding is happening in there.
Why it matters
Modern chatbots have made this debate practical. They can often fool people in short chats, so does that make them intelligent? The discussion teaches you to separate seeming intelligent from being intelligent, and that matters when deciding how far to trust an AI.
Key terms
- Alan Turing
- British mathematician (1912 to 1954) who laid the foundations of computer science and asked whether machines can think.
- Imitation game
- Turing's original name for his test, in which a machine tries to be mistaken for a person in a written conversation.
- Interrogator
- The human judge who asks questions and must decide which answers come from the machine.
- Chinese Room
- John Searle's 1980 argument that following rules to produce correct answers is not the same as understanding.
- CAPTCHA
- A website puzzle that checks you are human. It is a reversed Turing test, with a computer as the judge.
Explained in detail
The man and the question
Alan Turing was a British mathematician. In 1936 he described an imaginary machine that could carry out any calculation by following simple instructions, an idea that became the basis of every modern computer. During the Second World War he worked at Bletchley Park, where he helped break the German Enigma cipher. In 1950 he published an article called "Computing Machinery and Intelligence" in the philosophy journal Mind. Its first sentence was: "I propose to consider the question, 'Can machines think?'"
Replacing a vague question with a test
Turing saw that arguing about the meaning of "think" would lead nowhere. So he replaced the question with a game that could actually be played. A human judge exchanges typed messages with two hidden players, one a person and one a machine. The judge may ask anything. If, after the conversation, the judge cannot reliably say which is the machine, the machine has done well in the imitation game. Typing was chosen on purpose, so that voice and appearance would not matter, only the quality of the conversation.
Turing also made a prediction: that by about the year 2000, computers would play the game well enough that an average judge would have no more than a 70 per cent chance of identifying the machine after five minutes of questions.
Attempts to pass it
For decades no program came close. From 1991 an annual contest, the Loebner Prize, tested chatbots against judges. In 2014 a chatbot called Eugene Goostman, which pretended to be a thirteen-year-old boy from Ukraine, convinced a third of the judges at one event that it was human. Many researchers objected that pretending to be a child who spoke English as a second language was a trick to excuse its mistakes.
Large language models changed the picture. In a controlled study published in 2025, researchers at the University of California San Diego ran five-minute, three-way tests of the kind Turing described. One recent model, when told to adopt a particular human personality, was judged to be the human more often than the real people it was compared with. By the standard Turing set, machines can now pass.
Why passing does not settle the matter
The test measures whether a machine can imitate human conversation, not whether it understands, reasons correctly or is truthful. A system could pass by being good at deception. It could also be extremely useful, as a medical scanner is, without being able to chat at all. The philosopher John Searle's Chinese Room argument makes the point that producing the right symbols by following rules is different from understanding them. For these reasons, researchers today measure AI with many separate tests of reasoning, knowledge, safety and the ability to learn new skills, and the Turing Test is remembered mainly as the idea that started the discussion.
How it works
How the test works
- A judge chats by text with a hidden person and a hidden machine.
- After a set time, the judge says which is the machine.
- If judges are wrong about as often as right, the machine passes.
Criticisms
- It tests imitation, including imitating human mistakes.
- It depends on the judge and the length of the chat.
- It ignores vision, action and everything except conversation.
- A system can fool people using tricks, without much intelligence.
Try it yourself
Play out the Turing Test
A judge chats by typing, and can't see who is answering. Follow one question.
Human or machine reply?
A judge asked: "What did you have for breakfast?" Guess who wrote each reply, then read why clues can mislead.
"Breakfast is the first meal of the day, eaten in the morning."
"Poha, with far too much lemon. My brother squeezed it."
"As an AI, I don't eat, but I can suggest a healthy breakfast."
"Skipped it, overslept. Don't tell my mum."
Common mistakes
- Thinking that Turing claimed passing proves consciousness. He proposed a practical test in place of the vague question "can machines think?"
- Treating fooling someone once as a pass. Serious versions need many judges and longer chats.
Check yourself
Summary
- Turing Test: can a judge tell the machine from a person in conversation?
- Chinese Room: behaving correctly may not mean understanding.
- Passing the test is about behaviour, and doesn't settle the bigger question.
Finished reading?
Sources: Turing, Computing Machinery and Intelligence (1950) · ARC Prize: ARC-AGI