Glossary
What is the Turing test?
In short
The Turing test is a check in which a judge chats with a machine and with a person and tries to guess which one is the computer.
AI basics · Turing test
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine a game behind a screen. You write notes to two partners and cannot see them. One is your friend, the other is a computer. From the answers, you try to guess who is who. If the computer answers so much like a person that you often get it wrong, it played the game well. A scientist invented this game long ago to find out whether machines can think.
In plain words
The question “Can a machine think?” is hard to answer. So a simple check was proposed instead. A judge writes with two partners at once and can neither see nor hear them. One is a person and the other is a program. After the talk, the judge decides who is who. If the program is taken for a human at least as often as the real person is, the test is passed. Only the behavior in the conversation is tested: natural answers, jokes, human-like mistakes. This cannot show whether the program understands what it says.
For business
For a company, the test matters from the other side. A voice agent or chatbot can already sound like a live employee, and a customer may find it hard to tell who they are talking to, in text or on the phone. That is convenient, but it puts trust at risk: a customer who finds out the truth alone is more upset than one told directly. The honest practice is simple. The agent says at once that it is AI and hands the customer to a person on request. Sounding human is also not the same as quality: an agent can speak smoothly and still get a price or a date wrong. So before launch, teams measure the share of correct answers against the company’s scripts, and the rules for introducing the agent go into the voice agent’s script.
For a specialist
A. Turing described the imitation game in the paper “Computing Machinery and Intelligence” (Mind, 1950). In the original setup, an interrogator tells a man from a woman by written messages, and then a machine takes the place of one participant. In the three-party test, the judge talks to a human and a program at once and picks the human. Critics say the test measures imitation, not understanding: J. Searle’s “Chinese room” thought experiment (1980) shows that correct answers can follow rules without any grasp of meaning. C. Jones and B. Bergen (2025) ran two randomized, preregistered three-party tests with five-minute conversations. GPT-4.5 with a persona prompt was picked as the human in 73% of cases, more often than real people, while ELIZA reached 23% and GPT-4o 21%. Today model abilities are more often compared on task sets (benchmarks), and this test serves as a measure of persuasiveness and the risk of deception.
Example
An online store tests a chatbot on ten employees. Each chats with the bot or with an operator and guesses who answered. The bot is mistaken for a human in almost half of the chats, so the company adds the line “I am a virtual assistant” to its first message.
Related terms
Questions and answers
Has AI passed the Turing test?
Sources
- A. M. Turing, Computing Machinery and Intelligence, Mind 59(236), 1950academic.oup.com
- Andrew Hodges, The Alan Turing Internet Scrapbook: The Turing Test, 1950turing.org.uk
- C. R. Jones, B. K. Bergen, Large Language Models Pass the Turing Test, 31.03.2025arxiv.org
- J. R. Searle, Minds, Brains, and Programs, Behavioral and Brain Sciences 3(3), 1980doi.org
Where companies use it
Not sure whether your company needs this?
AI solutions map