The Turing test is a proposed way to judge whether a machine can exhibit intelligent behavior, in which a human judge holds a text conversation with an unseen partner and tries to decide whether it is a person or a machine. If the judge cannot reliably tell, the machine is said to have passed. The British mathematician Alan Turing described the idea in 1950 and called it the imitation game. It became the most famous benchmark in the history of artificial intelligence, and also one of the most debated.
History of the Turing test
Turing introduced the idea in his 1950 paper Computing Machinery and Intelligence, published in the journal Mind. He began by asking whether machines can think, then argued that the question was too vague and proposed to replace it with a practical game. In the original version, an interrogator communicates by text with a man and a woman in separate rooms and tries to identify which is which while both try to convince the interrogator. Turing then asked what would happen if a machine took the place of one player. He predicted that by the end of the twentieth century, an average interrogator would have no more than a seventy percent chance of making the right identification after five minutes of questioning, a prediction that proved optimistic but not wildly wrong.
How the Turing test works
The standard modern version is simple. A judge types questions to two hidden participants, one human and one machine, and receives written replies. The conversation can cover any subject, and the machine may use any strategy, including humor, errors and evasion. After a fixed time, the judge decides which participant is the machine. If the judge guesses wrong about as often as right, the machine is considered to have passed. The restriction to text is deliberate. It removes the machine’s appearance and voice from consideration, so that only its linguistic behavior is judged.
Famous attempts and results
Several programs have tried to fool judges. ELIZA, written by Joseph Weizenbaum in 1966, imitated a psychotherapist by turning a user’s statements into questions, and some users formed emotional attachments to it despite its simplicity. PARRY, written in 1972 to imitate a person with paranoid thoughts, fooled some psychiatrists in limited tests. The Loebner Prize, founded in 1990 by Hugh Loebner, held annual contests for years, and winners relied on tricks that mimicked human quirks. In 2014, a program called Eugene Goostman, which posed as a thirteen-year-old Ukrainian boy, was reported to have passed a test in London, but the claim was widely disputed because the persona excused errors and the test was short. Since 2022, large language models have produced far more fluent conversation, and some controlled studies have found that people often mistake their replies for human ones in brief exchanges.
Criticisms of the Turing test
The most influential objection is the Chinese Room argument by philosopher John Searle in 1980. He imagined a person who does not know Chinese following a rulebook to answer Chinese messages so well that outsiders think the person understands. Searle argued that manipulating symbols according to rules is not the same as understanding, so passing the test would not prove a machine has a mind. Other critics say the test rewards deception instead of intelligence, can be passed by tricks, ignores abilities such as perception and physical action and depends on the skill and expectations of the judge. Defenders reply that Turing intended a behavioral criterion and that explaining a convincing, open-ended conversation without intelligence is harder than it sounds.
Variants of the Turing test
Researchers have proposed many variations.
- Total Turing test: adds perception and physical action, so the machine must handle objects and sensory input as well as conversation.
- Reverse Turing test: a machine judges whether a user is human, as in the CAPTCHA tests used to block bots on websites.
- Domain-specific tests: judges examine performance in a single field, such as medical questions or art criticism.
- Lovelace test: asks whether a machine can create something original that its designers cannot explain, named after Ada Lovelace.
- Winograd schemas: ask questions that need common-sense understanding of pronouns, as an alternative that is harder to game with tricks.
Does the Turing test still matter?
Most AI researchers no longer treat the test as a practical goal. Modern work focuses on benchmarks for specific abilities, such as reading comprehension, coding and mathematical reasoning, and on safety evaluations. Yet the Turing test remains valuable as a way to think about what intelligence means, and as a reminder that fluent language is not proof of understanding. As chatbots become hard to distinguish from people in short text exchanges, the test has also taken on a social meaning, raising questions about disclosure, trust and the potential for deception, which regulators address by requiring that people be told when they are interacting with an AI system.
The Turing test and language models
Large language models have renewed attention to the test. They generate fluent, context-aware text, and in short chats some can be mistaken for people. Researchers who run controlled versions note that results depend heavily on the instructions given to the model, the length of the conversation and how experienced the judges are. A model told to adopt a casual persona and make small typing errors tends to do better than one that answers like an assistant. These results show that the test measures how convincingly a system imitates, which is not the same as reasoning ability, knowledge or understanding. Many researchers therefore pair conversation tests with benchmarks that probe specific skills and with checks of factual accuracy.
Turing’s other arguments
In the same 1950 paper, Turing answered a series of objections to the idea of machine thinking. He considered the theological objection, the argument that machines cannot be creative, and Lady Lovelace’s claim that a machine only does what it is told. He suggested that a machine might be built to learn, starting like a child and being educated, instead of being programmed with adult knowledge. This proposal anticipated the modern idea of machine learning. The paper remains widely read because it set out questions that are still open, including whether behavior is a sufficient criterion for intelligence.
FAQs
Has any machine passed the Turing test? Claims have been made, but results depend on the rules, the judges and the time allowed. No result is accepted as a definitive pass by the whole field.
Why did Turing call it the imitation game? Because the machine must imitate a human in conversation, just as, in the original party game, a man tries to imitate a woman.
Does passing the test prove consciousness? No. Even supporters of the test usually treat it as a test of behavior and not as evidence of consciousness.
More in Artificial Intelligence
Key facts
- Alan Turing proposed the imitation game in his 1950 paper Computing Machinery and Intelligence, published in the journal Mind.1
- Turing suggested replacing the question can machines think with a test based on observable conversational behavior.1
- John Searle's 1980 Chinese Room argument claims that passing the test would not prove that a machine understands.2
Related concepts
Quick checkWhere did Alan Turing first describe his imitation game?Show answer
In his 1950 paper Computing Machinery and Intelligence, published in the philosophy journal Mind.
Sources
- Stanford Encyclopedia of Philosophy, The Turing Test, https://plato.stanford.edu/entries/turing-test/
- Stanford Encyclopedia of Philosophy, The Chinese Room Argument, https://plato.stanford.edu/entries/chinese-room/
- Britannica, Turing test, https://www.britannica.com/technology/Turing-test
Editorially reviewed by Specialty Digest Editorial TeamLast reviewed November 17, 2026Researched and drafted with AI assistanceReport an issue
