AI Search 101
What is the Turing test?
The Turing test, proposed by Alan Turing in 1950, judges a machine by conversation. If a person exchanging written messages with a human and a machine cannot reliably tell which is which, the machine passes. It was never a measure of intelligence so much as a way of avoiding the question.
The basics
| Proposed by | Alan Turing, 1950 |
| Originally called | The imitation game |
| Published in | The paper Computing Machinery and Intelligence |
| Tests | Indistinguishability in conversation, not understanding |
| Relevant to | Strong and narrow AI |
What did Turing actually propose?
He began by observing that the question "can machines think" is too vague to answer, because thinking has no agreed definition. Rather than attempt one, he replaced the question with something testable.
An interrogator exchanges written messages with two hidden participants, one human and one machine, and tries to determine which is which. If the machine can sustain the deception as often as a human participant could, Turing argued, the original question has been answered as well as it can be.
The move is philosophical rather than technical. Turing was arguing that if behaviour is indistinguishable, insisting that one party is not really thinking becomes an unfalsifiable claim about inner states we cannot inspect in anyone, including other people.
Has the test been passed?
Depending on the conditions, arguably yes, and this turned out to matter less than expected.
Modern conversational systems routinely convince people in ordinary exchanges, and studies putting them in Turing-style setups have found participants unable to identify the machine at rates well above chance. Under casual conditions the test is not a meaningful obstacle.
What that revealed is less about machines than about the test. Producing convincingly human text turned out to be a narrower achievement than Turing's framing implied. A system can be indistinguishable in conversation while having no grasp of the world, no persistence, and no ability to do anything outside language. Passing has stopped being treated as a milestone.
Why did the test stop mattering?
Three reasons.
It rewards imitation. To pass, a system should hesitate, make errors and avoid answering arithmetic too quickly. Optimising for that is optimising for deception, not competence.
It measures one narrow ability. Conversation is one thing intelligent beings do. The test says nothing about reasoning under uncertainty, acting in the world, or learning from experience.
The judge is the weak point. Results depend heavily on how much time the interrogator has, how sceptical they are and what they know. That makes it an unstable measure.
Contemporary evaluation instead uses batteries of specific benchmarks measuring reasoning, factual accuracy, coding and safety. Less elegant, considerably more informative.
What is the Chinese Room?
The best-known objection, put by John Searle in 1980. Imagine someone inside a room following rules to manipulate Chinese symbols, producing responses fluent enough to convince a Chinese speaker outside, while understanding no Chinese at all. Searle's point is that symbol manipulation, however sophisticated, is not comprehension.
The counterargument is that understanding may be a property of the whole system rather than of the person inside it, and that the same reductive move applied to neurons would deny understanding to human brains. The dispute has run for forty years without resolution, and current systems have made it considerably more concrete without settling it.
How the Turing test is used in marketing
Production. The practical question is no longer whether generated text can pass as human but whether it should be presented as such. Disclosure norms are still forming and reputational risk sits with whoever gets it wrong.
Analysis. Detection tools claiming to identify generated text are unreliable in both directions, which is the Turing test's lesson arriving as an operational problem.
Distribution. A buyer reading a description of a company from an assistant cannot tell which parts are grounded in sources and which are fluent guesswork. Indistinguishability cuts both ways.
How this impacts your business
A language model does not look a company up in a register. It composes a description from everything it read during training and whatever it retrieves at the moment of the question. Trade press, directories, forums, competitor comparison pages, and somewhere in that mix the company's own website too.
So the description a buyer sees is assembled, not published. No company controls it, but every company supplies a share of the material it is built from.
The traffic impact is still small, roughly one percent of sessions for most companies. The influence is not small, because the reader is often a buyer forming a first impression of what a company does and who it serves, before any page gets a click.
Checking takes ten seconds. Type the company name into ChatGPT and read the answer.
Frequently asked questions
Who was Alan Turing?
A British mathematician and logician, foundational to computer science, who worked on wartime codebreaking at Bletchley Park and formulated much of the theory underlying modern computing.
Does passing the Turing test mean a machine is conscious?
No. The test deliberately avoids consciousness and inner experience, measuring only behaviour in conversation.
Is the Turing test still used?
Not as a serious evaluation method. It remains important historically and as a way of framing what intelligence tests can and cannot establish.
What replaced it?
Task-specific benchmarks measuring reasoning, factual accuracy, coding, instruction following and safety, reported as scores rather than as a single pass or fail.
What is a reverse Turing test?
A test designed to distinguish humans from machines, which is what a CAPTCHA is. The name is a joke that stuck.
Related definitions
- Strong and narrow AI
- What is artificial intelligence?
- What is the technological singularity?
- What is a large language model?
- AI Search 101
Written by Lari Numminen, Generate More. Also available in Finnish. Updated 31 August 2026.