Turing-teszt

in English: Turing test

basicwidespread

A test where a judge converses in writing with a human and a machine. If the judge cannot reliably distinguish them, the machine passes the test.

In plain terms

Imagine slipping notes under two closed doors, and a response comes from behind each door. A person sits behind one, a machine behind the other. If after a few minutes of correspondence you cannot tell which is the machine, then the machine has passed the Turing test.

The test does not measure whether the machine "really thinks", because this cannot be decided from the outside. It measures whether its behavior is distinguishable from a human's. Alan Turing invented it precisely to replace the intangible question "can a machine think?" with a testable game.

For decades it was considered a distant vision, but today it is more of a cultural milestone. Current conversational AI systems often deceive people in short conversations, so researchers use other methods to measure what a model or agent can do.

How it works

There are three participants: an interrogator (judge), a human, and a machine. The judge cannot see or hear them and communicates only in writing.
The judge can ask anything: ask for a joke, set arithmetic problems, or probe about emotions or yesterday's weather.
The machine's goal is to appear human, so it may intentionally make mistakes, delay, or use slang. The human's goal is to be recognized as human by the judge.
At the end of the specified time (typically a few minutes), the judge names which conversation partner they believe is the machine.
Many judges repeat the trial. If judges err as often as if they were guessing, the machine proved indistinguishable.

An everyday example

You are writing in a customer service chat window of an online store, and after the third response, you become unsure whether a human or an AI agent is replying. In this case, you are in a situation similar to a Turing test judge, except there is no definitely human conversation partner next to you to compare with. If you cannot decide, the system has passed an improvised, unofficial test on you.

Who is affected

everyoneIn more and more chats, emails, and phone calls, you must decide whether you are dealing with a human or a machine.

parent, student, teacherThe concept is a good starting point to discuss what intelligence means and why human-sounding text is not the same as understanding. → Education & Family

lawyer, citizenSince machines are now deceptively human-like, the EU AI Act requires you to be informed if you are interacting with an AI system, unless the situation makes this obvious. → Law

bank customer, investorScammers use AI conversations that feel human to build trust, so the feeling of "it was definitely a person" is no longer proof. → Money

A common misconception

MythIf a machine passes the Turing test, it has consciousness and thinks like a human.

In realityThe test only shows that a machine can deceive a human in conversation. It does not by itself prove understanding, consciousness, or general intelligence. What a successful test exactly means is still debated by philosophers. One of the most famous counterarguments is the 'Chinese room' thought experiment.

Limits and risks

It rewards deception, not knowledge: a machine can pass with clever nonsense and intentional errors.
The result depends heavily on the judge, the length of the conversation, and the topic, so there is no single official version accepted by everyone.
It examines only the conversation, so it says nothing about whether an agent reliably performs real-world tasks.
If a company advertises its product by saying it 'passed the Turing test,' ask under what conditions and for how long the trial lasted.

Similar, but not the same

CAPTCHA – This is the reverse situation: there, a machine automatically decides whether you are human, not a human judging the machine.
General AI – Artificial general intelligence would mean the machine performs almost every intellectual task at a human level, while the Turing test only measures deception in conversation.
Benchmark – A benchmark measures performance on a fixed set of tasks with a score, while the Turing test relies on the impressions of human judges.

Related concepts

Broader

Artificial intelligence, evaluation of AI systems

Narrower

Loebner Prize

On the edge

CAPTCHA, General AI, Large language model, Chatbot, Chinese Room, Benchmark

A brief history

British mathematician Alan Turing described the idea in his 1950 paper "Computing Machinery and Intelligence", published in the journal Mind, under the name "imitation game". He predicted that by around 2000 machines would play this game so well that an average interrogator would have no more than a 70 percent chance of correctly identifying the machine after five minutes of conversation. The name "Turing test" came later. The 1966 ELIZA program made many users feel that it understood them, using only simple tricks, although it was not evaluated under the rules of the test. The 2014 claim of success for "Eugene Goostman" was disputed by experts.

Where things stand

The concept is well known. In 2025, researchers at the University of California, San Diego, examined large language models in a controlled, three-party experiment involving five-minute written conversations. A model given appropriate instructions (GPT-4.5) was judged by human evaluators to be human in approximately 73 percent of cases, more often than the actual human participant. Therefore, at the end of 2026, this test is considered more of a historical milestone, with capabilities now measured by other methods.

Status: 2026 November

Also known as

Turing-teszt · Turing teszt · Turing test · Turing test · imitation game · imitation game · imitation game

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.