London Intercultural
AI history · Human intelligence · Philosophy
The Turing Test is a simple idea with a complicated legacy. Proposed by Alan Turing in 1950, it asks whether a machine can hold a conversation so convincing that a human cannot tell it apart from another person. It remains one of the most cited ideas in artificial intelligence, and one of the most argued over.
What is the Turing Test?
The Turing Test is a way of judging machine intelligence through conversation. A human judge exchanges typed messages with two hidden participants, one human and one machine. If the judge cannot reliably tell which is which, the machine is said to have passed.
Turing did not call it the Turing Test. He called it the imitation game, and he set it out in his 1950 paper “Computing Machinery and Intelligence,” published in the philosophy journal Mind.
Why Turing Changed the Question
The paper opens with a question: “Can machines think?” Turing thought this was too vague to be useful. Nobody could agree on what “think” meant.
So he replaced it with a question that could be tested. Instead of asking what goes on inside a machine, the imitation game asks whether its behaviour can be told apart from a person’s. This shift is why the test is still discussed today. It gives a practical standard rather than a definition of mind.
The cost is that a machine might pass without anything resembling understanding behind its replies. Passing measures performance, not inner life.
Turing’s Prediction
Turing made a specific forecast. He expected that within about fifty years, an average interrogator would have no better than a 70 per cent chance of correctly identifying the machine after five minutes of questioning.
The timing was optimistic. Whether his prediction has come true depends on who is judging and how the test is run.
Early Machines That Fooled People
Two programs from the 1960s and 1970s show how far surface fluency can go without understanding.
- ELIZA (1966): Built by Joseph Weizenbaum at MIT, ELIZA used simple pattern matching to reflect a user’s words back as questions. Its DOCTOR script imitated a Rogerian psychotherapist. Many users confided in it anyway. Weizenbaum reportedly found this disturbing, and the tendency to read understanding into a machine’s replies became known as the ELIZA effect.
- PARRY (1972): Developed by the psychiatrist Kenneth Colby at Stanford, PARRY simulated a person with paranoid schizophrenia. It responded with suspicion governed by a large set of internal rules. Some of its transcripts were later judged by clinicians, who had difficulty telling PARRY apart from a real patient.
Neither program understood anything. Both showed that convincing replies can come from rules alone.
The Eugene Goostman Controversy
In 2014, a chatbot called Eugene Goostman was widely reported to have passed a Turing Test competition organised by the University of Reading. Ten of the thirty judges were convinced it was human, around 33 per cent.
The bot posed as a 13-year-old Ukrainian boy. That persona conveniently explained evasive answers and grammatical slips. Critics argued the result showed the power of a clever disguise, not machine intelligence. The episode is still cited as an example of how the test can be gamed by strategy as well as by understanding.
The Chinese Room Objection
In 1980, the philosopher John Searle proposed a thought experiment that became one of the most influential challenges to the Turing Test.
Imagine a person who does not read Chinese sitting in a room with a detailed rulebook for manipulating Chinese characters. Written questions come in, and by following the rules, the person sends back appropriate answers. To someone outside the room appears to understand Chinese. The person inside understands nothing.
Searle’s point is that the test checks for the appearance of understanding. Turing anticipated objections of this kind. His 1950 paper addresses nine of them, including the “Lady Lovelace’s objection”, which draws on Ada Lovelace’s writings about the limits of machines.
What the Test Reveals About Us
Perhaps the most revealing thing about the Turing Test is what it says about human judges.
We usually judge other minds through conversation, because it is the only window we have into someone else’s thinking. The test extends that habit to machines. If we rely on fluency to judge intelligence, we may credit a fluent machine too quickly, and overlook the difference between fluency and thought.
Fluency, Assumptions, and Intercultural Judgement
The ELIZA effect is not only a question about machines. Across cultures, a polished and confident exchange can create an impression of shared understanding that has not actually been checked. A colleague who answers smoothly, speaks politely, or stays quiet may be read as competent, rude, or evasive, depending on norms the other person may not share.
The Chinese Room offers a useful distinction here. Someone can follow the rules of etiquette or language fluently and still not understand what those rules are for. Intercultural competence depends on understanding the reasons behind the surface behaviour not just reproducing it.
We are constantly judging others through conversation. Directness, humour, and silence all carry different meanings across cultures. Before concluding that someone lacks confidence, initiative, or clarity, it is worth asking who we are really evaluating: the person, or our assumptions about how a capable person should sound.
The Eugene Goostman episode is a reminder of how easily a persona can explain away a gap. In a business setting, organisations do something similar when a cultural mismatch is labelled as a personality trait or a performance issue. Inclusive leadership begins with noticing when a convenient explanation is doing the work that a real enquiry should do.
Why the Turing Test Still Matters
For years, the Turing Test was a thought experiment. Today it is a practical question with real consequences, and the answers are arriving faster than most people expected.
The test has now been passed in a rigorous study
In 2025, researchers at UC San Diego, Cameron Jones and Benjamin Bergen, ran a three-party version of the test with human interrogators. The interrogators chatted with a human and an AI at the same time, then had to pick out the human. GPT-4.5, given a persona, was judged to be the human 73 per cent of the time, significantly more often than the real humans in the study. Meta’s LLaMa-3.1-405B was judged human 56 per cent of the time, which was statistically indistinguishable from the humans. The study was published in the Proceedings of the National Academy of Sciences in 2026.
The most important detail is what made the difference. The systems did not win by being smarter or knowing more. They won by sounding more human, through tone, directness, humour, and small fallibilities. Turing would have recognised the point. His test was always about behaviour, and behaviour is exactly what the models learned to imitate.
Deepfakes and the impersonation economy
In 2024, a finance employee at the engineering firm Arup, based in Hong Kong, was reported to have transferred around US$25 million after a video call with what appeared to be the company’s chief financial officer and several colleagues. Everyone on the call was an AI-generated deepfake. The case is now a standard reference point in corporate security training.
Fraud data points the same way. A Pindrop study of 1.2 billion customer calls reported a 1,300 per cent year-on-year rise in deepfake-led fraud cases. The Turing Test asked whether a machine could pass as a human in conversation. Today’s fraudsters need only a few seconds of audio to pass as a relative, a boss, or a bank.
Political voices and robocalls
In January 2024, voters in New Hampshire received robocalls with an AI-generated voice imitating President Joe Biden, urging them not to vote in the primary. The US Federal Communications Commission later fined the consultant responsible $6 million. The case showed that a synthetic voice does not need to fool a whole electorate, only enough listeners to cause damage.
Regulation now asks the Turing question in law
The EU AI Act’s transparency duties took effect on 2 August 2026. Under Article 50, people must be told when they are interacting with an AI system, and synthetic content such as deepfakes must be labelled. The European Commission published a voluntary code of practice on labelling AI-generated content in June 2026.
The United States is moving in a more fragmented way. The FTC’s Impersonation Rule, effective since 2024, makes it unlawful to impersonate government bodies and businesses, including with AI-generated content. New York now requires disclosure when advertisements feature “synthetic performers” created by AI, from 9 June 2026.
The shift is notable. The Turing Test asked whether a machine could hide its nature well enough to pass. The new rules increasingly ask machines to declare it.
What people already believe
A representative survey of 1,040 people in Germany, conducted in September 2024, found that two thirds of those aged 16 and over who had heard of AI chatbots said it had become hard to tell a human from an AI in conversation. Among regular users, one in five said they had sometimes forgotten they were talking to a machine. The public has, in effect, already made Turing’s judgement, and found it unsettling.
Looking beyond 2026
Several directions seem likely:
- Proof of personhood: systems that verify a real human is present, rather than trying to detect a machine after the fact.
- Content provenance: standards such as C2PA, which attach verifiable origin information to images, audio, and video.
- Honesty over imitation: the more useful question may shift from “Is this human?” to “Is this system honest about what it is?”
The Turing Test asked whether a machine could pass as human. In 2026, the more pressing question is how humans, institutions, and laws will respond when it can.
