"Can machines think?" Alan Turing opened his 1950 paper with that question and immediately threw it away, declaring it "too meaningless to deserve discussion." Words like "machine" and "think," he argued, were so fuzzy that debating them was hopeless. So in "Computing Machinery and Intelligence," published in the philosophy journal Mind, he proposed a replacement: a game with a testable outcome. That substitution — swap a metaphysical argument for an experiment — is why a landmark paper from 1950 still frames arguments about artificial intelligence today.
The original version is stranger than most people remember. Turing started from a Victorian-style parlour amusement he called the imitation game: an interrogator exchanges written messages with two hidden people, a man and a woman, and must work out which is which while the man tries to deceive. Turing's twist was to replace one player with a digital computer. If an interrogator, communicating only by typed text, cannot reliably tell the machine from the human, then — Turing suggested — quibbling about whether the machine "really" thinks becomes empty. The typed-text channel was the crucial design choice: it strips away voice, face and body, forcing the judgement onto conversation alone.
Turing even ventured a forecast. By about the year 2000, he predicted, computers with roughly a billion units of storage would play the game well enough that an average interrogator would have no better than a 70 per cent chance of making the right identification after five minutes of questioning. He also, with characteristic dry wit, catalogued and rebutted nine objections to machine intelligence — theological arguments, the "heads in the sand" objection (thinking machines would be too dreadful, so let's hope they're impossible), arguments from consciousness, and "Lady Lovelace's objection," Ada Lovelace's century-old claim that a machine can only do what we order it to do and can never originate anything.
Chatbots, prizes and a fake teenager
The test escaped the philosophy journals quickly. In 1966 MIT's Joseph Weizenbaum built ELIZA, a simple program whose most famous script, DOCTOR, parodied a non-directive psychotherapist by reflecting users' statements back as questions. Weizenbaum was unsettled to find people confiding in it earnestly — his own secretary reportedly asked him to leave the room so she could talk to it privately — and he spent much of his later career warning about the ease with which humans project understanding onto machines. That tendency is now called the ELIZA effect. A few years later, psychiatrist Kenneth Colby's program PARRY simulated a person with paranoia; in an early informal milestone, psychiatrists shown transcripts often could not reliably distinguish PARRY from human patients.
From 1991 the Loebner Prize staged an annual, restricted version of the test, which mostly demonstrated how far chatbots could get on evasion, jokes and topic-changing. In 2014 a chatbot called Eugene Goostman, which adopted the persona of a 13-year-old Ukrainian boy — a character whose broken English and adolescent flippancy conveniently excused every failure — was declared by its promoters to have "passed" the test by fooling a third of judges in five-minute chats. AI researchers widely rejected the claim as a publicity stunt, but it illustrated something real: the test can reward deception and lowered expectations rather than intelligence.
Philosophers attacked from the other flank. In 1980 John Searle published the Chinese Room argument: imagine a person who speaks no Chinese locked in a room, following a rulebook to manipulate Chinese symbols so well that outsiders believe they are conversing with a Chinese speaker. The room passes the test, Searle argued, yet nobody and nothing in it understands Chinese — so passing proves simulation, not understanding. The debate over whether that argument works has filled libraries.
The test inverted
Meanwhile the test quietly turned inside out. The CAPTCHA — those distorted-letter and street-sign puzzles guarding websites — is a reverse Turing test: a machine judging whether you are human. Its name is an acronym for Completely Automated Public Turing test to tell Computers and Humans Apart. And as conversational AI systems have grown fluent enough to sail through casual, short-form conversation, the goalposts have shifted exactly as sceptics predicted they would: fooling a human in five minutes of chat now feels less like proof of thinking and more like proof of good mimicry. Perhaps that is the test's real legacy. Turing died in 1954, only four years after the paper; he never saw a machine come close. But he successfully moved the question from "what is thought?" to "what would convince you?" — and seventy-plus years on, we still argue about the answer in his terms.
Quiz nuggets
- Turing proposed the test in his 1950 paper "Computing Machinery and Intelligence," published in the journal Mind.
- He called it the "imitation game," adapted from a party game involving a man and a woman.
- ELIZA (1966), Joseph Weizenbaum's mock psychotherapist, gave its name to the ELIZA effect.
- John Searle's 1980 Chinese Room argument claims passing the test shows simulation, not understanding.
- CAPTCHA stands for Completely Automated Public Turing test to tell Computers and Humans Apart.