How the imitation game works
Turing introduced the test in his 1950 paper 'Computing Machinery and Intelligence,' published in the journal Mind. He wanted to replace the vague question 'Can machines think?' with something he could actually run. In his imitation game, a judge exchanges typed messages with two hidden partners, one human and one machine, and tries to tell them apart. If the machine fools the judge as often as a person would, Turing argued, we have no principled reason to deny that it thinks. He predicted that by 2000, machines would fool an average interrogator about 30 percent of the time in a five-minute chat.
The design is clever because it sidesteps arguments about what is going on inside. It does not ask whether the machine has a soul or feelings. It asks only whether its behavior is indistinguishable from a human's, judged by an ordinary person having a normal conversation.
Has any machine passed it?
Headlines have declared victory more than once. In 2014, a chatbot called Eugene Goostman, styled as a 13-year-old Ukrainian boy with shaky English, convinced 33 percent of judges at a London event, and organizers claimed a pass. Most researchers were unimpressed: the persona was a trick designed to excuse odd answers, the sessions were short, and the sample was tiny.
The more serious result came in 2025, when Cameron Jones and Benjamin Bergen at UC San Diego ran a rigorous three-party test. When GPT-4.5 was prompted to adopt a humanlike persona, judges identified it as the human 73 percent of the time, more often than they picked the actual human. By Turing's own standard, modern language models now pass. That is a real milestone in imitation.
Why passing isn't proof of thinking
Passing the test tells you the machine talks like a person. It does not tell you the machine understands, feels, or is aware of anything. A system can produce fluent, convincing language by predicting likely word sequences, with no inner experience behind the words. Many philosophers have pressed this point, and it is why the test is now often read as a measure of human gullibility and machine mimicry as much as of intelligence.
It also tests the judges. Skilled interrogators asking unusual questions do far better than casual ones. So the useful takeaway is not that the debate is over, but that fluent conversation has finally been decoupled from thinking. A thing can sound exactly like a mind without our knowing whether it is one, which pushes the hard questions onto tests the Turing game was never built to settle.