Artificial Intelligence & Machine Learning: The Turing Test
The Test That Was Never Meant to Be Passed
Alan Turing didn't design his famous test to prove machines could think — he designed it to make us stop asking the question.
The Idea
In 1950, Turing published a paper called 'Computing Machinery and Intelligence,' and he opened it with a provocation: rather than asking 'Can machines think?' — a question he found philosophically intractable — he proposed a practical substitute. Could a machine converse in text with a human judge, for five minutes, and fool that judge into thinking it was human at least 30% of the time? If yes, Turing suggested, we might as well move on. What's remarkable, and consistently overlooked, is that this was a rhetorical sidestep, not a definition of intelligence. Turing wasn't claiming that passing the test would mean a machine was conscious or truly thinking. He was saying that if we can't tell the difference in practice, the philosophical debate becomes less urgent. The test was a way of dissolving the question, not answering it. This distinction matters enormously now. When GPT-4 or Claude holds a conversation that feels uncannily human, people instinctively reach for the Turing Test as a measuring stick — as if passing it would settle something fundamental. But Turing himself would likely shrug. His test was always about behavioural indistinguishability, not inner experience. The hard problem of consciousness — whether there is 'something it is like' to be a machine processing language — remains entirely untouched by any chatbot's conversational smoothness. The test has also proven surprisingly easy to game and surprisingly hard to administer fairly. Humans, it turns out, are easy to fool — and also weirdly reluctant to be fooled, which distorts the results in the opposite direction.
In the World
In June 2014, a chatbot named Eugene Goostman — designed to impersonate a 13-year-old Ukrainian boy — was declared by its creators to have 'passed the Turing Test' at an event hosted at the Royal Society in London. Thirty-three percent of judges, over a five-minute text exchange, thought they were talking to a human. Headlines exploded. The moment was widely reported as historic. The AI research community was mostly unimpressed, and for good reason. The Eugene Goostman team had made a clever, almost cynical choice: by giving their chatbot a persona that was a non-native English speaker and a teenager, they built in a ready excuse for every grammatical oddity, every non-sequitur, every evasion. The bar wasn't 'sound like a thoughtful adult.' It was 'don't definitively sound like a machine.' That's a much easier target. More tellingly, the judges in these events are often untrained, given only a short window, and predisposed to be charitable. When researchers ran more rigorous versions — longer conversations, expert judges, no convenient persona shield — machines fared far worse. Fast-forward to today, and the conversation has inverted in a strange way. Modern large language models are so fluent that many people interacting with them assume they must be 'more' than they are — projecting understanding, intent, even emotion onto systems that are, at their core, extraordinarily sophisticated pattern-completion engines. We've gone from worrying machines will trick us into thinking they're human, to humans eagerly doing the tricking themselves.
Why It Matters
There's a practical trap hiding inside the Turing Test's legacy: we've inherited a framework that equates fluency with intelligence, and that confusion is quietly shaping enormous decisions right now — in hiring tools, in mental health apps, in legal research assistants, in customer service systems. When a system sounds confident and coherent, we extend it trust that should really be earned through accuracy, consistency, and genuine understanding. Language models can be wrong in ways that are perfectly grammatical and utterly convincing. The Turing Test, as a cultural benchmark, trains us to trust the surface — and the surface has never been smoother. Knowing that Turing intended the test as a philosophical sidestep rather than a definition of machine intelligence gives you a useful sceptical tool. The next time an AI interaction feels eerily human — and it will — you can appreciate the technical achievement while holding the deeper question open: what is actually happening inside this system, and how much should that change what I trust it with? Fluency is not understanding. Coherence is not knowledge. These feel obvious when stated plainly, but they dissolve surprisingly fast in a good conversation.
A Question to Ponder
If a machine consistently gives you better advice than most humans you know, does it matter whether it understands anything at all — and who gets to decide?
Get a new one of these every morning.
Start learning with Thinkable