AI passes the Turing test: what does this really mean?

When I came across the preprint titled Large Language Models Pass the Turing Test, I knew right away it would be a fascinating read. Although it’s aimed at an academic audience and dense with scientific jargon, I want to share the parts that really excite me.

Seventy-five years ago, Alan Turing asked a question that would shape the future of technology: Can machines think? His famous “imitation game” - now called the Turing test, was simple in concept but had huge implications. In this test, a human judge chats with both a real person and a machine, trying to work out which is which. If the judge cannot reliably tell the difference, the machine is said to have passed the test.

For decades, this was a theoretical milestone. But a new research paper (in preprint) by Cameron R. Jones and Benjamin K. Bergen at UC San Diego has just shown, for the first time, that a modern AI can pass the classic Turing test - and even outperform humans at sounding human.

The experiment: humans versus AI in a digital conversation

The researchers set up a series of text-based conversations, letting four AI systems compete against real people:

• ELIZA: A simple chatbot from the 1960s, famous for its “therapist” persona.

• GPT-4o: A recent, but less advanced, large language model (LLM).

• LLaMa-3.1-405B: A state-of-the-art LLM.

• GPT-4.5: The latest and most advanced model from OpenAI.

Participants - both university students and online workers - chatted with two “witnesses” at once: one human, one AI. After five minutes, they had to guess which was the real person. Sometimes, the AI was told to act like a specific type of person (a young, introverted, internet-savvy individual), and sometimes it was simply told to sound human.

The results: when AI outsmarts the human

Here’s where things get truly fascinating:

• GPT-4.5, when given a specific persona, was judged to be the human 73% of the time. In other words, people picked the AI as the real person more often than the actual human! This is a huge leap - no AI has ever done this well in the Turing test.

• LLaMa-3.1, with the same persona prompt, was picked as human 56% of the time. This is about the same as random guessing, but still impressive.

• Without a persona, both advanced AIs did worse (36–38% win rates), showing that how you prompt the AI really matters.

• The older models (ELIZA and GPT-4o) were easily spotted as machines, winning only about 21–23% of the time.

In short: with the right instructions, today’s best AI can not only fool people, but actually outdo real humans at sounding human in a text chat.

Why does this matter?

The Turing test is no longer just a thought experiment

For decades, the Turing test was a kind of philosophical benchmark - something to debate, not something AI could actually pass. This research changes that. It shows that, at least in conversation, AI can now stand in for a real person so convincingly that most people cannot tell the difference.

Prompting is powerful

The way you tell an AI to behave makes a huge difference. When GPT-4.5 was told to act like a specific kind of person, it became much harder to spot. This suggests that AI’s ability to mimic humans is not just about raw intelligence - it is about context, personality, and social cues.

Real-world impacts are coming

If AI can reliably pass as human, it could automate jobs that rely on conversation - customer service, sales, maybe even therapy. But there is a flip side: AIs that are this good at pretending to be people could also be used for deception, social engineering, or spreading misinformation. The line between human and machine is getting blurrier, and that raises big questions for society.

The takeaway

The Turing test has finally been passed - not by a simple chatbot, but by a sophisticated AI with the right persona. This is a milestone for artificial intelligence, and a sign that the future of human-machine interaction is arriving faster than many expected. As AI continues to evolve, the challenge will be not just building smarter machines, but figuring out how to use them wisely.

If you are interested in AI, now is the time to pay attention. The machines are not just coming - they are already here, and they sound a lot like us.