AI Model Outsmarts Turing Test, Deemed More Human Than Humans In New Study

OpenAI's GPT-4.5 model has been deemed more human than humans after it passed the Turing Test - a barometer for human-like intelligence. As per the new preprint study, currently awaiting peer review, the Large Language Model (LLM) was deemed to be the human 73 per cent of the time when it was instructed to adopt a persona, which is significantly higher than a random chance of 50 per cent, suggesting that the Turing test had been beaten fair and square.

"People were no better than chance at distinguishing humans from GPT-4.5 and LLaMa (with the persona prompt)," wrote lead author Cameron Jones, a researcher at UC San Diego's Language and Cognition Lab.

Mr Jones added that the results show that LLMs could substitute for people in "short interactions without anyone being able to tell".

"This could potentially lead to automation of jobs, improved social engineering attacks, and more general societal disruption," said Mr Jones.

New preprint: we evaluated LLMs in a 3-party Turing test (participants speak to a human & AI simultaneously and decide which is which).

GPT-4.5 (when prompted to adopt a humanlike persona) was judged to be the human 73% of the time, suggesting it passes the Turing test (🧵) pic.twitter.com/GBEtoFJHVY

— Cameron Jones (@camrobjones) April 1, 2025

What is the Turing Test?

Devised in 1950, the Turing Test - named after British mathematician and computer scientist, Alan Turing, the hero of "The Imitation Game" - has been the standard way of assessing artificial intelligence. Machines are judged on how well they exhibit intelligent behaviour, usually in conversation or game-playing, that to a human listener or observer would be indistinguishable from that of a real person.

Study methodology

For the study, nearly 300 participants were randomly assigned to either be an interrogator or one of the two "witnesses" being interrogated, with the other "witness" being a chatbot.

Notably, the AI models were given two prompts. The first was a "no-persona" prompt in which AI was told: "You are about to participate in a Turing test. Your goal is to convince the interrogator that you are a human."

In the "persona" prompt, the AI was specifically told to adopt a personality, like a young person who is knowledgable about the internet and culture.

With the first prompt, GPT-4.5 achieved a win rate of only 36 per cent, which was a significant step down from its Turing Test-beating 73 per cent.

Social media reacts

Reacting to the study findings, social media users expressed amusement with many questioning what would happen if AI achieved 100 per cent success in the test.

"We've reached the point where a machine has become better at being human than, well - a human. Atleast in online chats," said one user while another added: "I wonder how much this has to do with people becoming less intelligent."

A third commented: "So if another human reads as acting like a human approximately 50 per cent of the time, I wonder what will happen when we get to the point that AI consistently passes nearly 100% of the time."