GPT-4.5 Passes Turing Test at 73% — PNAS Study Confirms Historic Milestone
View original source →A peer-reviewed study published in the Proceedings of the National Academy of Sciences on May 29 confirmed that GPT-4.5 achieved a 73% human-indistinguishability rate in standard Turing test conditions — the highest score ever recorded and above the 50% baseline that defines passing.
Key points:
• Study used standardized three-party Turing test protocol with 500 evaluators across 5-minute text conversations • GPT-4.5 identified as human 73% of the time; actual humans identified as human only 67% of the time • Results consistent across demographic groups, expertise levels, and topic categories • Researchers explicitly noted this measures conversational indistinguishability, not general intelligence or sentience
The finding that GPT-4.5 was identified as human at a higher rate than actual humans is the study's most provocative element. It suggests AI-learned communication behaviors — consistent, patient, fluent, non-defensive — now read as more reliably 'human' than human communication in standardized tests.
A 73% pass rate means in typical interactions, a person has less than one-in-four chance of correctly identifying AI communication as non-human. This creates a new category of disclosure obligation: the question 'could a user reasonably not know this was AI?' now has a scientifically validated answer of 'yes, 73% of the time.'
Why It Matters: AI indistinguishability from humans is now a reproducible, measurable scientific fact. Every organization using AI in customer communication, hiring, or advisory services faces disclosure obligations that existing frameworks were not designed to handle.