2,587 readers judged short stories. They picked the AI ones as better — and could not tell which were which

Two researchers at Villanova University, Sydney Sears and Deena Skolnick Weisberg, ran three experiments with 2,587 adults and published the results this month in Judgment and Decision Making, the Cambridge University Press journal; the university's publisher announced the findings on 4 August. Participants read roughly thousand-word short stories, some written by people and some by ChatGPT, and were sometimes told the truth about the author and sometimes told the opposite. Readers rated the AI stories higher than the human ones on both quality and how absorbing they were — and rated them higher still when they had been told a person wrote them. Asked to name the author, they failed: 39.93 per cent correct in the second experiment, which is worse than guessing, and 51.97 per cent in the third, which is guessing. One trait predicted better detection, and it was not the obvious one. People who reported more experience with AI systems did better; people who reported more expertise in literature did not. 'AI writing tends to be clearer, more direct and easier to process,' Weisberg said. 'By contrast, human-written stories are often more subtle and complex.'

✓ Verified · 3 sources

Related stories