AI-written stories fool readers just fine. It is knowing that ruins it, study finds

In the past 18 months, the publishing industry has faced numerous scandals and increased concerns about the credibility of written works. In response, UK authors have called for labeling of human-written content, and Substack has introduced an AI-detection tool for readers to verify the origin of posts. Despite these efforts, a new study has found that readers often prefer machine-generated stories when they are unaware of their origin.

The study, published in the journal Judgment and Decision Making, involved 2,587 US participants who read three short stories published in literary journals alongside three AI-generated stories created using OpenAI's ChatGPT 4.0. The researchers discovered that participants rated AI-generated stories as superior in quality and more engaging, even though they were told they were human-written.

The study suggests that people may value AI-generated content due to its perceived fluency and ease of comprehension, rather than because it is actually superior.