Researchers develop a new method to catch AI chatbots impersonating humans by exploiting their perfect memory, which contradicts natural human cognitive limitations.

Online behavioral research faces validity threats as large language models (LLMs) can now solve traditional machine-detection challenges that were once human-exclusive

Scientists propose a reverse approach: use tasks that machines perform too perfectly to be human, specifically testing for limited working memory capacity

A serial recall cognitive task successfully distinguished human participants from LLMs, even when the AI was explicitly instructed to mimic human memory constraints

This method leverages well-established cognitive science phenomena as a reliable tool for humanness detection in online studies