The abstract is read at the publisher; the summary is JClub's.
Large language models are used across disciplines as human proxies, and their human-likeness is not one property because each role demands distinct validity criteria.
- Why it matters: The field needs a way to evaluate when LLM-based “human proxy” setups are scientifically trustworthy, despite lacking mechanistic explanations of human behavior.
- What they did: The authors review research that uses LLMs explicitly or implicitly as human stand-ins, grouping studies into four roles: believable agents, task agents, experimental subjects, and silicon samples.
- The result: Comparing these roles shows that similarity to humans varies by purpose, so each role supports different scientific claims and must be validated with role-specific criteria.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.