Large language models as human proxies.

Large language models can serve as human proxies in four distinct roles, each supporting different scientific claims and requiring unique validity criteria.

• Why it matters: Understanding how LLMs can approximate human behavior is crucial for advancing both scientific research and practical applications, but the criteria for their validity vary by role.
• What they did: The authors analyze interdisciplinary research that uses LLMs as human proxies, categorizing their functions into believable agents, task agents, experimental subjects, and silicon samples.
• The result: Findings suggest that similarity to humans is not a single feature; each role supports different claims and demands specific standards of validity, shaping future use of LLMs in science.