This research evaluates whether LLM-driven generative agents accurately represent human opinions in healthcare decision-making contexts.

• Significant differences discovered between real human survey responses and LLM-simulated responses on healthcare decisions
• LLM-generated responses showed systematic biases in medical decision-making scenarios
• Research reveals limitations in using AI agents as reliable proxies for human behavior simulation
• Provides critical insights for medical researchers considering AI-simulated human responses

This work has important implications for medical research ethics and methodology, highlighting the current limitations of LLMs in accurately modeling healthcare decision-making processes without introducing systematic bias.

Evaluating the Bias in LLMs for Surveying Opinion and Decision Making in Healthcare