The Predictably Unpredictable Patient: Exploring the Impact of AI-driven Personality Profiles on Medical Encounters

This project creates 5 different conversational AI agent personas based on the big five (OCEAN) personality traits and develops an evaluation framework to test how personality changes an AI agent's behavior. Every persona shares an identical medical illness and background history, so the personality difference can be isolated and emphasized for each one. The evaluation framework is rubric-based, with the goal of measuring of how the personas' unique traits affect the conversation, what symptoms they choose to share/hide, and how well they actually understand the medical advice. This project tests the behavior limits and flexibility of LLMs instead of just their ability to answer questions, and simulates the actual messy, anxious, and unpredictable ways real humans react to a medical issue.

Intern: Sarayu Aerpula

Mentors: Dr. Samantha Levy (AOS/QAT), Julia Thorne (AOS/QAT)