Objective:

This study outlines a step-by-step process for developing customized artificial intelligence (AI) powered virtual patient actors (VPs) for clinical simulation in medical education, including production, testing, and avoiding potential pitfalls that could limit its reliability and implementation.

Background:

AI has created new opportunities to enhance clinical training and assessment in healthcare education. Among the innovations, VPs generated by large language model (LLM) AI offer a cost-effective and scalable alternative to traditional standardized patients played by human actors, which have served as the standard for clinical skills simulation but are limited by cost and inter-actor variability. Advances in LLMs can simulate authentic patient interactions, with dynamic conversations for practicing history gathering and clinical reasoning.

Design/Methods:

We developed a five-step framework for creating AI VPs using ChatGPT. 1. Define learning objectives. 2. Create the patient prompt, including demographics, personality traits, and medical history to define the VP’s role. 3. Configure the VP actor to stay in character and regulate its information disclosure. 4. Test and refine responses to ensure operation within defined parameters. 5. Implement as a standalone clinical skills activity or embedded in a clinical reasoning assessment.

Results:

Production of VPs capable of natural conversation can be done efficiently and provide a novel tool for medical educators. Potential barriers to implementation include the need for careful quality assurance to avoid AI confabulation (“hallucinations”), minimize straying off prompt, identifying model bias, and ensure transparency with human participants.

Conclusions:

When carefully tuned and implemented, AI VPs offer accessible, scalable, and consistent, custom clinical education tools at low cost. By reducing resource barriers for educators, trainee education can be enhanced with improved and expanded simulation, assessment equity, and individualized learning. Future developments may include video and voice integration, integration into existing physical mannequin-based simulation platforms, and AI-powered analysis of exercises to coach participants in refining their clinical skills.