Skip to main content Skip to navigation

Synthetic Patients: Simulating Difficult Conversations with Multimodal Generative AI for Medical Education

Project Overview

The document explores the application of multimodal generative AI in medical education, specifically focusing on the creation of synthetic patients to enhance training for healthcare providers in managing complex patient conversations. This innovative tool enables learners to interact with realistic avatars that reflect diverse patient backgrounds, thereby improving communication skills in a cost-effective and scalable manner compared to conventional training methods. It details the development process of the AI system, highlighting its benefits, such as increased accessibility to training scenarios and the ability to simulate a wide range of patient interactions. The document also addresses the challenges faced during implementation and the overall outcomes of utilizing this technology in educational settings, underscoring its potential to transform medical training by offering immersive, hands-on experiences that prepare future healthcare professionals for real-world interactions.

Key Applications

Synthetic patients for simulating difficult patient-provider conversations

Context: Medical education for healthcare providers and trainees, particularly in palliative care

Implementation: Developed a web-based application integrating patient profiles, multimedia generation, and real-time interaction with AI-generated avatars.

Outcomes: High-fidelity simulations that enhance engagement and provide realistic conversation practice, which can improve trainee confidence and competence in handling sensitive discussions.

Challenges: Initial development required significant time and iteration; there were issues with the realism of generated patient avatars and multimedia, including stereotypical representations and video distortions.

Implementation Barriers

Technical

Challenges with generating realistic multimedia, including voices and visual representations of patients.

Proposed Solutions: Iterative refinement of patient profiles and the use of multiple tools for multimedia generation, alongside manual corrections.

Resource Allocation

Development of the system components was time-intensive and required significant investment in labor.

Proposed Solutions: Leverage existing generative AI models and open-source tools to reduce the need for specialized programming knowledge and lower costs.

Ethical

Concerns about reinforcing biases through stereotypical representations and the impact on traditional standardized patient actors. Ensure diverse representation in patient profiles and maintain collaboration with human actors in medical education.

Proposed Solutions: Implement strategies to ensure diverse representation in patient profiles and sustain collaboration with human actors in medical education.

Project Team

Simon N. Chu

Researcher

Alex J. Goodell

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Simon N. Chu, Alex J. Goodell

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies