The demand for evidence-based mental healthcare far exceeds the supply of clinicians trained to deliver it. Nearly half of all individuals with mental illness in the United States will never receive any form of care, and the quality of the care for those fortunate enough to receive services is highly variable. Unless there are significant changes to the behavioral health care workforce, the US Health Services and Research Administration estimates that these shortages will persist well into the next decade, leading to unnecessary morbidity and mortality and significant economic costs.
One reason for the shortage of trained mental health clinicians is a training bottleneck. Traditional models of clinical training require significant time by highly trained clinicians and opportunistic encounters with patients who presentations align with trainees’ learning objectives. This model constrains the number of individuals able to be trained at any given time as well as the diversity of interventions and populations those individuals can be trained in.
The goal of this project is to develop an artificial intelligence (AI) based simulated patient with whom health sciences trainees (e.g., medical students, residents, graduate students) can interact to practice evidence-based interviewing and psychological intervention skills. We will focus especially on being able to simulate a diverse range of patient interactions, psychologically plausible interactions (e.g., avoiding agreeable, instantly cured patients), and realistic multi-turn and multi-session simulations that integrate sensible patient trajectories and simulated life events (moving beyond the current, static, single-session roleplay that tends to get stuck in a single state).
We will train a large language model on real-life clinical interaction data (e.g., psychotherapy recordings), which will allow the system to “learn” how to better represent the types of clinical interactions trainees will experience in their training clinics. We will also develop an AI-based system to provide feedback to the trainee in real time, allowing them to refine their skills with iterative practice known to enhance learning.
To learn more, please contact Adam Kuczynski, PhD or Devon Sandel-Fernandez, PhD