Generative AI for Psychotherapy: The Development and First Trial of Therabot
Event Details:
Therabot: 6.5 Years and Over 100,000 Human Hours Building and Evaluating the First Generative AI Psychotherapy Chatbot
Please join us on March 18th via Zoom from 12-1pm to listen to Dr. Jacobson's follow up from his Grand Rounds talk (3/12).
Register: https://stanford.zoom.us/webinar/register/WN_42crm0AYTXucnWKnZmVetg
About the speaker: Dr. Jacobson's research focuses on using technology to improve mental health assessment and treatment. His lab develops AI systems that analyze passive sensor data from smartphones and wearables to detect and predict changes in anxiety and depression. In this talk, he'll describe the 6.5-year path to building Therabot, a generative AI chatbot for psychotherapy recently tested in a randomized controlled trial published in the New England Journal of Medicine AI. Early attempts training on peer support forums and psychotherapy transcripts failed. The team then built custom training data from scratch, investing over 100,000 hours of expert clinical supervision and refinement. The resulting chatbot was tested with 210 participants across depression, anxiety, and eating disorder risk groups. Compared to waitlist controls, users showed large and significant symptom reductions across all three conditions. Average engagement exceeded 6 hours per person across 24 separate days, breaking the typical pattern where users abandon mental health apps within the first week. He'll discuss what worked, what didn't, and implications for the broader mHealth field.
CME Credits: You may be eligible to receive CME credits for attending this Hubinar. More information will be provided at the start of the event. If you have questions, please reach out to the Tech Hub (techhub-psychiatry@stanford.edu)
Recording: This event will be recorded.
Register: https://stanford.zoom.us/webinar/register/WN_42crm0AYTXucnWKnZmVetg
