Abstract

Generative artificial intelligence (Gen-AI) chatbots have the potential to deliver scalable, personalized mental health treatments, addressing the global treatment gap as well as the engagement and retention challenges common in digital therapeutics. We present the development and evaluation of Therabot, a Gen-AI–powered therapy chatbot, which has been developed and refined over 6 years via over 100,000 human hours from members of our team. Therabot was developed through iterative finetuning and testing of many foundation large language models (LLMs) and many training data types. 

Therabot users demonstrated significantly greater symptom reductions in major depressive disorder compared to controls. This is the first RCT demonstrating the effectiveness of a fine-tuned fully Gen-AI therapy chatbot for treating clinical-level mental health symptoms. Iterative LLM refinement, including tailored, evidence-based psychotherapy training data, was critical to its success. Further research is warranted to confirm these findings in larger samples. Next steps should include establishing safety and efficacy benchmarks for Gen-AI therapy chatbots. Despite the growing number of foundation LLMs, users should exercise great caution when using general LLMs for mental health applications.

Speaker Bio

Michael V. Heinz is a research psychiatrist at Dartmouth College and Dartmouth Health, with an interest in scalable digital technologies for assessing and treating mental health problems. He is completing a research fellowship at the Artificial Intelligence and Mental Health Lab at the Dartmouth Center for Technology and Behavioral Health. His work there emphasizes the use of passively collected data, such as movement and heart rate, to understand mental health problems. Dr. Heinz has co-led multiple clinical trials, including an app-based digital intervention for co-occurring substance use disorders and the ongoing Therabot trial, which explores the effectiveness and feasibility of an AI-driven therapy robot. As a member of Dartmouth’s Psychiatry Immunology and Neurology Group, he engages in a multi-hospital collaboration investigating neurological and psychiatric disorders following infections or illnesses. His research has allowed for productive collaborations with various industry partners, including Microsoft Research and Artisight. Dr. Heinz sees adult clients at Dartmouth Health’s Hanover Psychiatry, specializing in the treatment of mood and anxiety disorders through medications and psychotherapy with training in interventional treatments, including electroconvulsive therapy (ECT) and transcranial magnetic stimulation (TMS).

Nicholas C. Jacobson is an associate professor in the departments of Biomedical Data Science, Psychiatry, and Computer Science at Dartmouth College’s Geisel School of Medicine. He serves as the Director of the Treatment Development and Evaluation Core within the Center for Technology and Behavioral Health and leads the AI and Mental Health: Innovation in Technology Guided Healthcare (AIM HIGH) Laboratory. Driven by a passion to transform mental health care through technology, Dr. Jacobson focuses on harnessing artificial intelligence and passive sensor data from smartphones and wearable devices to develop scalable, personalized interventions for anxiety and depression. His expertise lies in creating personalized just-in-time adaptive interventions and advancing quantitative tools that enable this work.