AI Chatbot for Mental Health

For one of the assignments of my Abnormal Psychology class, I gave a summary presentation on a peer-reviewed paper on application of AI chatbot in reducing symptoms of MDD¹, GAD² and FED³. Secondary outcomes include working alliance. The paper is by a team from Dartmouth University and published in the New England Journal of Medicine. I also gathered published responses to the paper.

A key insight is that they fine-tune a GPT-3 class⁴ AI model with actual therapist-client dialogues based on CBT. This is akin to reading Shakespeare then speak and write like Shakespeare. Another is safeguard. There are clinicians monitoring the conversations. Interventions did happen. The app can be use 24/7, so the monitoring is probably not real time.

My takeaways:

• For healthcare institutions. If there is data, AI can be trained to deliver some care.
• For clinicians. Relatively high-functioning individuals likely choose this due to low-friction. The study excludes participants with severe symptoms. Clinicians might see more “severe” clients.
• For clients. Choose a service that can optionally reach a human clinician to look over the conversations with the AI bot.

References:

• Slides (on Google Drive): https://drive.google.com/file/d/1GrJXgZTBRV_quixvDy5lKu7T8fGA77Az/view?usp=drive_link
• Paper: Heinz, M. V., Mackin, D. M., Trudeau, B. M., Bhattacharya, S., Wang, Y., Banta, H. A., Jewett, A. D., Salzhauer, A. J., Griffin, T. Z., & Jacobson, N. C. (2025). Randomized Trial of a Generative AI Chatbot for Mental Health Treatment. NEJM AI, 2(4). https://doi.org/10.1056/AIoa2400802