John Lyon Havlik

Hello! I'm the inaugural resident in Stanford's Humanities and Social Medicine Research Track. Psychiatric and substance use disorders are usually caught only after substantial, sometimes irreparable, harm has already occurred. My work is about changing that: I build artificial intelligence models that identify people earlier, and I evaluate whether the systems that receive them — health plans, treatment facilities, and now generative AI chatbots — are equipped to respond appropriately. I believe this kind of early, well-supported intervention will one day be commonplace and prevent a great deal of human suffering.

On the building side, I'm technical lead and first author on a supervised model that detects substance use disorders from EHR-like data, published in Nature Mental Health; related work uses multimodal data to flag who will go on to develop opioid use disorder and to predict imminent psychiatric emergency-department visits. I also co-authored a Nature Medicine paper on how to rigorously test the true capabilities of medical AI. I work across a wide range of deterministic and probabilistic methods, including the transformer architectures behind modern "chatbots."

In production — spinout AI-enabled clinical workflow automation software I developed is in use at the U.S. Department of Veterans Affairs and here at Stanford. The aim is always to move models out of publications and into the clinic.

On the evaluating side, I write on psychiatric services and health policy — recently in JAMA Health Forum on private-equity residential treatment and JAMA Network Open on the economics of telemedicine "house calls" — and I serve as an expert reviewer of AI-in-medicine trials for the Utah Office of Artificial Intelligence Policy. Alongside the research, I founded and lead Research Roundtable, a mentorship group whose undergraduates have published 10+ first-author papers. Before medicine, with Peggy Mason at UChicago, I found the human bystander effect in rats (Science Advances).