AI in Mental Health
Psychiatry is an unusually hard test case for AI. Its diagnostic categories are contested and partly socially constructed. Its “ground truth” labels are noisy and shaped by who gets access to care, and the people it serves often face stigma, discrimination, coercive treatment, and an overlap between the mental healthcare and carceral systems. Computational psychiatry has mostly asked whether models can predict diagnoses, symptoms, or treatment response accurately. So far, the field has delivered little clear clinical value and has often promised technical fixes for problems that are deeply social.
This working group brings computational psychiatry into conversation with the algorithmic fairness, mechanism design, and science and technology studies communities. Those communities have built tools for thinking about measurement, power, performativity (predictions that change the outcomes they predict), recourse, model multiplicity, and how clinicians and AI make decisions together. We want to ask what “useful” and “fair” should mean for mental health AI, rather than only “accurate.”
Alongside papers, data is a central focus. In each session we examine a dataset that currently powers mental health AI: neuroimaging consortia, population biobanks, electronic health records, smartphone sensing, social media, and clinical conversation corpora. We ask who is represented, how labels were produced, who consented to what, and which questions the data cannot answer. Over the semester we will build a shared, documented catalog of existing datasets and a wishlist of the datasets that should exist, along with the governance needed to make them safe to share.
We welcome researchers in machine learning, statistics, neuroscience, psychiatry, psychology, law, philosophy, and the social sciences, as well as clinicians and policy practitioners.
Working Group Organizer #
We welcome involvement in this working group! If you’re interested in joining, please reach out to the organizer:
| Name | Position | Affiliation |
|---|---|---|
| Saige Rutherford | University of Michigan |