The Future of Psychiatric Diagnostics: Philosophical Reflections on Artificial Intelligence and Mental Health Care

Abstract

Artificial intelligence is increasingly being used to detect diseases that remain invisible to human clinicians. But can these same technologies be applied to psychiatry? Is depression a disease in the same way as cancer? Unlike most of medicine, psychiatry lacks well-established biological disease mechanisms. This uncertainty shapes the kinds of AI models that can be developed, what they can predict, and what they can ultimately tell us about mental disorders. This dissertation explores the landscape of data-driven diagnostic models in mental healthcare through the lens of the philosophy of science and of ethics. It distinguishes between models that predict-, models that control-, and models that aim to explain psychiatric phenomena, arguing that these different approaches rely on different assumptions and should therefore be evaluated by different standards. Across a series of conceptual analyses, it examines supervised machine learning, behavioural prediction models based on digital phenotyping, unsupervised clustering techniques, network models of psychopathology, and complexity-theoretical approaches to mental disorders. The dissertation argues that predictive accuracy alone is insufficient for responsible clinical implementation. Instead, different AI models should be understood as serving different epistemic and clinical functions, each with their own benefits, limitations, and ethical challenges. Building on this analysis, it develops a philosophical framework for evaluating AI in psychiatry that integrates scientific validity with ethical considerations such as privacy, bias, explainability and clinical usefulness. In doing so, the dissertation offers both a conceptual map of the field of AI in mental healthcare and practical guidance for the development of ethical AI technologies.