PhD defence Anna van Oosterzee: Responsible AI in psychiatry

Artificial intelligence is increasingly being used to detect diseases that are invisible to doctors. Can such AI models also be applied to psychiatry? On Wednesday 26 August, Anna van Oosterzee will defend her PhD thesis ‘The Future of Psychiatric Diagnostics: Philosophical Reflections on Artificial Intelligence and Mental Health Care’.

AI in psychiatry

Unlike most of medicine, psychiatry lacks well-established biological disease mechanisms, Van Oosterzee writes. These are the underlying biological processes that explain the onset and course of a disease. This uncertainty shapes the kinds of AI models that can be developed for psychiatry, what they can predict, and what they can ultimately tell us about mental disorders.

An overarching challenge is the DSM, the diagnostic and statistical manual of psychiatric disorders. AI models are trained on information from the DSM, but the manual has structural problems, such as heterogeneous categories and overlap between diagnoses. These problems limit what the models can predict, explain, or monitor.

Responsible use of AI

In her dissertation, Van Oosterzee explores the landscape of data-driven diagnostic AI models through the lenses of philosophy of science and ethics. This perspective addresses different questions: philosophy of science asks what knowledge AI models actually produce, for example, while ethics examines what is needed to use these models responsibly in care.

The fact that certain AI models can make accurate predictions is not, in itself, sufficient for responsible implementation, Van Oosterzee says. Each model, she argues, has its own benefits, limitations, and ethical challenges.

A framework for assessing AI in psychiatry

To give researchers, developers, and healthcare institutions insight into the value of AI models for psychiatry, Van Oosterzee developed an assessment framework. This philosophical framework helps assess whether a model predicts accurately, the data used are representative, patient privacy is protected, outcomes are understandable to clinicians, and the model genuinely adds value to care.

In doing so, Van Oosterzee offers both a conceptual map of the field of AI in mental healthcare and practical guidance for the development of ethical AI technologies.

Prior to the defence, Van Oosterzee will give a layman’s talk starting at 16:00.