Eike Buhr

Division of Medical Ethics, Department of Health Services Research, University of Oldenburg

The use of AI-based systems in psychiatry, for example in the context of predictive models or continuous behavioural data collection, is accompanied by a reconfiguration of the temporal structures of psychiatric diagnosis. Whilst psychiatric practice integrates point-in-time findings into diagnostically relevant longitudinal contexts, AI-supported applications often tend to generate diagnostic evidence by aggregating temporally fragmented data points, patterns and risk scores. This paper examines the epistemic and ethical consequences of these differing temporal logics for clinical decision-making processes.

A central argument put forward is that AI-based modelling can imbue diagnostic processes with its own, sometimes irrelevant, temporal logic, whereby mental disorders appear as aggregated risk profiles rather than as temporally embedded trajectories. This shift is ethically relevant, as clinical assessments of the severity of illness, the need for treatment and well-being rely fundamentally on temporally extended and context-sensitive understandings of mental states. This is exemplified by affective phenomena such as grief, the significance of which only becomes apparent over time and which can be prematurely pathologised through isolated measurements. Conversely, a more aggregated analysis may obscure time-limited manic episodes, which are clinically central to distinguishing depressive disorders from personality disorders. These examples illustrate that different algorithmic temporal logics generate diagnostic and normative risks, without calling AI-supported diagnostics into question across the board.

Based on a conceptual analysis of psychiatric classification logics (e.g. temporal criteria of diagnostic categories) and psychopathological concepts of disease course (e.g. episodic or chronic disease courses), this paper demonstrates that a temporally sensitive medical ethics is required to provide a normative framework for the use of AI-based diagnostics. It must clarify under what conditions temporally fragmented evidence can constitute an appropriate basis for clinical decisions, where biographical temporal dimensions remain indispensable, and to what extent such temporal dimensions can be incorporated into AI-based diagnostics.