Encephale. 2026 Sep 19:S0013-7006(26)00179-X. doi: 10.1016/j.encep.2026.08.004. Online ahead of print.
ABSTRACT
OBJECTIVES: Precision psychiatry seeks to narrow the gap between knowledge generated at the group level and decisions made for an individual patient. This ambition responds to the limitations of current diagnostic categories, which bring together patients with markedly different clinical, biological, cognitive, and longitudinal profiles. Artificial intelligence may support this shift by combining multiple sources of data, identifying complex patterns, and tracking individual trajectories over time. This first part of the article, Tailoring Care to the Mind: The Emerging Role of Artificial Intelligence in Precision Psychiatry, clarifies what artificial intelligence means in psychiatry, the types of data on which it relies, and how it may reshape clinical reasoning.
MATERIALS AND METHODS: A narrative and conceptual review was conducted around three main themes: the major learning paradigms used in artificial intelligence and their epistemological limitations; the different sources of data that may contribute to precision psychiatry; and the potential transformation of diagnostic, prognostic, and therapeutic reasoning. The analysis considers clinical data, electronic health records, free-text clinical notes, digital and behavioral data, genomiques and other multi-omiques approaches, neuroimaging, and electrophysiology.
RESULTS: Artificial intelligence can identify complex relationships within data that are not always specified in advance. Depending on the method used, it may predict clinical outcomes, uncover latent subgroups, extract relevant features from raw data, exploit unlabeled information, or optimize sequences of therapeutic decisions. Supervised learning is mainly used for prediction, whereas unsupervised learning may reveal previously unrecognized patient profiles. Deep and self-supervised learning can derive representations from imaging, physiological signals, or unstructured text, while reinforcement learning may support adaptive treatment strategies. Transfer learning and federated learning may help address limited sample sizes and fragmented multisite data. However, these possibilities remain highly dependent on data quality, representativeness, interoperability, and longitudinal completeness. Moreover, most models identify statistical associations without, by themselves, establishing causal relationships. No single data source appears sufficient. Clinical, biological, neurophysiological, imaging, and digital information provide complementary perspectives on the patient. Their multimodal integration therefore represents both the major promise and one of the central challenges of the field. Artificial intelligence may help psychiatry move beyond a static clinical snapshot toward a more dynamic understanding of trajectories, facilitate the identification of transdiagnostic biotypes, and support therapeutic decisions that are repeatedly updated as new information becomes available. World models and digital twins extend this perspective by aiming to construct dynamic representations of the patient in which different clinical scenarios and treatment strategies could be simulated before their real-world application.
DISCUSSION: The value of artificial intelligence lies not only in improving prediction, but also in its potential to change how psychiatric phenomena are observed, connected, and followed over time. Yet the accumulation of additional data does not necessarily produce greater precision. Without a robust clinical and conceptual framework, complex models may reproduce, in algorithmic form, the same limitations and ambiguities already present in current diagnostic systems.
CONCLUSIONS: Artificial intelligence may contribute to a more dynamic, multimodal, and individualized psychiatry. Its clinical value will depend on its ability to clarify meaningful patient trajectories without confusing predictive performance, causal understanding, and clinical relevance.
PMID:42763257 | DOI:10.1016/j.encep.2026.08.004