[Artificial intelligence in the early detection of psychotic disorders: opportunities and limitations of clinical implementation].

Early detection of psychotic disorders is complicated by nonspecific symptoms, heterogeneous trajectories and the limited ability of established clinical instruments to generate accurate individual-level predictions. Data-driven artificial intelligence can integrate complex clinical and biological data and achieves moderate to good predictive performance, which often declines on external validation. Despite substantial methodological advances, a marked translational gap between model development and clinical application persists, mainly due to limited generalizability as well as practical, ethical and regulatory requirements. Artificial intelligence can complement clinical expertise and facilitate access to appropriate psychiatric care provided that it is implemented transparently and close to real-world care, subject to health economic evaluation and developed with the active involvement of experts by experience.