Author argues psychiatry big data needs theory-driven, within-person intensive sampling and generative AI integration to address atheoretical predictive research

The author argues that big data research in psychiatry is currently dominated by atheoretical prediction, with theory and models taking a backseat. Relative to Stein et al.'s critical review of big data use, this article further frames the problem as a shift from 'all theory and no data' to 'all data and no theory,' using this as the premise for proposed improvements. This is a commentary argument based on descriptions and examples of the field's current state, not new empirical data.

The author argues that big data should be used more to test and inform theories of mental disorders, illustrating current model limitations with suicide risk prediction. The text notes that prediction models using numeric ICD codes in EHRs can help identify patients at greatest risk of suicide attempt in a health care system, but cannot inform how, why, or when suicidal thoughts and behaviors emerge, or how best to treat them. Argued through an existing study (Nock et al., JAMA Netw Open 2022) as an example; this is an opinion-based citation rather than new data from this article.

The author advocates developing idiographic psychiatry using deep digital phenotyping for intensive longitudinal study of individuals. The text reviews early experimental psychologists (Ebbinghaus, Pavlov, Wundt) and their idiographic methods, and notes that deep digital phenotyping offers unprecedented opportunities to strengthen such designs, emphasizing proactive planning for data collection that answers specific questions rather than default use of all available data streams. This is a historical review and methodological argument citing existing work, but the article provides no new idiographic data results.

The author advocates systematic examination of how to use generative AI safely and effectively in psychiatry, covering understanding humans, aiding clinical decision-making, and directly helping patients. The text raises the need to attend to how, where, and when humans should be involved in the loop, and notes that generative AI can help meet patients' unmet need for care while ensuring safety and effectiveness. This is a forward-looking perspective citing Torous et al. (World Psychiatry 2025) and others, but the article reports no empirical effects of AI interventions.

This article is aimed at psychiatric researchers, clinicians, and AI developers, and applies to discussions of big data research design, within-person monitoring, and generative AI applications in psychiatry. Implementing its proposals requires research environments capable of proactively planning data collection, as well as clinical or research institutions with digital phenotyping and AI deployment capacity.

This is a commentary article that provides no new empirical data or systematic review, so the actual effects of the three directions still require subsequent research. Additionally, the loaded text reflects an incomplete reading scope, lacking figures and full reference details, which may affect complete understanding of some argumentative context.