From the DSM to ChatGPT: When the Framework Becomes the Boundary of Attention

PREPRINT — Not peer reviewed. Clinical frameworks help clinicians decide which information deserves attention. Diagnostic categories, differential diagnosis references, checklists, algorithms, and generative artificial intelligence serve different purposes, but each organizes a limited view of the patient. Building on a related discussion of AI-mediated attentional displacement, this Perspective examines the broader relationship between clinical frameworks and continued inquiry. Three propositions guide the argument: frameworks should direct attention without becoming its boundary; accuracy and attentional safety are distinct; and safe frameworks must remain permeable to findings outside their current explanation. Hypothetical clinical examples illustrate how a correct principal diagnosis may coexist with consequential unexplained findings, and how workload can discourage further inquiry. Practical questions include what remains unexplained and what would prompt reconsideration. Potential supports include source access, explicit uncertainty, opportunities for colleagues and patients to challenge an account, and purposeful diagnostic pauses. These are proposals for evaluation, not validated safeguards. Assessment should consider decision quality, recognition of consequential exceptions, and workload together. Clinical judgment includes knowing when to use a framework, revise it, or leave it behind. Related work: This Perspective extends the author’s related manuscript, “Artificial intelligence-induced attentional displacement in clinical diagnosis,” submitted to JAMIA and publicly available as an unreviewed preprint (doi:10.5281/zenodo.22727468). The present article examines clinical frameworks generally; the related manuscript focuses on AI-mediated shifts in attention. The relationship is described in the accompanying journal submission cover letter. AI assistance: OpenAI ChatGPT and Codex were used during development and revision to help retrieve sources, organize the author’s prior notes and correspondence, draft and edit prose, check references, and prepare documents. Codex assisted with the final revision and publication preparation on September 15, 2026. The author originated the central argument, directed the scope and revisions, reviewed the manuscript, and takes responsibility for its content. No empirical results were generated. All clinical scenarios are hypothetical.