JMIR Publications

ABSTRACT

Artificial intelligence (AI) increasingly retrieves, organizes, synthesizes, and interprets medical information and generates recommendations. What, then, must physicians still know without assistance? We propose cognitive externalizability as a function-based curricular heuristic: the degree to which a specific knowledge element or cognitive operation can be supported externally without compromising the construction of expertise, timely clinical action, error detection, or critical professional judgment. Unlike cognitive offloading, distributed cognition, AI entrustment, or collaborative intelligence, cognitive externalizability asks not whether cognition can be externally supported or a task delegated, but what must remain internally available to the physician when external support exists. Four domains structure this appraisal—structural value, immediate availability, supervisory value, and reliability of external support—applied at the level of the specific knowledge element or cognitive operation and weighted by learner stage, clinical context, scope of practice, and technology. We argue for multidisciplinary, periodically revisable curricular decisions; deliberate protection of competence vulnerable to deskilling and never-skilling; assessment of both unassisted reasoning and supervised human-AI collaboration; and empirical testing of the framework. Tomorrow’s physicians will need a robust internal cognitive architecture precisely because their tools will be powerful: to understand, act in time, detect error, manage uncertainty, and exercise accountable clinical judgment.