AI in medical education: a faculty’s playbook for the next five years

Generative AI did not ask permission to enter your curriculum — students brought it. The decision in front of academic leadership is no longer whether to have a position, but which one, and the next accreditation cycle will ask for it in writing. The schools that do this well will treat it as a curriculum design problem with four distinct layers, not a single ban-or-allow policy.

If a take-home essay can be produced by a model in forty seconds, the essay is no longer measuring the student. The response is not surveillance — detection tools lose that arms race — but redesign: put the weight on performances that require the student to be present and reasoning. An observed clinical conversation. A defence of a diagnostic decision, questioned in real time. A simulation attempt with its transcript attached.

The irony is that these were always the better assessments; they were just too expensive to run for every student. Recorded, rubric-scored simulation is what makes the defensible version of assessment affordable at cohort scale.

Teaching changes in one specific way: recall is no longer the scarce skill. What stays scarce is knowing when the model is wrong — the pharmacology interaction it missed, the plausible-sounding contraindication it invented. That is a teachable skill, and it looks like practice: give students AI output about a case and grade the critique, not the summary.

The content layer is the upside almost no school has claimed yet: faculty using AI to turn their own materials — slide decks, protocols, clinical notes — into simulations, cases and interactive modules, with the faculty member as clinical editor rather than production bottleneck. The institutions that build this muscle first will iterate their curriculum at a pace the others cannot match.

The accreditation layer is the least glamorous and the most urgent, because it has a deadline you do not control. A defensible written position covers four things: which uses are allowed and expected of students, how assessment integrity is protected by design rather than by detection, how faculty-created AI content is clinically reviewed before it reaches a student, and when the policy is next reviewed. One page is enough. The difference between having it and not having it is the difference between a deliberate policy and a scramble in front of a review panel.

The content and assessment layers are where a school can act this year without waiting on policy: faculty-authored AI simulations that train and assess the reasoning a written exam no longer can. It is the most concrete first move, and the easiest to show a board.

See the platform faculties are using →