Simulation Faculty Need Shared AI Governance Rules, Not More Tool Demos

Simulation educators are calling for structured frameworks to govern AI use in CME rather than relying on scattered tool demos and extended expert modeling.

AI simulation needs a map before it scales

A simulation journal club discussion this week framed AI use in healthcare simulation as scattered rather than systematized, with examples ranging from scenario writing to avatars, data analysis, feedback, and dissemination. The proposed structure separates five simulation domains — education, assessment, faculty development, translational simulation, and research — from cross-cutting applications such as design, delivery, data collection, evaluation, and dissemination (Simulcast Journal Club).

For CME providers, the important part is not the taxonomy itself. It is the governance layer beneath it: ethics, bias, consent, AI literacy, cybersecurity, and accountability. A related urology conference conversation made the same problem concrete outside simulation: AI can make clinical information more accessible, but it also raises risks around hallucination, misrepresentation, and unauthorized or biased reuse of guideline material (GU Cast; YouTube).

The caveat is clear: these are mostly educator and conference voices, not a broad sample of independent practicing clinicians. But the implication is still portable. Simulation-based CME that uses AI for cases, avatars, feedback, summaries, or assessment should not treat those uses as isolated production choices. Each use should have a named purpose, a faculty-development requirement, and a risk check.

We saw a related pattern in an earlier brief on clinicians naming the real barriers to AI adoption in practice: trust problems are often operational before they are philosophical. For CME teams, the question is simple: can faculty explain where AI enters the learning experience, what it is allowed to do, and how learners or instructors should challenge it?

Procedural formats should spend less time showing

The second signal came from a randomized dual eye-tracking study in sonography education. Learners assigned to extended coaching outperformed those assigned to extended modeling by 12% in dynamic image interpretation, with better time efficiency in tasks that required translating perception into action (Medical Education Podcasts).

This is not a license to generalize across all procedural CME. The study was small, sonography-specific, and based on a brief training period. The authors also noted that coaching benefits were less clear for static tasks. Still, it gives CME teams a useful design challenge: many hands-on workshops still protect too much time for expert demonstration because it feels efficient and faculty-centered.

For dynamic procedural skills, efficiency may come from earlier supervised learner action, not longer modeling. That means agenda design should identify which minutes are for expert thinking aloud, which minutes are for learner attempts, and which minutes are for targeted correction. A strong workshop is not the one with the most polished demonstration; it is the one where faculty can see enough learner performance to intervene.

The concrete question for CME teams: in the next procedural activity, what proportion of faculty time is spent performing expertise versus watching learners perform and coaching the next attempt?

What CME Teams Should Reconsider

• Audit AI-enabled simulation activities against purpose, faculty role, learner disclosure, bias checks, and escalation rules.
• Require faculty-development support before asking instructors to use AI-generated cases, avatars, feedback, or summaries.
• For procedural workshops, separate agenda time for modeling from agenda time for supervised learner attempts and feedback.
• Add at least one transfer-oriented measure to coaching-heavy pilots, such as dynamic interpretation, time efficiency, or observed performance quality.

What to reconsider next

The common thread is control. AI tools and expert demonstrations both look useful when they are visible, polished, and easy to schedule. But this week’s education-led signals point CME providers toward the less glamorous work: defining faculty rules, protecting learner autonomy, and measuring whether simulation time produces better performance rather than better presentations.

Provides concrete examples of AI applied to design, delivery, evaluation and dissemination phases.