Trainees Adopt AI Faster Than Supervisors Can Guide Them

Trainees are adopting large language models faster than supervisors can establish timing rules, critical-thinking prompts, and measures of preserved reasoning. Clinician threads from internal medicine and radiology make the gap concrete enough for CME teams to revisit how they prepare both learners and faculty for AI-supported education.

AI use needs timing rules, not just access

The sharpest clinician concern was not that trainees are using AI. It was that they may be using it before they have enough independent reasoning practice to know when the output is wrong, incomplete, or simply too easy to accept.

One internal medicine clinician shared concern after receiving trainee work that had been generated by AI and left insufficiently checked, adding: “AI should be a tool for learning, not a substitute for critical thinking.” (source) A separate clinician post pointed to the same training issue through the lens of clinical supervision: learners may be more comfortable with large language models than their supervisors are, which changes the basic supervision task. (source)

Radiology added a useful parallel. In an RSNA podcast discussion of AI-supported screening mammography, the speakers described why implementation cannot stop at model performance. They discussed how AI risk categories and markings can change visual search, including concerns about automation bias and reduced search outside marked regions. (source) The specialty context is radiology, but the education issue is broader: AI can change what learners attend to, not just what answer they produce.

For CME providers, the implication is that AI education should include a sequence. Learners may need an unaided attempt first, followed by AI comparison, then a structured explanation of what they accepted, rejected, or revised. That builds on an earlier brief on targeted verification drills, but this week’s signal moves the issue upstream: verification habits need supervision, timing, and faculty participation.

The concrete question for CME teams: where in your AI modules do learners have to demonstrate reasoning before AI assistance, and where do faculty evaluate whether that reasoning improved or narrowed?

Faculty preparation is too important to leave episodic

The second signal came mainly from provider-owned CPD content, so it should be read as a field example rather than independent clinician consensus. Still, the operational details are useful.

In an Alliance Podcast episode on faculty development, the discussion centered on familiar but persistent problems: engagement, overloaded slides, inconsistent use of audience-response tools, role clarity for moderators and panelists, and the need to help faculty use technology and AI in ways that support learning rather than decorate a session. (source) The examples came from neurosurgery and orthopedic surgery CPD, but the faculty-development problem is portable.

The important move is from speaker invitation to faculty pathway. A planning call with the chair, role-specific guidance, review of slides, polling that has a real learner purpose, Q&A planning, role play when appropriate, and speaker self- or peer-evaluation are not extras. They are part of the educational intervention.

This matters because the AI supervision issue above depends on faculty readiness. If faculty are expected to teach learners how to question AI, they also need preparation on how to structure that questioning. CME teams should ask whether faculty development is a recurring capability inside the organization or a set of reminders sent shortly before an activity.

What CME Teams Should Reconsider

• Review AI literacy courses for three missing elements: timing of AI exposure, faculty-trainee supervision prompts, and measures of preserved reasoning.
• Add an unaided-before-assisted step to AI exercises, then require learners to explain when they accept, reject, or modify AI output.
• Treat speaker preparation as a pathway: chair planning, role-specific facilitation guidance, technology prep, and learner/faculty feedback.

What changed this week

The AI conversation moved from tool competence to supervision competence. That is a harder education problem because it crosses learner behavior, faculty readiness, and outcomes measurement. For CME providers, the next test is not whether an activity mentions AI. It is whether the activity teaches clinicians when to pause, reason independently, and supervise the tool rather than simply use it.