Argue that reviewing and editing AI-generated clinical notes is a teachable competency that medical education should address, and propose simulation-based education as a setting for teaching and evaluating it.
Approach:
Evidence reviewed: The article summarizes reported effects of ambient AI scribes on documentation time and cognitive workload, alongside evidence about AI-generated note errors and human review.
Educational rationale: It contrasts the reasoning involved in independently authoring notes with the potentially less effortful process of reviewing fluent AI-generated drafts.
Proposed setting: It identifies simulation-based education as a venue for structured comparisons, controlled exposure to errors, and standardized measurement of trainee development.
Key Findings:
In a five-center assessment of 8,581 clinicians, adopters reduced EHR time by a mean of 13 minutes per 8-hour workday; resident physicians saved 94 minutes per 8 hours.
In a Yale pilot of 47 trainees, NASA-TLX workload scores decreased from 56.3 to 43.3 after two months of use (p < 0.001), and adoption reached 91.5%.
Existing deployment studies have emphasized time and burden, not whether clinicians detect errors in AI-generated notes or how use affects trainee learning and reasoning.
A systematic review reported error rates above 50% in conversational and multispeaker scenarios. Comparative assessments found AI notes rated higher for thoroughness and organization but lower for accuracy and internal consistency; omissions were the predominant error.
The article reports that no study has assessed the accuracy of notes trainees approve, their detection of embedded errors, or whether reasoning developed through independent note writing is retained, redirected, or displaced.
Interpretation:
The authors argue that editing AI-generated documentation should be taught deliberately: reviewing does not necessarily engage the same reasoning as authoring and may expose trainees to automation bias. They propose simulation-based education to develop and assess this competency.
Limitations:
The available literature does not establish whether clinicians identify errors in AI-generated notes or whether independent documentation supports trainee reasoning.
Reported error rates are derived predominantly from validation studies rather than real-world deployment.
The provided context ends during the discussion of current note-authoring education and does not include the full competency framework.
