How AI Could Change Surgical Training
Updated: 3 days ago
Surgical residents are typically judged by whichever faculty member happened to be in the room.
That model has changed little since residency training was formalized over a century ago.
Case exposure also depends on volume, scheduling, and competing trainees, not on what an individual resident needs to learn.
A Perspective published in npj Digital Surgery on 1 September 2026 by researchers at Harvard Medical School and Mass General Brigham proposes a different approach.
Practice Questions Without a Senior Surgeon
Large language models can run oral board style practice at any hour.
Oral board practice normally needs a senior surgeon, and the most experienced board examiners are often unable to run mock exams for trainees.
A language model can also generate unlimited case scenarios, so residents are not limited to the pathology that happens to appear during a rotation.
Feedback From Every Operation
Computer vision can mark safe and dangerous dissection zones in the operative field.
After the case, the same system can produce structured feedback, even when no expert faculty member watched.
Skills scoring software adds a second layer by tracking instrument motion and economy of movement.
In published studies, this scoring classifies proficiency at levels approaching expert human raters.
One Record Per Trainee
The authors argue that these three tools matter most when they are combined.
Knowledge gaps from language models, decisions from video analysis, and technical execution from skills scoring would feed one continuous record.
That record could show a specific gap early, instead of at a periodic review.
Smaller and community programs could then receive feedback that has mostly existed at large academic centers.
What Programs Should Ask First
The paper describes this model as prospective, so no program should treat it as a finished product.
The authors' own cautions point to three questions before adopting any of these tools.
First, was the tool validated against real educational and operative outcomes, and not only against expert raters?
Second, does it teach the process of surgical reasoning, or does it only hand over answers?
Third, who owns and controls the operative video data?
The Deskilling Risk
The paper cites a multicenter observational study on deskilling risk among endoscopists after exposure to AI in colonoscopy.
The authors conclude that training built around AI must still produce surgeons who can operate and reason without it.
Until these tools are validated, they describe them as decision support, not as independent judges of competence.
Which surgical AI tools have published evidence behind them, and which rely on company claims, is tracked in the NextInSurgery database.
