AI in the OR can sharpen skills, but humans make the lessons stick

Editor's Note

Artificial intelligence (AI) can track surgical performance with pinpoint accuracy, but true mastery still requires a human teacher, American Council on Science and Health August 20 reports. A randomized trial of an AI-powered surgical tutoring system found that while algorithms provided real-time error detection, the best learning happened when human instructors personalized that feedback—even if it left students frustrated.

As detailed in the article, the study centered on the Intelligent Continuous Expertise Monitoring System (ICEMS), an AI tool designed to assess neurosurgical training. ICEMS continuously quantified performance on a scale from novice to expert by analyzing movements of key surgical instruments. The system flagged deviations from expert benchmarks and categorized performance across safety, efficiency, quality, movement, and bimanual coordination.

Per the outlet, 87 medical students were randomized into three groups: those receiving direct AI feedback, those guided by residents simply repeating the AI’s feedback, and those trained by residents who received AI alerts but delivered tailored, humanized instruction. After practice sessions, students performed a simulated brain resection designed to test skill transfer under realistic conditions.

All groups improved but none reached expert-level proficiency. Human-delivered instruction, even when repeating algorithmic prompts, outperformed AI-only feedback. The best results came from AI-augmented personalized instruction, where residents adapted their coaching to the learner’s needs. Students in this group not only scored highest on performance but also transferred skills most effectively to the complex resection task.

The study also found frustration was common across all groups, and highest with personalized coaching. Rather than impeding progress, this discomfort appeared to enhance learning, supporting the concept of “desirable difficulty” where challenges drive deeper skill acquisition. Students improved in reducing tissue injury and bleeding risk, and patient-like outcomes were measurably better after the more difficult training.

The outlet notes the trial highlights both promise and limits of AI in surgical education. While ICEMS delivered precise analytics at a scale humans cannot match, only human instructors could translate those metrics into meaningful, nuanced guidance. Even the presence of a human voice improved engagement compared with automated prompts. Ultimately, the article concludes, AI may become an essential surgical training partner, but it cannot replace empathy, context, or the capacity to push learners through the right amount of struggle.