Artificial Intelligence in Surgery

Translated by Specialist Level I Doctor Trần Sang

AI-augmented learning has emerged, and many believe it surpasses all previous methods. AI’s ability to provide objective, real-time, and standardized feedback marks an unprecedented turning point.

Giglio et al. investigated whether AI feedback could accelerate the acquisition of surgical skills using virtual reality simulation. Medical students (N = 87) were randomly assigned to three groups:

• Received computer-generated verbal feedback based on AI performance assessment.
• Received human verbal feedback but using an identical script to AI.
• Received human verbal feedback personalized based on AI performance assessment.

Results showed that group 3 performed the best; however, group 2 also outperformed group 1 on the fifth attempt and in the final virtual reality simulation task.

In addition, student engagement improved when replacing synthetic computer voice (group 1) with human voice (group 2), even though the wording was identical. This result highlights that the way of interacting with AI is just as important as the accuracy and automation of the system. Other studies in clinical diagnosis and radiographic image interpretation have also shown that AI results alone do not necessarily improve human performance. In short: delivery is the decisive factor.

A noteworthy research direction is to leverage the depth and natural nuances of human feedback—timely eye contact, praise, changes in intonation, emphasis on content. This was not fully explored in the study by Giglio et al., indicating the need for multimethod research to capture instructors’ tone and manner of expression.

Large language model (LLM) systems are often optimized to create a positive and agreeable impression. However, early evidence suggests this may reduce cognitive load but at the same time weaken students’ critical thinking ability. This is a factor to consider when integrating AI into medical education. The concept of “desirable difficulty” has been shown to be an important factor in creating durable learning.

Another notable achievement of the study is the development of an objective scale from novice to expert. Parameters such as force, path length, and speed were benchmarked against expert standards, allowing instructors to ask: “Has the student nearly reached proficiency?” instead of merely “How fast did they do it?”. The race toward precision education requires standardized objective data; otherwise, both humans and AI are prone to giving qualitative, subjective feedback that reinforces bias.

Illustration Image

Although AI cannot replace mentorship, emotions, or direct transfer into the operating room, its ability to provide standardized real-time analysis at scale is something humans cannot achieve. Merely optimizing algorithms without attention to real-world implementation will yield only limited benefits. AI can support learning, but fostering surgical expertise still requires a human-centered interface—one that preserves friction, nuance, and motivation for improvement.

Refer to the full post on JAMA Dermatology here