Surgery has always been described as equal parts science and craft. A surgeon spends years learning anatomy from textbooks, but the deeper education happens at the operating table â in the feel of tissue under a blade, the split-second judgment when a vessel bleeds unexpectedly, the quiet recalibration when the body inside the patient does not match the body in the diagram.
Now researchers are asking whether that craft can be transferred to a machine.
Artificial intelligence has already found a foothold in medicine. Algorithms read scans, flag suspicious lesions, draft clinical notes and help triage patients in crowded emergency departments. Surgery, though, has long been treated as a frontier apart. It is physical, improvisational and unforgiving â the kind of work that resists being reduced to a dataset.
That assumption is being tested. Robotic systems have been used in operating rooms for years, but almost all of them are teleoperated: a surgeon sits at a console and the robot faithfully translates hand movements into precise motions inside the body. The robot does not decide anything. The ambition now is different. Researchers are training systems on video of real operations, hoping that machines can learn procedures the way a resident does â by watching, over and over, until patterns emerge.
The technical appeal is obvious. A single surgeon might perform a given operation a few thousand times in a career. A model can, in principle, absorb footage from thousands of surgeons, including the rare complications most clinicians encounter only once or twice. Supporters argue that this could eventually standardize quality, reducing the uncomfortable reality that outcomes often depend on which hospital a patient reaches and which surgeon happens to be on call.
The obstacles are equally obvious. Surgical video is messy, with obscured views, smoke, blood and cameras that move. Anatomy varies from person to person. And unlike a chatbot that can produce a wrong answer harmlessly, a surgical system that misjudges a boundary between tissue types can cause irreversible harm in seconds. There is no undo button in an abdomen.
There is also the question of judgment, which may be the hardest thing to encode. Much of surgical expertise lies not in executing a maneuver but in deciding whether to execute it at all: when to stop, when to convert to a different approach, when the safest move is to close and try another day. Those decisions draw on context that extends well beyond the operative field â the patient’s age, frailty, wishes and the trade-offs they are willing to accept.
Regulators have signaled caution. Approving an autonomous surgical system would require evidence of safety in a domain where failures are catastrophic and rare events matter most. Questions of liability remain unsettled: if a machine errs, responsibility may fall on the surgeon supervising it, the hospital that bought it or the company that built it.
Most people working in the field describe a gradual path rather than a leap. Near-term systems are likelier to assist than replace â highlighting structures a surgeon should avoid, warning of deviations from a standard approach, handling repetitive subtasks such as suturing under close supervision. Autonomy, if it arrives, will likely come procedure by procedure, beginning with the most routine and best-documented operations.
Whether that amounts to learning the art of surgery, or merely imitating its mechanics, is the question the next decade will answer. For patients, the measure will be simpler: fewer complications, and surgeons who are better, not absent. Read More
Leave a Reply