and it would primarily be used with routine surgeries and would default to human intervention under uncertainty. Most of these surgeries are basically identical and the AI could improve by sharing data with all surgeries performed around the world. Not to mention how bad humans are at surgery, especially when fatigued.
AI is the only way to fix healthcare, a doctor/surgeon can work effectively only a few hours a day and takes nearly 30 years to educate from childhood. An AI could do surgery and diagnose patients nearly 24/7, 7 days a week.
Surgical robots -- at least the ones that actually exist and work today -- are designed more like traditional safety-critical robotics platforms and less like a one-off case study for a nice visualization/table for an end-to-end deep RL paper. There is no "the model".