Surgeons accepted 3 in 5 incorrect AI incision suggestions in a controlled experiment, so oversight cannot be assumed to catch AI errors.
- Surgeons accepted incorrect AI overlays in 60.6% of cases (43 of 71), rejecting only 39.4%.
- Correct AI suggestions were accepted 76.9% of the time, and overall behavioral accuracy was 65.4%.
- Design: controlled two-phase experiment, 19 surgeons, 20 robotic cholecystectomy frames, mock (not real) AI overlays.
Where it fits: this is a simulation with static frames and a mock AI, so it cannot show how real intraoperative systems will perform. It does show that human review is not a reliable safety net, and that AI guidance tools should be tested for automation bias before clinical use.
Journal Article by Morais MC, Godbole AA (…) Filicori F et 4 al. in Surg Endosc
© 2026. The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.