November 17, 2026
During a laparoscopic cholecystectomy, a surgeon identifies a structure as the right hepatic duct. An artificial intelligence (AI) tool identifies the same structure as the cystic duct.
Charles E. Binkley, MD, FACS, used this hypothetical disagreement to frame two questions: How should the surgeon respond when judgment and AI output diverge, and what should the patient have been told about the tool before surgery?
Dr. Binkley, director of AI Ethics and Quality at Hackensack Meridian Health in New Jersey, addressed those questions during the October 22 ACS AI & Surgery: Leading in Policy Development webinar on transparency and ethics. Tyler J. Loftus, MD, PhD, FACS, from the University of Florida in Gainesville, moderated the discussion. Dr. Binkley presented approaches for considering patient disclosure and building a surgeon–AI team; he did not present them as established ACS policy.
Dr. Binkley distinguished three levels of patient involvement: disclosure that AI will be used, a discussion that seeks the patient’s verbal assent, and written informed consent. He argued that most AI applications need disclosure rather than separate written consent. Which level is appropriate depends in part on what patients would find material to a decision about their care.
Dr. Binkley offered an ambient AI scribe as an example outside the OR. Before recording a visit to generate a clinical note, he said, the clinician should explain what the tool will do and seek the patient’s assent. That conversation differs from simply informing a patient that AI may be used somewhere in their care. His point was that meaningful transparency depends on the role of the specific tool, rather than on a blanket rule that all AI use calls for the same consent process. Several factors inform that judgment, such as the degree of AI autonomy, whether the tool adds risk compared with existing care, the strength of evidence for its benefits, the availability of reasonable alternatives, and how patient data will be used, Dr. Binkley said.
The distinction between a technical choice and a treatment choice also matters. He illustrated that distinction with heart valve surgery. Selecting the size of a valve ring is a technical decision, whereas choosing between a mechanical and a tissue valve involves trade-offs that matter to the patient. No single factor, including autonomy, determines the appropriate conversation.
As an example of disclosure, Dr. Binkley described a tool that predicts the likelihood of a blood transfusion during surgery. For a tool used to assess potentially positive margins during lumpectomy, he proposed discussing its use with the patient and documenting that conversation, without necessarily obtaining separate written informed consent.
The surgeon could explain how the tool’s findings would inform the operation and discuss the possibility of re-excision. He suggested focusing that discussion on how information from the tool will affect care, rather than on the technical details of how the model was developed. A more experimental tool warrants greater consideration of written informed consent.
Dr. Binkley urged health systems to consider these categories before deploying a tool, with input from surgeons, rather than leaving each surgeon to make the determination independently for each case. A consistent approach across users of a tool, he said, could also avoid unnecessary variation in practice. He distinguished clinical consent from consent for a research protocol. He also said that certain secondary uses of identifiable patient data would require separate authorization rather than inclusion in surgical consent.
The consent approach may change as evidence and practice change. During the discussion, Dr. Binkley returned to the lumpectomy example and considered what would happen if evidence eventually established that an AI tool improved care and it became a commonly accepted way to assess margins. In that circumstance, he suggested, its use might come to be regarded as a technical decision. If declining the tool were feasible but increased risk, Dr. Binkley suggested considering informed refusal. He framed these as possibilities to assess as experience with the tool grows, not as conclusions about current practice.
Returning to the cholecystectomy example, Dr. Binkley considered three proposed responses to disagreement: deferring to a highly accurate AI tool, always favoring the surgeon’s judgment, or seeking a second opinion. He said none was an ideal way to resolve intraoperative disagreement. A second surgeon, for instance, might not be available when needed. Disagreement is not the only concern. A surgeon and AI might reinforce the same incorrect conclusion, or the surgeon might dismiss AI output without considering it. Dr. Binkley described these as false confirmation and silent dismissal.
His proposed response draws on principles surgeons already use to build effective OR teams. The surgeon remains the designated leader, with defined tasks for the AI tool. Surgeons should understand the tool’s strengths and predictable failures, as well as the limits of their own judgment. Teams should consider simulation and credentialing before bringing a tool into the OR. During care, they should acknowledge AI output and apply clinical judgment rather than silently accepting or ignoring a recommendation. Recording these interactions, particularly disagreements, would give teams material for quality review and help them identify where either the tool or the team needs to improve.
Dr. Binkley compared this process with the way surgical teams learn from cases at morbidity and mortality conferences. An interaction record could help identify how the tool failed, how surgeons incorporated its output into decisions, and where their combined work needs attention. He acknowledged concern that recording disagreements might create liability exposure, but argued that failing to preserve them would limit opportunities to evaluate and improve care. He also emphasized assessing the performance of the combined team. A tool’s accuracy on its own does not establish that surgeons using it will make better decisions.
Dr. Binkley also recommended that surgeons form an initial judgment before seeing AI output when possible, especially in training. He expressed concern that seeing an AI assessment at the same time could encourage uncritical reliance and weaken clinical reasoning. He called for an environment in which team members can acknowledge and examine differences with AI output and use those experiences to learn. He proposed assessing how the surgeon and tool perform together, rather than treating the two as competitors.
During the discussion, Dr. Loftus and Dr. Binkley emphasized the need for education that helps practicing surgeons use AI tools effectively. Dr. Binkley urged the ACS to develop practical guidance on questions such as consent, training, and surgeon–AI collaboration. Dr. Loftus pointed to surgeons’ experience leading OR teams as a starting point for that work.
The next webinar in the ACS AI & Surgery: Leading in Policy Development series, planned for early 2027, is expected to continue the discussion of how surgeons can help shape the use of AI in patient care.