AI is entering surgical practice, but its authority remains conditional. The central question is no longer whether surgeons use AI, but when the output is trustworthy, actionable, and safe enough to influence a high-stakes decision.
Audience: Practising Surgeons
Countries: USA, UK, Canada, France, Italy, Germany
SGID: 8794665
– Hero finding
Quick Read — Key Findings
Explore how surgeons balance algorithmic insight with clinical accountability, patient context, and the right to override.
// The backstory
The early promise of surgical AI focused on pattern recognition, planning, and efficiency. The harder question begins when the tool moves closer to decisions about whether to operate, which approach to take, or how to manage a high-risk patient.
WHO guidance on AI for health places human autonomy, transparency, safety, and accountability at the centre of responsible use. FDA, Health Canada, and MHRA transparency principles similarly emphasize the performance of the human-AI team, not the algorithm in isolation.
Sources: World Health Organization, Ethics and Governance of Artificial Intelligence for Health; FDA, Health Canada, and MHRA, Transparency for Machine Learning-Enabled Medical Devices.
We asked practising surgeons across the United States, United Kingdom, Canada, France, Italy, and Germany how AI enters their decisions, where they trust it, and what happens when the machine and the surgeon disagree.
Finding 1
AI enters surgical decision-making at least occasionally for 63.2% of respondents, yet only 26.4% describe routine or frequent use. This is not a simple adoption gap. It suggests that many surgeons are testing AI selectively, usually when a case generates uncertainty, additional information, or a need to compare options.
What the pattern means: AI is behaving more like an occasional second lens than a stable member of the surgical workflow. Selective use can be appropriate, but it also means experience, governance, and confidence develop unevenly across specialties and institutions.
Finding 2
Complex and multimorbid cases attract the strongest reported influence, while treatment-pathway recommendations are the most commonly used output. These are precisely the decisions where context, competing risks, and patient goals are hardest to compress into an algorithm.
What may be behind it: Surgeons may turn to AI when conventional pathways become less decisive. That creates a paradox: the tool is most attractive when complexity is highest, but complexity is also where generalizability and explainability matter most.
Finding 3
37.1% say trust varies substantially case by case, while 22.6% generally do not trust AI outputs. The data does not describe blanket rejection. It describes a professional instinct to revalidate the tool each time the clinical context changes.
Why this matters: A model can perform well overall and still be unhelpful in an individual case. Surgical adoption therefore depends not only on accuracy, but on whether the surgeon can understand the limits, intended use, and consequences of the output.
Context: WHO guidance emphasizes autonomy, transparency, accountability, and safety in health AI. FDA, Health Canada, and MHRA transparency principles focus on the performance of the human-AI team and clear information for users.
Finding 4
More than half rely on clinical judgement when recommendations conflict. Others seek tests, re-evaluate, or discuss the case within a multidisciplinary team. Only 2.4% say they follow AI when uncertainty is high.
The deeper insight: The most common response is not rejection of technology. It is escalation of scrutiny. That is a healthy safety behaviour, but it requires time, access to additional diagnostics, and clear documentation pathways.
Finding 5
64.7% encounter outputs that seem accurate but are not clinically actionable at least sometimes. This separates information quality from decision usefulness.
What may explain the gap: The output may be too generic, disconnected from the operative pathway, poorly timed, or unable to account for patient preference and institutional capability. The next phase of surgical AI may depend less on producing more signals and more on producing signals that fit a defensible decision pathway.
Finding 6
54.1% identify automation bias as the greater long-term risk, and another 27.0% consider over- and under-use equally significant. This concern sits alongside limited institutional integration, suggesting surgeons are thinking ahead of current workflow maturity.
What the pattern means: Adoption cannot be measured only by how often a tool is used. Safe maturity also requires the ability to override it, recognize when it is outside scope, and preserve independent reasoning even when institutional pathways begin to normalize AI-supported choices.
Practice implications
Keep the decision boundary visible
Teams should define where AI is advisory, where independent verification is mandatory, and which decisions remain outside the tool’s validated scope.
Document disagreement, not just agreement.
When surgeons override or question an output, the reason can reveal important failure modes and improve local governance.
Measure actionability and decision impact
Accuracy alone does not show whether the tool improves care. Evaluation should include whether the output changes management appropriately, reduces uncertainty, or creates extra diagnostic and cognitive burden.
//People Also Ask
Direct answers to the questions professionals most often search around this topic.
Direct answers to the questions healthcare professionals are most likely to ask about these findings.
