Responsible AI in Surgery: Starting with the Right Questions

Before asking what AI can do in the operating room, we should ask what problems are worth solving and what evidence we would need to trust a solution.

Not interested in hype — interested in what holds up under scrutiny.

The conversation about AI in surgery tends to oscillate between uncritical enthusiasm and reflexive skepticism. Neither is productive. What's needed is a disciplined approach to identifying where AI might genuinely help, what evidence would be required to trust it, and what risks come with adoption — including risks we haven't thought of yet.

The most promising applications aren't the flashy ones. They're the boring ones: automating documentation, flagging imaging findings that might be missed in a busy workflow, predicting which patients are at higher risk for complications based on patterns too subtle for human pattern recognition. These are problems worth solving because they're real, measurable, and currently handled poorly.

The dangerous applications are the ones that promise to replace clinical judgment without understanding what clinical judgment actually is. Judgment isn't pattern matching — it's the integration of pattern recognition with context, experience, uncertainty, and values. Any AI system that claims to replicate that without acknowledging its limitations is selling something, not solving something.

I'm interested in AI that makes surgeons better at what they already do, not AI that pretends surgeons aren't necessary. The former is useful. The latter is fantasy.