The Surgeon Training AI on the Decisions That Happen Before the First Incision

By Dr. William Cross — 2025-11-15

Twenty years of surgery has given Dr. William Cross a particular perspective on AI in medicine: the most consequential decisions happen before anyone enters the operating room. Whether a patient's symptoms warrant surgical intervention. Whether the risk-benefit calculus favors operation in an elderly patient with comorbidities. Whether a finding on imaging is the cause of the problem or an incidental passenger. These are the questions AI struggles with most — and the ones where getting it wrong causes the most harm.

The irreversibility of surgery

"Surgery is irreversible," he says. "You can adjust a medication dose. You can stop a chemotherapy regimen. You can't un-operate. That makes the decision to operate — or not to operate — one of the highest-stakes judgment calls in medicine. AI surgical decision support that doesn't deeply understand that asymmetry is dangerous."

Cross joined IXO to work on surgical indication evaluation — reviewing AI outputs on cases where the decision to operate is genuinely uncertain. His annotations focus not just on whether the AI's recommendation was correct, but on the reasoning it used to arrive at it. A correct recommendation for the wrong reasons is, in his view, almost as dangerous as an incorrect one.

Right answer, wrong reasoning

"I had a case where an AI tool recommended observation rather than surgery on an appendicitis presentation. The recommendation was actually defensible for that presentation. But the reasoning was wrong — it had weighted imaging findings that were actually incidental and discounted the clinical examination findings that were most significant. If you deploy that model, it will be correct sometimes for the wrong reasons and incorrect when the presentation shifts slightly."

Postoperative prediction

He also contributes to postoperative complication prediction model evaluation. "Identifying which patients are highest risk for anastomotic leak, for SSI, for readmission — done well, that's enormously valuable. But done poorly, it creates false reassurance. I'd rather be involved in making sure it's done well."

Read The Surgeon Training AI on the Decisions That Happen Before the First Incision on the IXO blog