Emergency Medicine in the Age of AI: One Physician's View from the Front Line

By Dr. Michael Grant — 2025-11-10

Emergency medicine is practiced under conditions that most AI evaluation doesn't simulate: incomplete information, time pressure, simultaneous competing priorities, and the near-constant possibility that the obvious diagnosis is wrong. Dr. Michael Grant has worked in a Level I trauma center for eleven years. He joined IXO because he saw AI entering his domain without understanding its most fundamental operating conditions.

The reality of the ED

"The ED isn't a controlled environment," he says bluntly. "A patient arrives with three symptoms, an unclear history, and a family member who speaks limited English. You make decisions with what you have, fast, knowing you might be wrong and watching for the signs that you are. AI trained on clean retrospective cases doesn't know how to do that."

His IXO work focuses on two areas. Triage evaluation — testing AI triage tools against ED triage standards, specifically looking for undertriage and overtriage. And diagnostic reasoning under uncertainty — evaluating how AI handles cases where the presenting symptoms are consistent with both benign and life-threatening diagnoses.

The aortic dissection problem

"The aortic dissection problem is one I care about a lot. Classic presentation is chest pain and pulse differential. But most aortic dissections I've seen presented atypically — back pain, abdominal pain, neurological symptoms. AI trained on classic presentations will miss the atypical ones. Those patients die. Flagging that gap in training data is direct patient safety work."

Finding balance

Grant does 6-8 hours of IXO work weekly, mainly between shifts. "The pace of emergency medicine means my brain needs different types of work. Careful, deliberate annotation is genuinely restorative in a way that the ED isn't."

Read Emergency Medicine in the Age of AI: One Physician's View from the Front Line on the IXO blog