Blog & news

Who answers for the AI at 2 AM? Inside The State of AI in Emergency Medicine 2027

I asked 41 leaders in emergency medicine, health systems and AI five hard questions and printed their answers. Here is where they split, and what your ER should settle before go-live.

Who answers for the AI at 2 AM? The State of AI in Emergency Medicine 2027, with photos of the contributors and the editor

I still work ER shifts. At 2 AM, when a tool flags a patient or drafts a note, the software doesn't sign the chart. I do. That's the question I took to 41 leaders in emergency medicine, health systems and AI for The State of AI in Emergency Medicine 2027, which I published today.

Every contributor answered the same five questions. The result is 145 answers and 27 predictions.

They didn't agree, and I think that's the most useful part of the report.

Accountability splits three ways

When ED AI gets it wrong, who answers? I coded 29 answers to that question. Fifteen said the responsibility has to be shared by the clinician, the health system and the vendor. Nine said it sits with the clinician who signs. Five said to fix the process before arguing about blame.

When ED AI gets it wrong, who answers? Of 29 coded answers, 15 said shared responsibility, 9 said the clinician who signs, and 5 said fix the process first.

Patrick R. Coyle, who certified financial statements as a CFO, put the risk plainly: “Certification without the ability to inspect is not accountability.” David C. Rhew, MD, of Microsoft, sees it differently. “Human accountability does not require reviewing every AI action,” he wrote. In his view, people answer for the limits, the monitoring and the moment a human steps in.

Ambient scribes: time back, or burden moved?

Two physicians on ambient AI scribes: Jay Woody, MD says 100% of his charts are done when his shift ends; Robert Hoyt, MD says there is no evidence scribes are huge time-savers, and they are expensive.

Jay Woody, MD, finishes every chart before he leaves. Robert Hoyt, MD, reads the evidence and doesn't see large time savings, at a real cost. Prashant Mahajan, MD, chair of emergency medicine at the University of Michigan, gave the fairest summary I've seen: “Clinicians trade typing for reviewing.” If you're buying a scribe this year, measure whether the work disappeared or simply moved.

Triage, patients and the year ahead

On triage, Amy Toopes of the Emergency Nurses Association drew the line clearly: “AI should strengthen clinical judgment, not replace it.” On what to tell patients, contributors came closest to agreeing. Miguel Villagra, MD, put it in four words: “Disclosure costs us one sentence.”

For 2027, Ryan A. Stanton, MD, President of the American College of Emergency Physicians, says get involved and don't sit on the sidelines. Jesse M. Ehrenfeld, MD, MPH, Past President of the American Medical Association, says put AI into one defined care pathway, measure whether patients benefit, and refuse to treat a good demo as proof.

What you can do on the report page

  • Where do you stand? Vote on each question and compare your answer with the panel. No name, email or signup.
  • Debate mode. Pick any two contributors and read their answers side by side.
  • Ask the report. Type a question and see the contributor answers that match it.
  • Stress-test my go-live. Describe your AI rollout and see what contributors said about plans like yours.
  • A seven-point go-live checklist, 27 bold predictions I'll grade in public in January 2028, and the free PDF.

Read the full report free

Thank you to all 41 contributors.

Bring this conversation to your team

I've turned the report into a keynote and workshop, “Who answers for the AI at 2 AM?” Watch the speaker reel or check availability.

Contributors speak only for themselves. This post is educational and is not medical, legal or regulatory advice.