Skip to Main Content
In Depth

Our AI Partner

4 Minute Read

Hi Everyone:

Last week, while pre-charting for clinic, I found Epic serving up AI-generated patient summaries:

Ms. Smith is a 79-year-old woman with bronchiectasis…When seen on June 2, 2026, she reported cough with clear sputum…A chest CT showed bronchiectasis, unchanged from previously…She was advised to use an acapella...

The summaries unearthed facts I would have overlooked, like the patient who missed clinic because he couldn’t afford transportation—a detail noted by a social worker. AI didn’t save me time; I still sifted through the notes, messages, and test results, but the summaries proved accurate, thorough, and concise.

The new Epic feature made me wonder how residents should use these tools and, specifically, how AI will impact your education.

AI is becoming indispensable, for senior physicians and trainees alike. It answers questions fast. If given the right prompts, it constructs comprehensive, prioritized differentials and suggests diagnoses we might have missed. Some claim AI can outperform physicians, at least for certain cognitive tasks. The accuracy and value of AI grows every day.

But AI poses a hazard to medical training. It will be tempting to seek rapid answers from OpenEvidence (or ChatGPT) without exploring and absorbing nuanced text. It will be enticing to rely on computer-generated summaries instead of filtering the EMR. And it will be tantalizing to use these tools to write notes and discharge summaries. Like the Homeric Sirens, AI may prove irresistible.

But there are no shortcuts to learning medicine. Just as we need to pump iron to build muscle, we need to work hard to become doctors. We can’t learn acid-base if OpenEvidence interprets our blood gases, nor can we learn ID if Doximity selects our antibiotics. If we don’t truly learn medicine, we will be fooled by LLM-generated hallucinations.

And do we really want AI writing our notes? How well can we know our patients if we don’t decide which details to include or exclude? How can we understand our patients’ illnesses if we don’t review their data ourselves? Can we gain experience without experience?

To learn medicine, we have to create AI-free zones. In clinic, we will continue to require interns to write their own notes before graduating to ambient scribes. On rounds, we will work through ABGs and antibiotic choices without leaning on OpenEvidence for answers. We will hold ourselves accountable for knowing our patients’ stories, including the details computer-generated summaries may miss. We will create our own human-generated differentials and treatment plans.

AI can improve how we practice medicine, but to fully realize its benefits, we have to become—and remain—sophisticated users. We must commit to hours of studying and probing our knowledge. We must untangle webs of data and solve diagnostic dilemmas without outsourcing the work. Used appropriately, AI will be our partner—augmenting, but never replacing, the knowledge and skills of human physicians. There are no shortcuts.

Enjoy your Sunday, everyone. Heide and I are going to the New York Botanical Gardens, followed by dinner on Arthur Avenue. Tomorrow I’ll be starting a week in the Stepdown Unit.

Mark

P.S. What I’m reading:

Biking on the Farmington CanalCredit: Mark D. Siegel, MD

Article outro

Author

Mark David Siegel, MD
Professor of Medicine (Pulmonary)

Media Contact

For media inquiries, please contact us.

Explore More

Featured in this article