The Stethoscope Didn’t Replace the Physical Exam. Will AI Replace the Medical Educator?

Ambient AI is transforming how clinicians document care. But as hospitals rush to adopt it, a critical question becomes: what happens when medical students and residents hand their most formative learning tasks to an AI?

Every generation of physicians has navigated a new technology that changed how medicine is practiced. The stethoscope, the X-ray, the electronic health record — each one raised questions about what clinicians might stop doing themselves, and what might be lost in the process.

Ambient AI is the latest in that line. These tools, which listen to patient-clinician conversations and automatically generate clinical notes, are already in use across thousands of clinical settings. Studies show they reduce documentation time, cut after-hours charting, and meaningfully improve physician satisfaction and burnout. Patients report feeling more seen when their doctor is looking at them instead of a screen. The benefits are real, but there are many questions that we have not yet answered. I wrote about the changes even across my own career.

The potential for a wide-ranging impact of this technology is not yet clear. A group of physicians from Suki, the University of Miami Miller School of Medicine, and Dawson Med recently raised a critical question that medical educators must grapple with urgently in a Viewpoint published by JMIR.

Why learning is different from care, even when both occur in the same room with the same people

Seeing a patient and writing a clinical note is not just busywork. For a medical student or resident, it is one of the primary ways they learn to think. As the saying goes, to write clearly, you have to think clearly.

The act of taking a history and documenting the concerns of the patient forces trainees to organize what they observed, identify what they don’t yet understand, build a differential diagnosis, and commit to a plan. It is a thinking exercise wrapped in a writing task. Educational researchers call this “cognitive apprenticeship” — the idea that expertise develops through the deliberate, structured act of doing.

When that process is handed to an AI, the notes may be better — but what happens to the learning? Ambient AI, used uncritically in training, could turn that active process into a passive one. A student who watches AI draft the assessment and plan misses the step where synthesis happens. It could also help the learning get better because the trainees can now focus on the relationship instead of worrying about writing down all the details. Which is it? Does it change based on when in the learning continuum the technology is introduced?

These concerns run across the full training continuum. In undergraduate medical education, students are still building baseline skills in history-taking, communication, and documentation. In residency, those skills are being refined under real stakes. In both settings, thoughtful incorporation of ambient AI is necessary to maximize the benefit and minimize the potential for unintended consequences before the technology becomes an entrenched habit.

These are unanswered questions today.

What We Don’t Know — and Why That Matters

Here is what makes this moment genuinely urgent: almost none of this has been studied.

There are no major studies examining ambient AI’s impact on medical students’ acquisition of clinical documentation or reasoning skills. There are small pilot studies among residents suggesting improved satisfaction and reduced burnout — but no studies tracking whether residents who use ambient AI develop stronger or weaker patient relationships, or whether their clinical reasoning improves or atrophies over time.

In a new Viewpoint published by JMIR, The Promise of Ambient AI Technology in Medical Education: Opportunities and Guardrails, we identify several specific gaps the research community needs to close:

• Does ambient AI use during training improve or impair the development of clinical reasoning?
• What happens to note quality and synthesis skills when documentation is automated early in training?
• How does reduced documentation burden affect time spent in direct patient care?
• Are there specialty-specific effects — for example, in pediatrics, where developmental context and caregiver relationships add layers of communication complexity?
• How do equity considerations play out, given that speech recognition tools are known to perform worse for some accents and dialects?

We call for rigorous, multi-center study designs — randomized crossover trials, step-wedge designs — that can actually isolate the effect of ambient AI on learning outcomes. It’s a high bar, but one the field should aspire to meet before widespread deployment in training environments becomes irreversible.

In the Viewpoint, we also discussed a set of guardrails worth considering because this technology can make medicine better for clinicians and for patients, but everyone, from developers to educators, carries a responsibility to ensure we understand the impact on the next generation of physicians.

That means engaging with medical schools and residency programs as genuine research partners, not just distribution channels. It means welcoming independent, vendor-neutral research into how these tools affect learning — even when the findings are complicated. It means building the kinds of study designs that can actually answer hard questions about cognitive offloading and skill atrophy, not just satisfaction scores and documentation time.

It also means acknowledging that a tool that is unambiguously good for an attending physician may need to be introduced differently — or introduced later — for a third-year medical student. Context matters. The stage of development matters. The act of documentation is not the same thing at every point in a clinical career.

That is not a caution against ambient AI. It is a challenge to the ambient AI industry to do the work that will earn medicine’s trust in these tools for the long term. I think we should take it seriously.

Reference

JMIR Viewpoint: Shafazand S, Bowers U, Jayaraman S. “The Promise of Ambient AI Technology in Medical Education: Opportunities and Guardrails.” Viewpoint (published).

Medcity article: A Transformation Doctors Didn’t Expect, But Desperately Need

Sudha Jayaraman, MD, MSc, is Medical Director of Clinical Strategy and Research at Suki and an Adjunct Professor at the University of Utah Departments of Surgery and Biomedical Engineering. The views expressed here are her own.