Yes, AI is changing medicine – and AI in medical education means the curriculum has to change with it. A viewpoint published today by Dr. James W. Schroeder in Chief Healthcare Executive argues that medical schools should teach less memorization and more empathy. As a community medicine researcher who studies how medical students learn and behave, I think he is right – but the curriculum needs more than empathy alone. It needs AI literacy built on evidence, not hype.
The argument that started this piece
Schroeder, who is Chief Academic Officer at the Hospital for Special Surgery, makes a straightforward case. Eight in ten doctors already use AI for tasks like note-taking and catching up on research. The technology brings real benefits, but it has limits: biases, shortcomings, moments where the tool gets it wrong. So medical educators should do two things at once – double down on the human elements of care like compassion, communication and bedside manner, and teach students how to use AI while recognising its flaws.
His point about empathy carrying clinical value is one I would defend with the literature. Over half of patients say they want compassion and good bedside manner from their doctors, and there is real evidence that physician empathy is associated with better outcomes in some conditions. No language model can sit with a family receiving a life-altering diagnosis. That part of the job is not being automated.
Read Dr. Schroeder’s full viewpoint on Chief Healthcare Executive.
What my own research adds to this debate
I study how medical students think, learn and make decisions – from a mindfulness study (n=344) to a self-medication survey (n=699) to the psychometric validation of the PHQ-9 and DASS-21 (n=602). Across all of them, one pattern keeps repeating: knowing is not doing.
In my self-medication study, medical students knew the risks better than anyone – yet self-medicated just as often as everyone else. The knowledge-behavior gap was real and measurable. I wrote about it in detail here. The same principle applies to AI in medical education. Teaching students about AI – listing its risks in a lecture – will not make them careful users of it. Training has to change behavior: hands-on practice, structured reflection, supervised use in clinical settings. If we learned anything from decades of health-behavior research, it is that a slide on “AI hallucinations” changes nothing by itself.
What the curriculum should actually change
Here is what I would argue for, as a researcher rather than a clinician:
1. AI literacy as a core competency, not a guest lecture
Students need to understand what AI tools do well (pattern recognition, summarisation, documentation) and where they fail (confabulation, encoded bias, overconfident errors). This belongs in the curriculum the way statistics belongs in the curriculum – as foundational method, not enrichment.
2. Protected practice with human supervision
An editorial in the Future of Medical Education Journal maps the near-term opportunities – personalised practice at scale, workflow augmentation, high-fidelity simulation – but warns that most studies are small, single-site and brief, rarely measuring downstream competence or safety. The full editorial (PDF) also flags hallucinations, privacy concerns and threats to assessment integrity. The lesson for educators: adopt AI in teaching, but study it like any intervention – with rigorous evaluation before scaling.
3. Teach the human skills AI cannot replace – on purpose
Schroeder is right that empathy, communication and navigating vulnerable moments deserve more curricular weight. My one addition: teach these as assessable competencies, not as vague “professionalism” checkboxes. If it is not assessed, it is not taught.
4. Assessment redesign before the integrity crisis
The same editorial warns about assessment integrity – and rightly so. Exams built around recall are exactly the kind AI answers best. If we keep testing what a chatbot can do, we will measure the chatbot, not the student. Curriculum reform that keeps 20th-century assessments will fail quietly.
Frequently asked questions
Will AI replace medical education?
No. AI can scale practice, feedback and simulation, but it cannot replicate supervised clinical judgment, empathy or professional responsibility. It changes what must be taught, not the need for teaching.
Should medical students use AI in their studies?
With guidance, yes. The risk is not use but uncritical use. Students who learn to verify AI outputs, spot bias and know when not to trust the tool will be safer doctors than those who were simply told to avoid it.
What does the evidence actually say about AI in medical education?
Short-term knowledge and skills gains look promising but the evidence base is fragile – small, single-site studies with brief follow-up. We need the same standards for AI interventions as for any educational method: controlled designs, long-term outcomes, attention to equity and safety.
My bottom line as a researcher
Medical education has always lagged a decade behind medical practice. This time we cannot afford that gap, because the tools are already in students’ pockets. The task is not to add a module on AI. It is to re-examine what a doctor is for – judgment under uncertainty, compassion in vulnerable moments, responsibility for decisions – and to build a curriculum around that, with AI literacy woven through it. Less memorization, more empathy, yes. But also more rigor: we should study our AI-era reforms the way we would study a drug, and publish what we find.
Dr Taimoor Asghar is a physician and community medicine researcher studying mental health in medical students. His work spans self-medication behavior, mindfulness, and the psychometric validation of mental health instruments. Connect on LinkedIn or find his publications via his ORCID and Google Scholar profiles.