“Explain the pathophysiology of paracentesis-induced circulatory dysfunction.” I type this prompt into Doximity Ask as I review for my internal medicine exam. Within seconds, I get a detailed explanation of the challenging concept, with several appropriate peer-reviewed articles cited. Still a little confused, I ask a follow-up question, and I get a slightly different explanation that better fits my learning style. Then, I ask another question: “When to give albumin after paracentesis?” In addition to journal articles, I get verified responses from hepatologists using Doximity’s PeerCheck. This is not the future anymore. This is the reality of medical education today. This rapid integration of AI has created a “calculator moment” for the next generation of medical students, offering a powerful tool for preclinical learning.
When calculators entered the classroom, teachers worried students would lose critical math skills. For the most part, we are not any less functional in our daily lives without the same expertise in long division. But, as a medical student, I find myself among my peers straddling the line between traditional clinical intuition and the tempting efficiency of AI.
In the preclinical curriculum, the utility of AI is difficult to dispute. It reframes a concept through the lens of a clinical scenario you already understand and adjusts its explanation when you push back. It is, in the truest sense, a 24/7 personalized tutor. For exam studying, this translates into something meaningful, less rote memorization with Anki and more genuine comprehension. When I understand the underlying physiology, I retain it, apply it, and build on it. AI accelerates this process for me in ways that traditional resources alone cannot.
While AI-driven tutoring can simplify complex concepts into personalized analogies, the clinical transition during third year introduces “unwritten rules” and a looming fear that outsourcing differential diagnoses could erode the very critical thinking skills we came to medical school to develop. When traversing my clinical rotations, there were no explicit rules surrounding AI use. The general understanding is that students should not use AI for documentation or clinical reasoning. But, the temptation is real. When a 40-year-old woman presents with fatigue, diffuse joint pain, and a mild rash, the clinical differential is broad and the time to think is short. Doximity Ask can generate a reasonable differential in seconds. I don’t use it for this, but I think about the fact that some of my peers might, and I wonder what that means for medical education. As a student, it is important to reason through a differential and develop clinical intuition.
This is the crux of the AI dilemma in clinical medicine. Ambient AI scribes that reduce documentation burden are increasingly welcomed, even by attending physicians who once resisted them. They handle the administrative exhaust of modern medicine. But, that feels different from asking an AI what to think about a patient. Documentation is what you record after thinking. A differential diagnosis is the thinking itself.
If some students are using AI to preround, synthesize patient data, and polish oral presentations and notes while others are not, are we being evaluated on an even playing field? Without clear policies, the answer is unclear. A student who uses AI to construct a more polished assessment and plan may receive better evaluations, though not because they understood more. The absence of explicit guidance creates an uneven playing field, and the students most likely to use every available advantage are not always the ones most likely to flag it.
There is a concept in educational psychology called desirable difficulty. This is the idea that learning is strengthened when it involves a degree of effortful retrieval, rather than passive reception. Struggling to generate a differential diagnosis, even imperfectly, builds the skills that make you a better diagnostician. Being handed a differential diagnosis by an AI bot as a medical student without that existing foundation of knowledge is a shortcut that undermines our learning.
With all this said, I would push back on any binary framing for or against AI in medical education. The students who use AI to prepopulate a note template and never think critically about the patient are outsourcing their education. But the students who use AI to explore the pharmacology of an unfamiliar medication or understand the evidence behind a treatment they just observed in rounds are doing something different. Those students are learning deeper and more efficiently with AI as a tool.
AI etiquette in medical education is sparse. Students need clear frameworks to distinguish between AI as a learning tool and AI as a clinical decision surrogate. We need to ask, seriously, what we want physicians to be able to do without AI and protect those skills accordingly.
When do you think medical students should use AI? Share in the comments.
