Our Take
AI isn't breaking medical education; it's revealing that the system was already broken and designed for a different era.
Why it matters
Physicians in training face $300,000+ debt, curricula that lag medical knowledge by years, and selection processes that optimize for test scores instead of empathy or teamwork. As AI tools arrive in residencies and patient care today, the misalignment between what training teaches and what practice demands becomes impossible to ignore.
Do this week
Attending physicians: model responsible AI use on rounds this week (stress-testing outputs, tracing references, naming hallucinations) so residents learn judgment, not abstinence.
The AI mirror shows an older problem
Medical education is in crisis, but not because of artificial intelligence. The crisis exists because the system hasn't kept pace with the work it's supposed to prepare physicians to do. LLMs now pass the USMLE, a fact that sounds impressive until you ask what the exam actually measures. A patient doesn't walk into an exam room with a formatted clinical vignette and five answer choices. They arrive with decades of life history, incomplete information, and communication shaped by everything before that appointment.
Early research shows the gap in uncomfortable detail. When AI performs well on isolated clinical scenarios, accuracy drops significantly once a real human with messy communication enters the loop. Context, which practitioners learn through years of human interaction, cannot be optimized away through test performance. The real problem isn't that AI exists in residencies and patient care—it already does, in the pockets of residents on rounds and in the hands of patients in exam rooms. The problem is that medical education was never designed to produce physicians who could use such tools wisely.
Structural debt didn't come from LLMs
Medical literature doubles roughly every 73 days, yet curricula haven't kept pace. Student selection still prioritizes individual academic achievement when medicine is fundamentally a team sport. Graduates face debt loads of $300,000 to $400,000, which shapes specialty choices and geography in ways unrelated to where physicians are actually needed. Primary care and underserved regions suffer while debt drives graduates toward higher-paying specialties.
These failures predate any language model by decades. They exist because medical education was built in a different era and has resisted structural change through multiple attempted disruptions: duty hour restrictions, USMLE pass/fail Step 1, virtual care adoption. AI doesn't create these problems. It simply makes clear that the return on investment in time, money, and years of a physician's life is no longer being maximized by a system designed for a world that no longer exists.
What responsible integration actually requires
Pushing AI out of training environments is neither possible nor desirable. Teaching physicians to practice without AI tools would be like training pilots without flight simulators. The goal instead is to build intentional integration into curricula, not let it happen around the system.
Faculty must model what responsible AI use looks like: stress-testing outputs, tracing references back to primary literature, recognizing when a tool is hallucinating with confidence. Clinical reasoning doesn't disappear in an AI-enabled world. It becomes more important, because someone must decide whether to trust the machine.
A medical student today can use LLMs to simulate thousands of additional patient encounters and clinical scenarios that previous generations never accessed. This is not deskilling—it's an equalizer, if built intentionally rather than left to chance. The next generation of physicians who will thrive aren't the ones who memorized the most or scored highest. They're the ones who learn continuously, use tools critically, and do what no model has figured out: walk into a room, read a face, and make a person feel seen.