What happens when medical students rely on AI – and never develop their own judgment? | Simar Bajaj and Joseph Sakran
Two US physicians warn that medical trainees who lean on AI tools such as OpenEvidence before developing their own clinical judgment risk "never-skilling" rather than simply deskilling – losing the capacity to reason that they never actually built. Unlike an experienced doctor who might temporarily lose sharpness from over-reliance on technology, a trainee who never learns independent diagnostic reasoning may be unable to recover it later, which matters because the medical apprenticeship model depends on trainees struggling through incomplete answers to internalise clinical thinking.
The authors note that around two-thirds of US doctors already use OpenEvidence, an AI chatbot for clinicians, to check symptoms, drug interactions and guidelines, and that trainees increasingly use it the same way, but at a far more formative stage of their careers. Where students once had to build differential diagnoses themselves, learning from gaps and mistakes, they can now get a near-complete answer instantly, which can impress supervisors while masking the very deficits training is meant to expose. The piece cites a Nature Medicine study finding such literature-based AI tools can be less accurate than general-purpose chatbots, and questions who will critically supervise AI-assisted reasoning if the next generation of doctors was trained alongside – and dependent on – the machines themselves.
- Medical trainees risk never developing clinical reasoning skills if reliant on AI
- Two-thirds of US doctors already use AI tool OpenEvidence for clinical queries
- Study found such AI tools can be less accurate than general chatbots
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Those who caution against heavy AI reliance argue that clinical judgement is forged through the effortful, sometimes uncomfortable process of forming a differential diagnosis, sitting with uncertainty, and learning from one's own errors under supervision. If trainees lean on AI outputs before that foundation is built, they may become operators who cannot tell when a tool's reasoning is flawed, verify its suggestions, or cope when it is unavailable or wrong. Given that patient safety ultimately rests on a clinician's own judgement in unpredictable, high-stakes moments, they see safeguarding the slow, difficult work of skill formation as essential rather than old-fashioned.
The case against
Those more open to AI integration argue that these tools can strip away rote memorisation and administrative burden, freeing trainees to focus on higher-order skills such as communication, prioritisation and complex decision-making. They point out that medicine has absorbed earlier technological shifts, from decision-support software to search engines, without collapsing clinical competence, and that the answer lies in teaching AI literacy rather than shielding students from tools they will inevitably use throughout their careers. On this view, resisting integration risks producing graduates who are unprepared for the reality of modern practice, where thoughtful human-AI collaboration is the norm.