Medical Students Are Using AI Before They've Learned to Think Like Doctors
A popular AI tool already used by two-thirds of US doctors has reached trainees who are still forming basic clinical judgment. Experts warn the risk isn't losing a skill. It's never building one.
Key points
- Roughly two-thirds of US doctors actively use OpenEvidence, an AI chatbot built for clinicians, to look up symptoms, drug interactions and treatment guidelines.
- Medical educators now worry that trainees using AI before developing clinical judgment may never build that reasoning ability at all.
- The concern has a name: "never-skilling," which is distinct from "deskilling," the already-documented risk of trained professionals forgetting skills they once had.
- A doctor who never learned to reason independently has no baseline to recover from.
A tool that helps experienced doctors work faster may be doing something quite different when a first-year medical student picks it up.
OpenEvidence is an AI chatbot built specifically for clinicians, similar in style to ChatGPT. It fields questions about puzzling symptoms, drug dosages and clinical guidelines, returning answers in seconds, anchored in current research. The Guardian first reported the growing concern among medical educators that this kind of tool, used at the wrong stage of training, could quietly hollow out the next generation of doctors.
What is "never-skilling" and why does it matter?
It is the risk that a trainee who leans on AI before developing core judgment never actually develops it. "Deskilling" describes what happens to an expert who stops practising a skill they once had: that person has a foundation, and recovery is possible. A student who skips the foundation is in a different position.
Clinical reasoning, the mental process of weighing symptoms, arriving at diagnoses and making judgment calls under uncertainty, takes years of deliberate practice. It isn't learned by reading an answer. It is built by working through a problem, being wrong, and understanding why. When a tool produces the right answer in three seconds, a trainee can pass the test without doing the thinking.
Our 10 August story on AI and peer review touched on a related problem: how AI can short-circuit the analytical work that makes a reader, or a clinician, actually understand what they are looking at.
Who is affected right now?
Medical students and post-graduate residents and fellows are at greatest risk. These are people at precisely the stage where clinical reasoning is supposed to be forming.
About two-thirds of practising US doctors already use OpenEvidence regularly, which means trainees are picking up a tool that feels professionally normal. The concern isn't that the tool is wrong. Often it isn't. Correct answers, handed over without the work of reasoning, may produce doctors who are fluent at asking AI but poor at thinking without it.
What happens next?
No regulatory body has yet set formal rules on AI use in medical training. That gap matters. Medical schools are largely navigating this individually, without shared standards for when or how trainees should use these tools.
When AI is unavailable or simply wrong about an unusual case, the doctor in the room needs to reason through it alone. Educators calling attention to this are not arguing against AI in medicine. They are arguing for sequencing: build the judgment first, then introduce the tool.
Common questions
Does this mean AI is bad for medical training?
Not necessarily. The concern is about timing, not the technology itself. An experienced clinician using AI to double-check their own reasoning is a different act from a trainee using it to skip that step entirely.
Could medical schools simply ban AI tools for students?
Bans are difficult to enforce and may not be the right answer. Educators are more focused on designing training that requires students to show their reasoning process, not just arrive at a correct answer.
