The rise of AI in medicine is both exhilarating and deeply unsettling, especially when you consider its impact on the next generation of doctors. Personally, I think we’re standing at a crossroads where technology could either elevate medical training or inadvertently undermine it. What makes this particularly fascinating is how AI tools like OpenEvidence are reshaping the very foundation of clinical reasoning—a skill that has traditionally been honed through years of struggle, failure, and hands-on experience.
The Illusion of Competence
One thing that immediately stands out is how AI can create the illusion of competence. Trainees can now bypass the painstaking process of diagnosing a patient by simply querying an AI. On the surface, this might seem efficient—and it is. But what many people don’t realize is that this efficiency comes at a cost. The struggle to diagnose, to wrestle with uncertainty, and to learn from mistakes is where true clinical judgment is forged. If you take a step back and think about it, AI isn’t just answering questions; it’s potentially short-circuiting the development of critical thinking skills.
From my perspective, this isn’t just about deskilling—it’s about never-skilling. A doctor who loses a skill might regain it with practice, but a trainee who never develops it in the first place? That’s a far more daunting challenge. This raises a deeper question: Are we creating a generation of doctors who are more supervisors of AI than independent thinkers?
The Apprenticeship Paradox
Medical training is, at its core, an apprenticeship. Students learn by doing, by making mistakes, and by being corrected. But AI is disrupting this dynamic. A detail that I find especially interesting is how trainees are now using AI to appear more competent than they actually are. They can present a flawless list of diagnoses, impress their supervisors, and move on—all without truly understanding the reasoning behind those diagnoses. What this really suggests is that we’re not just training doctors; we’re training performers.
This isn’t to say AI is inherently bad. In fact, when used thoughtfully, it can be a powerful tool. But the timing matters. Trainees should be required to reason first, consult AI second. This sequencing is crucial because it forces them to engage with the problem independently before relying on the machine. In my opinion, this friction is essential for learning. As learning scientists Elizabeth and Robert Bjork argue, “desirable difficulties” enhance long-term retention and skill transfer.
The Aviation Analogy
A comparison that I find particularly illuminating is aviation. Pilots aren’t taught to avoid autopilot; they’re taught to maintain manual flying skills by periodically disengaging automation. Medicine needs a similar approach. Trainees should be required to work through cases without AI, not just to prove they can, but to reveal gaps in their reasoning. What this really implies is that we need to treat AI as a tool to augment human judgment, not replace it.
The Trust Trap
Another layer to this issue is the misplaced trust in AI. A recent study in Nature Medicine found that AI tools like OpenEvidence can be less reliable than they appear. This is alarming because trainees are already relying on these tools without fully understanding their limitations. Personally, I think this highlights a broader cultural issue: our tendency to equate technological sophistication with infallibility. If we don’t teach trainees to interrogate AI critically, we risk creating a feedback loop where flawed reasoning perpetuates itself.
A Structural Solution
The solution, in my view, can’t rely on individual restraint. It has to be structural. Medical schools and residency programs need to set clear expectations about when and how AI should be used. For example, requiring trainees to submit a “pre-AI assessment” before consulting the tool could ensure they’re actively engaging with the case. Additionally, programs could run drills where trainees analyze AI-generated assessments with subtle flaws. This wouldn’t just test their knowledge; it would teach them to question the machine’s reasoning.
What this really suggests is that we need to rethink how we integrate AI into medical education. It’s not about banning the technology but about using it in a way that complements human reasoning rather than supplanting it.
The Human Element
Ultimately, medicine is as much an art as it is a science. A trainee who has seen pneumonia that looks like heart failure, or heart failure that looks like pneumonia, develops a bedside judgment that no AI can replicate. That nuanced understanding—what to notice, what to question, and when to distrust a familiar pattern—is what medical training aims to cultivate. AI should augment this process, not replace it.
In my opinion, the real danger isn’t AI itself but how we choose to integrate it into training. If we get this wrong, we risk creating a generation of doctors who are technically proficient but clinically shallow. But if we get it right, we could usher in an era where human judgment and machine intelligence work in harmony to deliver better care.
The question is: Will we prioritize efficiency over depth, or will we find a way to balance the two? Personally, I think the answer will define the future of medicine.
