In today's rapidly evolving healthcare landscape, a pressing concern arises: what happens when medical students, the future of medicine, rely heavily on AI without developing their own clinical judgment? This question is not just theoretical; it's a real-world dilemma with potential far-reaching consequences.
The AI-Assisted Trainee
Imagine a medical trainee, armed with an AI tool like OpenEvidence, which provides instant answers to complex medical queries. While this may seem like a boon, it raises a critical question: are these trainees truly learning, or are they becoming reliant on a machine's judgment?
The traditional medical training path is an apprenticeship, a journey where mistakes and uncertainties are not just accepted but essential. Each step, from student to resident to fellow, is marked by these experiences, shaping the physician's clinical reasoning. However, with AI, this process is being disrupted.
The Danger of Never-Skilling
The concept of deskilling, where a skill is lost over time, is concerning. But the idea of never-skilling, where a skill is never truly developed, is even more alarming. A doctor who has forgotten how to reason can relearn, but one who never learned in the first place faces a different challenge entirely.
When trainees use AI tools, they may get perfect answers, but they miss out on the process of learning. They don't experience the struggle, the failure, and the gradual building of clinical intuition. This is a critical part of medical training, and without it, we risk creating supervisors who are adept at using AI but lack the fundamental skills of clinical reasoning.
The Trap of AI Dependence
Many trainees are aware of this trap, understanding that AI tools can become a crutch. Yet, they feel pressured to use these tools to keep up with their peers. It's an arms race, where opting out feels like a disadvantage. This highlights the need for a structural solution, not just individual restraint.
Shaping AI Use in Medical Training
Medical schools and residency programs must play a pivotal role in shaping how and when trainees use AI. While it's impractical and unnecessary to police every search, simple expectations can be set. Trainees should be encouraged to reason first, consult AI second, and make their initial unaided assessments visible. This could involve writing a brief pre-AI assessment after examining a patient, or pausing to discuss changes in diagnosis or treatment plans before consulting AI.
Learning from 'Desirable Difficulties'
The concept of 'desirable difficulties' in learning science is key here. While it may slow performance in the moment, it improves long-term retention and skill transfer. In the context of AI, this means using the technology after an independent attempt, allowing it to serve as a tutor, highlighting areas of improvement.
Learning from Aviation's Example
Aviation provides an interesting precedent. Pilots in training are not taught to avoid autopilot; instead, they're taught to maintain their manual competence. Similarly, medicine should ensure that trainees periodically work on no-AI cases, assessed on their unaided reasoning. This reveals potential drift and ensures that clinical judgment remains sharp.
Interrogating AI
Trainees should also be taught to critically evaluate AI outputs. This could involve running medical equivalents of flight simulator drills, where subtle flaws are introduced, and trainees are assessed on their ability to identify and rectify these errors. It's about teaching disciplined judgment, not reflexive skepticism.
The Core Competency of Clinical Reasoning
In the end, the struggle to independently reason through a patient's case is not about virtue signaling or hazing. It's about developing a core competency, a rich bedside judgment that can distinguish between similar conditions and identify when a familiar pattern should be questioned. AI can augment this process, but it should never replace it.
Conclusion
AI is here to stay, and its benefits to patients are undeniable. But we must ensure that doctors can stand apart from the machine, able to recognize when it's wrong, incomplete, or right for the wrong reasons. It's a delicate balance, and one that medical training must navigate carefully.
