AI tools are becoming integral to clinical workflows and education. That creates a specific risk for trainees: not losing skills they once had, but never acquiring them. This article outlines what that risk looks like and offers concrete programmatic steps medical schools and residency programs can take to protect independent clinical judgment.
The “Never-Skilling” Dilemma for Trainees
Never-skilling occurs when learners rely on AI outputs before they develop core reasoning. Unlike deskilling, where practice atrophies, never-skilling is a failure of formation. Trainees who accept suggestions from systems like OpenEvidence without practice may reach later stages of training with gaps in history taking, hypothesis generation, and differential diagnosis construction.
Beyond Automation: AI’s Unique Cognitive Impact
AI differs from prior tools because it participates in cognitive work. It can produce plausible, confident answers that mask uncertainty and bias. Trainees shaped by that feedback loop may struggle to interrogate outputs or recognize subtle errors. The risk is not only wrong decisions, but an erosion of habits that underpin safe, reflective practice.
Strategies for Building Resilient Judgment
Reason First, AI Second
Require unaided reasoning before AI use. Implement pre-AI assessments where learners document their history, exam findings, working diagnoses, and management plans. Make those responses part of summative assessment and learning portfolios.
Purposeful Friction
Design regular no-AI scenarios and observed structured clinical examinations. Use graded stations where technology is prohibited to measure manual competence. Rotate responsibilities so trainees practice skills that AI might otherwise perform.
Interrogating AI
Teach methods to probe and test AI outputs. Run simulated exercises with intentionally flawed AI suggestions. Instruct learners to request provenance, challenge assumptions, and perform independent checks. Log when AI was consulted as part of performance reviews.
Augmenting Human Expertise, Not Replacing It
AI offers speed and breadth of information and can be a powerful tutor when used deliberately. The goal for educators is clear: use AI to expand learning opportunities while structuring training so that judgment, skepticism, and clinical craft are learned first and reinforced later.
Adopting these structural practices will help produce clinicians who can use AI wisely and still stand accountable for decisions that machines may never fully explain.
