Never-skilling

The older fear was deskilling — a radiologist who leans on an AI flag long enough to lose the pattern-recognition she spent a decade building. A pair of recent medical-education papers name a different one, aimed not at seasoned clinicians but at the people still in training. Call it never-skilling: a resident who lets a language model write every differential logs the hours of residency but bypasses the mental work those hours exist to force, and so never builds the reasoning in the first place. The term was coined in a 2025 NEJM review and turned into its own thesis by a 2026 Nature Medicine Perspective, which adds a name for the result — 'false proficiency,' competence that looks indistinguishable from the real thing right up until the AI isn't there.

'False proficiency' — competence that looks indistinguishable from the real thing right up until the AI isn't there. — Nature Medicine Perspective

The honest part is buried in the abstract: the authors sounding the alarm concede there is no direct evidence it is happening in medical training yet. The case rests on learning theory plus signals from outside the clinic — most vividly an MIT study in which students who used AI to write an essay believed the work was their own but couldn't recall a single line of it moments later. So the before-and-after anchoring a claim about future doctors is actually about undergraduates and essays.

That gap is the point, not a weakness to paper over. This is a precautionary framework arriving before the phenomenon it describes — an attempt to name a risk while curricula are still being written, rather than to document one already loose in the wards. Whether never-skilling turns out to be real or a well-argued false alarm, the distinction it draws is the useful part: losing a skill, never acquiring one, and — the third case in the triad — internalizing an AI's confident errors as fact are three different failures, and a training system that only guards against the first will miss the other two.