The Gruntwork You Outsource Is the Expertise You Never Build
AI doesn't just risk making us lazy. It risks making us look competent while the expertise never forms — and the fix is not less AI, but the intentional use thereof.
Offloading the labor that turns a novice into an expert is an excellent way to never become an expert.
01 / THE MECHANISM
De-skilling / Never Skilling
Answers to complex questions are instantaneous nowadays. De-skilling is losing a skill you once had; never skilling is what happens when the tool arrives before the skill ever forms. The judgment required to recognize a wrong answer takes effort to develop. You only get it by doing the reps — building the differential, committing before you're corrected, being wrong in a way you remember. (I remember getting kicked out of a patient room once for not knowing the global period for a toenail avulsion - I sure know it now!) The effort required to develop real expertise must not be delegated to AI. The struggle isn't friction on the way to skill. The struggle is the skill.
02 / WHY IT HIDES
Cognitive Spoofing
AI tools produce work that looks like competence, so no one — including the learner — can tell it didn't form in a long-lasting way. Looking something up inflates your sense of what you personally know; the tool's knowledge gets miscredited to the self. Felt competence, decoupled from the real thing. A counterfeit that fools the person holding it.
The Calculator Objection — and the Answer
"Every tool triggered this panic. We adapted. Why is AI different?"
What it offloads
Not computation or recall — judgment and reasoning. The faculties that make a clinician, not a technician.
The breadth
Earlier tools took one task. AI reaches every cognitive task at once — history, differential, plan, note.
The failure mode
Wrong GPS, you're late. A confidently wrong model with no one able to catch it — you can cause patient harm.
GPS Shows the Difference
GPS de-skilled an entire generation away from navigating by paper map. Every generation after may now never skill that ability at all — and in some cases, that is probably fine. Some skills can be safely replaced. But medicine is not paper navigation: when AI replaces the reps that build clinical judgment, the long-term risk is not inconvenience. It is humanity forgetting how to verify the machine.
Twenty Years Down the Wire — the version to avoid
A cohort of fluent prompters who never built the substrate beneath. Productive on paper, but unable to verify what they're handed — because verification needs the judgment that irresponsible AI use prevented them from developing.
The Fix Is Intention
Use It Deliberately
01Generate first, then check.Build your own plan, then run it past the model. The order is the entire discipline.
02Widen, don't decide."What am I missing?" keeps your reasoning. "What's the plan?" outsources it.
03Attest, don't accept.Treat output like a junior's note you cosign — your name means you verified it.
04Run the spoof test.Could you reproduce the reasoning without the tool? If not, you're renting it, not owning it.
Who Instills This?
The mindset doesn't install itself. Trainees won't reintroduce friction that AI removes — not without someone showing why the friction and the struggle was the point all along. (Journey Before Destination) That job falls to educators. But educators are as much at risk of deskilling as learners. Educators have all the more important imperative to teach responsible and intentional AI use.
The goal is not less AI. It's using AI responsibly in ways that cultivate the skills that forge you into an expert instead of preventing you from becoming one.