Deskilling is what happens to a practised skill when a system performs it instead. The word predates AI by decades and was documented in aviation when autopilots became standard and in navigation when GPS did. The clearest recent case is in gastroenterology: nineteen experienced endoscopists in Poland began routine work with an AI polyp-detection system, and within about three months their adenoma detection rate in procedures done without the AI had fallen from about 28 percent to about 22 percent.
It matters for governance because every oversight design assumes a reviewer who could still do the work. Human-in-the-loop, sampling audits, escalation paths and sign-off all depend on someone whose unassisted judgement remains sound. If that judgement has eroded while the tool was on, the oversight is nominal and the organisation does not know it, because the assisted metrics look fine.
It is distinct from automation bias, which is a trust failure in the moment by someone who still has the skill. Deskilling is the skill itself going. The two compound: as skill fades, deferring to the tool becomes the rational choice, which removes the last of the practice. The remedy the endoscopy study points to is dull and effective: measure unassisted performance on a schedule and treat a fall as a finding.