Our Take
The deskilling worry is real and widespread, but hospitals haven't moved from anxiety to mitigation—most are still figuring out how to measure skill erosion once it happens.
Why it matters
Healthcare workers are the bottleneck in AI adoption, not the models. A widespread perception that tools degrade judgment will slow clinical deployment and force vendors and health systems to design differently, with human oversight and feedback baked in from day one.
Do this week
Clinical leadership: map which decisions you're automating via AI in the next 12 months, then design a quarterly audit of clinician performance on those same decisions when done without the tool—before skills actually atrophy.
Three-quarters of clinicians worry AI will dull their judgment
A survey by Wolters Kluwer Health found that nearly 75% of clinicians identified losing critical thinking or decision-making skills as one of the greatest risks of adopting artificial intelligence in healthcare settings (per Wolters Kluwer). The concern surfaces as AI adoption accelerates across provider networks and electronic health record systems, with tools now assisting in diagnosis, treatment planning, and administrative triage.
The worry is not hypothetical. When workers offload judgment to tools for extended periods, cognitive skills atrophy measurably. Radiologists who rely heavily on AI-assisted reads show measurable declines in independent diagnostic accuracy within months if the habit goes unchecked. Similar patterns appear in fields from piloting to law, where automation bias and skill loss have documented precedent.
Clinician confidence is the real constraint on adoption
Healthcare AI deployment has not hit a capability ceiling—the models work. The constraint is human. If three-quarters of the workforce believes a tool will erode the judgment they depend on to practice safely, adoption stalls, adoption becomes performative, or both.
This matters for vendors and health systems equally. A clinician who feels deskilled is one who either refuses the tool, uses it passively without engaging its output, or demands so much oversight that the time savings vanish. The perception of risk becomes a real drag on ROI. Health systems will need to move from abstract "human-in-the-loop" language to concrete design: clear audit trails showing when and why the AI recommendation differed from a clinician's independent call; regular structured practice on decisions without the tool; transparent performance data on both clinician and AI accuracy over time.
Start measuring before automating
The gap between concern and action is wide. Clinicians report anxiety but most health systems have no baseline of decision quality before deploying AI assistance. You cannot know if judgment has eroded unless you measured it beforehand.
Clinical operations teams should establish decision audits in priority areas before rolling out AI tools. Track diagnostic accuracy, treatment plan quality, and time-to-decision for a cohort of clinicians working without assistance. Then, once the AI tool ships, repeat the same audit quarterly on a rotating basis (clinicians working with and without the tool on parallel cases). If scores drop after six months of tool use, you have early signal to adjust training, tool design, or deployment scope. If scores hold steady, you have proof to show skeptics that the tool augments rather than replaces judgment. That data will drive adoption faster than any vendor pitch.