Could leaning on AI at work quietly wear down the very judgment we need to catch its mistakes?

This explores whether leaning on AI tools slowly wears down workers' own skills and judgment, and why that decline might go unnoticed. The most direct evidence in the collection comes from a study that mapped 8,356 workplace AI risk scenarios. It found that keeping a human 'in the loop' does not by itself protect anyone Does AI augmentation protect workers from skill erosion?. The usual case for augmentation over automation is that humans stay in charge and catch mistakes. That only works if they keep the skills that let them catch mistakes. When people rely too heavily on AI agents, those skills fade, and the oversight that was supposed to make augmentation safe fades with them.

The less obvious finding is that people usually can't feel this happening. One line of research describes an 'LLM fallacy': when AI-assisted work comes out smooth and polished, users count it as proof of their own ability and come to believe they have skills they don't Do AI-assisted outputs fool users about their own skills?. A companion paper names four mechanisms behind this: it's unclear who did what, fluent output feels like understanding, thinking gets handed off to the tool, and the workflow hides where the AI's contribution starts and stops. The paper argues these multiply each other rather than simply adding up How do AI tools trick users into overestimating their own skills?. So the erosion doesn't show up as feeling less capable. It shows up as feeling more capable while actually becoming less so.

Judgment has its own version of this. One framework describes LLMs as fast, intuitive 'System 1' thinking at scale. When users treat fluent answers as reasoned ones, three traps compound: mistaking the model's description of the world for the world itself, mistaking intuition for reasoning, and having existing biases confirmed Why do people trust AI outputs they shouldn't?. Security research describes the same pattern from the other side. The riskiest systems are the ones that work well most of the time, because their competence quietly wears down the skepticism people need to catch the rare failure How do competent systems quietly undermine safety oversight?. Reliability itself trains people to stop checking.

A social factor makes this harder to spot inside organizations. In four experiments with 4,439 participants, people who used AI expected colleagues to see them as less competent and less diligent, and they were less willing to tell managers they had used it Do people fear judgment when they use AI at work?. If workers hide their AI use, managers lose the information they would need to notice declining skills. At the scale of whole societies, the 'gradual disempowerment' argument goes further. Institutions stay aligned with human interests partly because they depend on human workers who care about outcomes. As AI takes over that work, the human check weakens, possibly beyond the point of reversal Does incremental AI replacement erode human influence over society?.

The collection does not yet include long-term studies that measure one group's skills declining over months of AI use. The case rests on mechanisms and risk mapping, not on long-term measurements. Labor research suggests where to look. Delegation to AI is concentrated in information-heavy jobs Where have workers actually delegated tasks to AI?. When AI affects only a few tasks in a job, workers tend to shift toward the tasks it doesn't touch Does concentrated AI exposure enable workers to adapt and reallocate?. This raises an open question: does erosion hit hardest where AI covers so much of the job that there's nothing left to shift to?