When AI does the making and you just grade its work, why does that smaller role make you care less?

This explores why people can become less engaged, and less invested, when AI does the making and their role narrows to checking and judging what the AI made. The corpus doesn't have a single study that isolates this mechanism. What it does have, read together, points to a clear answer: evaluation is a thinner kind of involvement than production. It gives people less control, less feedback and less ownership. It also tends to grow faster than anyone's ability to keep up with it.

The idea comes from decades of research on automation, not from AI. Simkute and colleagues take established Human Factors principles, first developed for cockpits and control rooms, and apply them to generative AI Can Human Factors design fixes restore GenAI productivity losses?. That older research found that when machines take over the doing, humans become monitors. Monitoring is hard to sustain because it gives you less feedback and less sense of what's going on. The proposed fixes show what gets lost: continuous feedback, interfaces that keep the task visible, stable task structure and clear division of who does what. One caveat matters: these fixes are proposals, not tested remedies. The diagnosis is better supported than the cure.

The ownership research adds a second piece. People feel a piece of AI-generated text is theirs in proportion to how much they shaped it Does user control over AI text shape feelings of ownership?. A personalized AI model didn't change that feeling. Only real influence over the output did. So when someone moves from writing to approving, the work feels less like theirs, and people put less care into things they don't feel they own. Tool design can make this worse. Generated interfaces in AI-assisted analysis make results clearer but harder to change mid-task Do generated analysis UIs really work better than chat?. That pushes users further toward approving or rejecting and away from shaping.

Then there's scale. When AI produces faster than people can judge, evaluation becomes a backlog instead of a craft. One paper calls this "epistemic hyperinflation": confidence in knowledge collapses the way money loses value in hyperinflation Can AI generate knowledge faster than humans can evaluate it?. The same pattern shows up in academic publishing, where AI-scaled paper production drives AI-automated review, which in turn invites manipulation and countermeasures Does AI create a coupled arms race in research production and review?. Once checking is handed to machines too, people are pushed one step further from the work.

One result you might not expect: "engagement" can drop for an almost opposite reason. In Nextdoor's experiments, better AI summaries *reduced* clicks Does better summary writing actually increase user engagement?. The AI had already done the reader's work of judging whether a post was worth opening, so they had no reason to click. Together, these findings suggest engagement depends on whether the user still has something meaningful left to do. Take away the making and people disengage from the work. Take away the judging as well and they stop showing up at all.