Researchers find AI interaction leaves measurable marks on the human

ArXiv researchers shifted the lens on human-agent interaction. Instead of measuring what the agent did, they measured what changed in the human afterward. The paper names these changes and tracks them across repeated use. A person who delegates work to an agent may grow less confident in their own judgment. A person who reviews agent output under time pressure may grow less attentive. The paper draws on multiple studies to show these patterns are measurable and consistent. That grounding is what makes this useful to a product team: the claims are testable, and the paper offers a starting point for measurement. Most teams define success by task completion or output quality. They do not measure what the interaction does to the person over repeated use. For UX leadership, the implication is direct. Interaction patterns designed around efficiency or accuracy miss a second question: what does this pattern teach the human to do, and what does it erode? A design that improves short-term throughput may degrade long-term human capability or judgment. The paper gives teams a vocabulary and a measurement approach for a dimension of agent product quality that most teams ship without measuring.