Most evaluations of human-AI interaction focus on what the artificial agent enables during use—task completion, efficiency, or accuracy. A new arXiv preprint argues this perspective overlooks a crucial effect: interaction can also alter the human who acts next. The paper proposes a framework called Residual Human States to represent these changes.
The framework is trajectory-based, meaning it tracks how a human's state evolves through interaction rather than treating the user as a static input or output. The central idea is that an AI system leaves a residual trace in the human—something that persists after the session ends and shapes subsequent behavior.
The work is grounded in a multi-study analysis, though the abstract does not detail the methods or domains of those studies. As a preprint, it has not yet undergone peer review, so the framework's empirical support and generalizability remain to be assessed.