A Model of Integrated Information Processing in Human-AI Interaction

AI systems should be designed around how humans actually regulate their actions and make decisions, not just technical capabilities. The paper argues that effective human-AI interaction requires understanding psychological mechanisms of action regulation, meaning designers need to consider how people process information and adjust their behavior in real-time.

Three specific design qualities determine whether humans can effectively work with AI: input adequacy (humans receive the right information), reference consonance (humans and AI share the same goals/understanding), and output operativity (humans can actually act on AI recommendations). When these three qualities are present, humans experience greater transparency and controllability over AI systems.

Explainable AI (XAI) techniques should be evaluated based on whether they improve human action regulation, not just whether they provide explanations. The paper suggests that simply making AI "more explainable" doesn't automatically make systems more trustworthy or usable—what matters is whether explanations help humans understand what to do next and maintain appropriate control.

Current approaches to human-AI interaction often describe system design but fail to predict how that design actually changes human behavior. The paper identifies a gap in existing frameworks: they explain how systems are built but not why those design choices lead to better human performance and decision-making in practice.

Human-AI teams should be modeled as coupled control loops rather than separate entities, meaning the system design must account for continuous feedback between human actions and AI responses. This cybernetic perspective suggests that effective AI systems require ongoing adjustment and alignment between human and machine objectives.