New IIP Model Unifies Human-AI Interaction with Three Key Qualities

Tim Schrills and Thomas Franke have introduced the Integrated Information Processing (IIP) model, a theoretical framework for Human-AI Interaction (HAII) published on arXiv. The model conceptualizes humans and AI as coupled control loops engaged in shared tasks, using a unified modeling language to describe both agents. It addresses a gap between existing automation frameworks and psychological mechanisms of action regulation, providing a process-oriented account of how interface design affects human behavior.

The core of the IIP model lies in three integration qualities: input adequacy (how well the AI receives relevant information), reference consonance (alignment of goals between human and AI), and output operativity (clarity of AI's actions for human control). These qualities directly influence human-centered benchmarks like transparency and controllability. The authors demonstrate applications to XAI techniques and interface design, offering a systematic way to map design choices to expected user behavior. This work is significant for researchers and practitioners building autonomous AI systems that must collaborate effectively with humans.