Interaction-Centered Intelligence: Toward Interaction as the Primary Unit of Analysis in Co-Creative AI and Human-AI Systems

A new academic framework proposes interaction as the primary unit of analysis for understanding intelligence in human-AI systems, shifting focus from isolated computation within individual models to the relational dynamics that emerge through collaborative engagement. The paper synthesizes decades of research across distributed cognition, embodied cognition, and computational creativity to argue that intelligence, creativity, and meaning arise from evolving interaction patterns rather than internal computation alone.

This paper represents a significant theoretical shift in how researchers conceptualize artificial intelligence, moving beyond the dominant paradigm that evaluates AI systems through isolated metrics like prediction accuracy or benchmark performance. Traditional AI frameworks treat intelligence as computation occurring within bounded agents, evaluated primarily through outputs and optimization metrics. The proposed Interaction-Centered Intelligence framework challenges this assumption by positioning human-AI collaboration itself as the fundamental unit of analysis, drawing from decades of work in cognitive science, HCI, and participatory sense-making.

The theoretical foundation for this approach emerges from multiple disciplines converging on relational accounts of intelligence. Rather than treating intelligence as a property contained within individual systems, this framework recognizes that meaning, creativity, and adaptive behavior emerge through dynamic coordination between humans, AI systems, and their shared environments. The paper references concrete systems like the Drawing Apprentice and AI Drawing Partner to demonstrate how intelligence manifests through interaction trajectories and coordination patterns over time, not just final outputs.

For the AI development community, this framework offers practical implications for designing more effective human-AI systems. Instead of optimizing for isolated model performance, developers would prioritize interaction quality, adaptive participation, and shared understanding. This affects how AI systems should be evaluated, explaining their reasoning through interactive dynamics rather than post-hoc output analysis. The framework has particular relevance for explainable AI and hybrid intelligence systems where human understanding and engagement directly influence outcomes.

• →Interaction-Centered Intelligence proposes that intelligence emerges from relational dynamics between humans, AI systems, and environments rather than from isolated computation.
• →The framework synthesizes decades of research across distributed cognition, embodied cognition, and computational creativity into a cohesive theoretical model.
• →Evaluation metrics shift from output-focused benchmarks to interaction trajectories, coordination patterns, and adaptive participation over time.
• →The approach has direct implications for designing explainable AI and hybrid intelligence systems that prioritize human understanding and engagement.
• →This represents a paradigm shift in AI research methodology, moving away from the dominant agent-centric computational model toward relational accounts of intelligence.