Unpacking Interaction Profiles and Strategies in Human-AI Collaborative Problem Solving: A Cognitive Distribution and Regulation Perspective

A research team from multiple institutions, led by Zhanxin Hao, has published a groundbreaking study on arXiv titled "Unpacking Interaction Profiles and Strategies in Human-AI Collaborative Problem Solving." The research applies frameworks of distributed cognition and regulation of learning to analyze how college students collaborate with AI on complex problems. Through detailed cluster analysis of interaction patterns, the team identified three distinct collaborative problem-solving modes: Delegated Reasoning (where humans offload reasoning to AI), Concerted Interpretation (joint sense-making), and Delegated Elaboration (offloading detailed work).

The study's key finding is that the Delegated Reasoning (DR) group achieved the highest overall task performance, significantly outperforming the Concerted Interpretation (CI) group. Furthermore, the semantic similarity between human and AI discourse was notably highest in the DR group, suggesting more aligned thinking. Interestingly, the CI group reported significantly greater use of self-regulation strategies by the human learners. This reveals a critical tension: the most efficient distributed system (DR) may come at the cost of deeper human cognitive engagement and regulation.

These insights provide a crucial empirical foundation for the future of human-AI teamwork, particularly in educational settings. The research moves beyond simplistic notions of AI as a tool or tutor, instead framing it as a collaborative partner in a distributed cognitive system. The identified profiles offer a blueprint for designing next-generation AI-empowered educational tools that can adapt to or encourage specific collaboration styles, balancing performance outcomes with learning depth and user agency.