Learner–AI interactions in collaborative learning: Patterns and dynamics of co- and shared regulation in human-human and human-AI learning processes

Collaborative learning depends on more than exchanging information. Learners need to question, monitor, respond to one another, manage disagreement and sustain engagement together. As artificial intelligence (AI), particularly generative AI, becomes increasingly conversational and socially expressive, an important question is whether these same processes can also emerge with AI as a learning partner.

This dissertation examines what happens to learning and regulation when the collaborative partner changes from another person to AI, and shows that the shift from using AI as a tool towards engaging it as a co-participant is shaped by the interaction that develops between learner and AI.

The dissertation consists of three studies examining human-human group collaboration, spoken interaction with generative AI, and interaction with an embodied AI agent in mixed reality. The studies involved Finnish upper-secondary students, Vietnamese university students, and university students in Finland. Video and interaction data were analysed using a process-oriented approach to trace cognitive and socio-emotional interactions over time.

The findings show that productive learning and regulation depend not only on what participants contribute, but on how their contributions connect and develop through interaction, whether between humans or between a human and AI. In human groups, adaptive regulation was characterised by deliberation, shared monitoring and coordinated responses to challenges. With speech-based AI, reciprocity became important for moving beyond simple question-and-answer exchanges, while embodied AI added gaze, posture and gesture that further supported responsiveness, socio-emotional engagement and learner initiative. Taken together, the studies suggest that as AI becomes more socially expressive, it can create stronger conditions for reciprocal and partner-like interaction. However, this shift is not automatic. It depends on how learners perceive and engage with the AI, and on how responsively the AI participates in the interaction.

Overall, the dissertation identifies reciprocity, socio-emotional engagement, learner initiative and perceived AI responsiveness as central to the move from tool use towards co-participation. Methodologically, it shows how process-oriented interaction analysis can capture these changes as they unfold. For education, the findings suggest that AI should support questioning, reflection and shared regulation while preserving learners’ active role in directing and evaluating their learning.