Beyond AI Teammates: Collective Vicarious Cognition as a Framework for Educational Multi-Agent Systems

Abstract

Artificial intelligence is assuming an increasingly prominent role in educational contexts, yet the notion of the “AI teammate” is often reduced to the generative capabilities of Large Language Models (LLMs). Such a view preserves a tool-centred logic in which AI supports individual productivity without substantially contributing to collaborative learning processes. This paper argues that meaningful forms of human–AI collaboration in education require socio-technical architectures capable of organizing cognitive plurality and making reasoning processes visible, interpretable and open to evaluation. Drawing on Social Learning Theory, the concept of vicariousness, and perspectives on collective agency, the paper introduces the concept of Collective Vicarious Cognition (CVC) as a pedagogical framework for understanding learning as engagement with multiple, interacting cognitive trajectories. Within this perspective, learning emerges not simply through exposure to information, but through the observation, comparison, interpretation, and regulation of distributed forms of reasoning. Multi-agent systems based on LLMs (LLM-MAS) are examined as a possible educational infrastructure capable of supporting these conditions by distributing cognitive roles across interacting agents and rendering epistemic plurality observable. The paper argues that the educational significance of AI teammates lies less in their capacity to generate responses than in their potential to support organized cognitive diversity, reflective judgment, collective agency, and collaborative knowledge construction within hybrid human–AI learning environments.