Many hands make light work? Individual and collaborative interaction with a
GenAI
‐powered multi‐agent system: Effects on learning performance and critical thinking
Xiaoyan Chu, Hongzhu Dai, Xuesong Zhai Abstract
The rapid proliferation of generative artificial intelligence (GenAI) has made fostering learners' critical thinking an increasingly pressing educational challenge. While single‐agent GenAI applications are prone to limited perspective diversity and echo‐chamber effects, multi‐agent systems (MAS) offer a promising alternative by enabling multi‐perspective interactions. Meanwhile, the pedagogical potential of collaborative human–GenAI interaction, particularly when peers jointly engage with MAS to promote epistemic diversity, remains underexplored. Moreover, little is known about how critical thinking unfolds within such interactions at a process‐oriented level. This study investigates how different configurations of peer collaboration and GenAI‐powered multi‐agent support are associated with learning performance, critical thinking awareness and interaction processes. A quasi‐experimental study was conducted with 87 undergraduate students enrolled in an instructional design course. Participants were assigned to a peer collaboration condition without AI support, an individual human–MAS interaction condition or a collaborative human–MAS interaction condition. Learning performance was assessed using knowledge tests and design projects, critical thinking awareness was measured via a questionnaire and critical thinking processes were examined through lag sequential analysis and epistemic network analysis. The results indicate that collaborative interaction with the MAS corresponded to higher project performance, whereas individual interaction was associated with greater gains in critical thinking awareness. Process analyses further suggest distinct cognitive trajectories across configurations, with individual interaction characterised by more focused analytical–evaluative processing and collaborative interaction involving more distributed and integrative reasoning. These findings highlight the importance of aligning MAS design with appropriate human–AI collaboration configurations to better facilitate critical thinking in GenAI‐supported learning environments.
Generative artificial intelligence (GenAI) is increasingly being incorporated into educational settings, raising ongoing interest in how AI‐supported learning environments can engage learners in higher‐order cognitive processes, including critical thinking. Many existing AI‐supported learning environments rely on single‐agent systems, which may unintentionally encourage answer‐seeking behaviours and reinforce echo‐chamber effects. Multi‐agent systems (MAS) provide role‐differentiated support and exposure to diverse perspectives, and have been suggested as a promising approach for supporting higher‐order thinking in learning.Practitioner notes
What is already known about this topic
What this paper adds
By developing a GenAI‐supported multi‐agent system, this study empirically compares learning outcomes across peer collaboration without AI support, individual interaction with the MAS, and collaborative interaction with the MAS, showing that individual interaction is linked to higher critical thinking awareness gains, whereas collaborative interaction corresponds to stronger performance on complex design‐oriented tasks. Through the integration of lag sequential analysis and epistemic network analysis, the study reveals how overt interaction behaviours and implicit cognitive structures jointly reflect critical thinking processes in multi‐agent learning environments. The findings provide initial process‐level insights suggesting that multi‐agent systems may function not merely as information providers, but as reflective and socially mediating supports that facilitate how learners regulate, negotiate and integrate ideas during learning.