#66: Designing Multi-Agent AI Systems for Collaborative Learning: Strategic Questioning Patterns and Cognitive Development in Large-Scale Online Learning

As large-scale online learning environments increasingly integrate AI technologies to support collaborative learning, understanding how learners strategically interact with different AI agent roles becomes critical for effective system design. Grounded in Bloom’s Taxonomy and Graesser and Person’s question taxonomy, this study examines questioning strategies employed by 411 doctoral students interacting with three distinct AI agents—Teacher, Assistant, and Companion—in a large-scale online course, and compares cognitive development trajectories between AI-mediated and human peer dialogues. Using clustering analysis on 224 AI-directed questions and sequential analysis of cognitive levels coded via Bloom’s Taxonomy, we identified three strategic patterns: Teacher-Focused Quick-Response Strategy (35.7%), Assistant-Focused In-Depth Strategy (27.7%), and Teacher-Focused In-Depth Strategy (36.6%). AI-involved dialogues showed higher cognitive-level variability (average progress of 0.48) compared to human dialogues (0.02), suggesting AI agents accelerate cognitive transitions while human interactions provide stability. These findings illuminate design principles for multi-agent systems that balance AI efficiency with human collaborative depth.