Promoting the advanced development of critical thinking through the construction of a human-GenAI collaboration learning framework: a design-based implementation research

Generative artificial intelligence (GenAI) is rapidly entering higher education, yet its capacity to generate fluent responses does not by itself explain how human–AI collaboration should be designed to support critical thinking. Guided by hybrid intelligence, hybrid human–AI regulation, socially shared regulation, hybrid intelligent feedback, and critical-thinking progression, this study constructed and iteratively refined a Human-GenAI collaborative learning (HGCL) framework supported by a generative agent system using Design-Based Implementation Research (DBIR). The study was conducted over 14 weeks in an undergraduate Information Technology Education course with 50 educational technology students. The study used discourse coding, frequency analysis, three-dimensional epistemic network analysis (3D ENA), and repeated-measures ANOVA to trace developmental patterns across cycles. The findings reveal nonlinear critical thinking development, asynchronous growth across dimensions, and increasing coherence in students’ cognitive networks. Metacognitive abilities developed cumulatively, with significant gains between Cycles 1 and 3. Self-efficacy followed divergent trajectories with individual efficacy changed earlier, group efficacy later. Additionally, further analysis shows a possible reciprocal association between metacognitive regulation and self-efficacy, which may be associated with learners sustained cognitive engagement and strategy adjustment in complex human-machine collaborative contexts. Given the single-group DBIR design, these findings reflect process-oriented developmental trends across implementation cycles.