认知与元认知何以协同?OA

The emergence of LLM-based pedagogical agents offers new possibilities for personalized learning while bringing risks of cognitive outsourcing and metacognitive laziness.In response,this study constructed a pedagogical agent integrating a dual cognitive-metacognitive diagnostic mechanism.It dynamically identifies learners' cognitive levels and metacognitive phases to adap-tively match corresponding cognitive and metacognitive scaffolds.To verify its effectiveness,a quasi-experimental study was conduct-ed with 66 fifth-grade students from two intact classes over eight instructional periods.The experimental group carried out agent-sup-ported algorithm problem-solving learning,whereas the control group received teacher-supported algorithm problem-solving learning.Results showed that the experimental group significantly outperformed the control group in academic achievement and metacognitive awareness and demonstrated significantly lower cognitive load.Epistemic network analysis further revealed that the core difference be-tween high-and low-performing students lay in whether they could efficiently achieve deep understanding of new knowledge.Their cognitive development trajectories demonstrated advanced characteristics marked by transitions from linear trial-and-error to multi-dimensional association and from partial comprehension to the integrated construction of holistic frameworks.These findings suggest that the key to intelligent educational support lies not in providing more answers,but in dynamically adapting scaffolds to learners' re-al-time states so as to trigger cognitive restructuring and metacognitive awakening.The findings provide empirical evidence for un-packing the underlying mechanisms of human-AI collaborative learning.