Critical Monitoring and Cognitive Offloading in GenAI-Mediated STEM Learning: A Moderated Mediation Model of Deep Engagement

Generative artificial intelligence (GenAI) can increase students’ interaction frequency and task completion without necessarily deepening learning engagement. This study examined how AI use, cognitive offloading, critical monitoring, dialogic reconstruction, and deep engagement are associated in STEM learning. A moderated mediation model was examined using structural equation modeling and hierarchical regression with 326 valid responses from STEM undergraduates. Usage intensity was negatively associated with deep engagement, whereas critical monitoring and dialogic reconstruction showed positive associations. Cognitive offloading risk mediated the relationship between AI use and deep engagement. Critical monitoring also moderated the link between usage intensity and cognitive offloading, with a weaker positive association among students reporting higher monitoring. The findings suggest that the educational value of AI-assisted learning depends not only on frequency of use, but also on whether students critically evaluate and reconstruct generated content. The study highlights the importance of preserving independent reasoning, strengthening critical monitoring, and structuring dialogic interaction in STEM education. It also underscores the need to distinguish AI-assisted performance from independently developed understanding.