The Anatomy of AI Integration in Student Learning: A Psychological Network Analysis of AI Appraisal and Self-Regulated Learning Across Use-Frequency Groups

Abstract

Artificial intelligence (AI) is increasingly embedded in students’ learning practices, yet little is known about how AI engagement evolves from an external technological aid into an agentic component of self-regulated learning. This study applies psychological network analysis to examine the structural relations among students’ knowledge of AI, perceived value and perceived cost of AI, intention to use AI, and three core self-regulated learning processes—forethought, performance control, and self-reflection—across different levels of AI use frequency. The study was conducted on a sample of 673 university students and early-career graduates. Networks were estimated using EBICglasso for the full sample and separately for low-, moderate-, and high-frequency AI users. Across all models, a stable two-system organization emerged, consisting of an AI appraisal subsystem (knowledge, value, cost, intention) and a self-regulation subsystem (forethought, performance control, self-reflection). However, the connectivity between these subsystems differed systematically by usage frequency. Among low-frequency users, perceived cost was more prominently positioned within the appraisal subsystem, suggesting that cost-related concerns may be more salient in lower-frequency use contexts. In contrast, in the moderate- and high-frequency groups, performance control appeared more centrally positioned at the interface between appraisal and self-regulation, suggesting stronger alignment between AI-related appraisals and performance-level regulatory processes in these groups. Students’ knowledge of AI displayed context-dependent structural roles across networks, consistent with a variable relational position across use-frequency groups. Overall, the findings suggest that AI appraisal and self-regulated learning form partially distinct but interconnected subsystems, and that their configuration may vary across AI use-frequency groups. Because subgroup comparisons were descriptive and formal stability analyses were not conducted, these findings should be interpreted as exploratory. The results do not support causal or developmental inference and require replication using bootstrapped stability analyses and formal network comparison procedures.