Answers are grounded in this paper and cite their sources.
Objective While the use of generative artificial intelligence has become increasingly pervasive in higher education, the overall linear relationship between students’ perceived cognitive offloading and their academic engagement remains non-significant. Grounded in distributed cognition theory, the extended mind thesis, and cognitive load theory, this cross-sectional study examines whether multidimensional cognitive load allocation operates as countervailing indirect pathways linking perceived cognitive offloading to university students’ academic engagement, while evaluating the moderating boundaries of perceived learner agency and AI literacy. Methods A valid sample of 1,050 university students with generative AI experience was analyzed (representing a 93.75% retention rate from 1,120 submitted records). Data were evaluated using hierarchical linear regression (controlling for specific AI task types), non-parametric bootstrap parallel mediation modeling, moderated mediation analyses, and cross-disciplinary comparative regressions across humanities and social sciences ( n = 368) and STEM ( n = 465) cohorts. Results The findings indicated that: (1) Perceived cognitive offloading had no direct linear association with academic engagement, accounting for an inconsistent mediation pattern of two countervailing indirect associations: a negative indirect pathway via extraneous cognitive load (estimate = −0.025) and a positive indirect pathway via germane cognitive load (estimate = +0.053); (2) Sub-dimensional analyses indicated descriptive divergences across task categories: memory retrieval offloading was positively associated with extraneous verification load, whereas analytical processing offloading was linked to germane cognitive load; (3) Perceived learner agency positively moderated the germane load pathway, whereas AI literacy demonstrated direct associations with cognitive load without significantly moderating these indirect pathways; (4) Cross-group comparative analyses detected no statistically significant slope differences in this cognitive load allocation pattern across humanities/social sciences and STEM cohorts. Conclusion Perceived cognitive offloading in intelligent learning environments reflects a cognitive trade-off between operational verification friction and conceptual schema scaffolding. Rather than imposing blunt administrative prohibitions, these findings highlight the conceptual utility of supporting learners’ epistemic agency and critical AI literacy, assisting students in channeling technological offloading toward productive, germane engagement. Nevertheless, as these cross-sectional findings capture subjective cognitive trade-offs rather than longitudinal learning gains or objective memory alterations, instructional interventions should be tested empirically.
No takes yet. Share an insight, caveat, or question.
贾巾月 (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: