Generative AI and cognitive load in education: a systematic review of WoS/SSCI-indexed studies through the lens of cognitive load theory

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Generative artificial intelligence (GenAI), together with closely related educational AI systems retrieved by the executed search, has been rapidly adopted across education, prompting empirical work on its consequences for learners' cognitive load. This systematic review of English-language, SSCI-indexed journal articles in the Web of Science Core Collection examined 39 empirical studies through the tripartite architecture of Cognitive Load Theory (CLT): intrinsic, extraneous, and germane load. Studies were dual-coded across 14 dimensions. The corpus was strongly experimental (28 of 39 studies) and dominated by undergraduate samples. Sixteen studies measured extraneous load as a distinct subtype: seven reported lower extraneous load, three reported no difference, one reported higher extraneous load, and one reported a curvilinear pattern under the reported AI/GenAI contrast; four did not provide a directional comparison. A separate 22 studies reported an overall or undifferentiated index, for which 11 reported lower load, five reported no difference, and two reported higher load, with four non-directional cases. These groups must not be pooled as evidence that AI or GenAI specifically reduces extraneous load. Eleven studies included a germane-load subscale, but only five yielded a directional comparison (four higher and one curvilinear). The modal overall verdict was conditional rather than uniformly beneficial (21 conditional, 16 beneficial, two mixed): benefits depended on scaffolding, dosage, learner prior knowledge, and task design. Only 13 studies framed AI/GenAI through cognitive offloading. We therefore argue that future research should measure the CLT components separately, report the AI/GenAI contrast transparently, and use an offloading lens to distinguish support for learning from substitution for learning.