Integrating Artificial Intelligence and Learning Sciences to Reduce Cognitive Load and Achievement Gaps in Data-Driven K-12 Instructional Systems

Abstract

Persistent achievement gaps in K–12 education are increasingly linked to unequal cognitive load imposed by data-driven instructional systems that scale content without adequately modeling learner cognition. This paper proposes a technical framework that integrates artificial intelligence with established learning sciences to systematically reduce extraneous cognitive load while improving learning equity across diverse student populations. We introduce a novel algorithm, Cognitive Load–Adaptive Instructional Network (CLAIN), which combines graph-based knowledge tracing, multimodal cognitive state inference, and reinforcement learning–driven content sequencing. CLAIN dynamically adjusts instructional complexity, pacing, and representational modality based on real-time estimates of intrinsic, extraneous, and germane cognitive load.The proposed approach is empirically evaluated against widely used baselines, including Bayesian Knowledge Tracing, Deep Knowledge Tracing (LSTM-based), Item Response Theory–driven adaptive testing, and contextual multi-armed bandit recommenders. Using longitudinal classroom datasets with heterogeneous socio-academic profiles, we compare models across predictive accuracy, learning efficiency, and equity-focused metrics such as variance reduction in outcome distributions across demographic groups. Results show that CLAIN achieves superior performance, with lower cumulative cognitive load indices, faster concept mastery curves, and statistically significant reductions in achievement dispersion. Graph-based visualizations illustrate how adaptive pathways differ across learners and how the proposed model reallocates instructional effort to high-risk cognitive bottlenecks. Beyond performance gains, the paper contributes a technically grounded bridge between cognitive load theory and AI system design, demonstrating how learning-science constraints can be encoded directly into model architectures and optimization objectives. The findings suggest that cognitively aware AI systems can move beyond accuracy-centric personalization toward equity-oriented instructional intelligence, offering a scalable pathway for data-driven K–12 platforms to close achievement gaps without increasing learner burden.