Cognitive-Aware AI Systems for Learner-Centric Education

Abstract

Cognitive-aware AI systems integrate models of human cognitive processes -- working memory load, cognitive load theory, dual-process reasoning, and metacognitive regulation -- into educational AI design to create learning environments that adapt not just to what learners know, but how they are currently thinking and processing information. By monitoring cognitive load indicators (response latency, error patterns, physiological signals, gaze behaviour) and dynamically adjusting instructional complexity, modality, and pacing, cognitive-aware systems aim to maintain learners in their optimal cognitive engagement zone -- challenging enough to promote learning, manageable enough to prevent overload. This paper proposes the Cognitive-Aware Educational AI Framework (CAEAF), evaluating five cognitive monitoring approaches and four instructional adaptation strategies across 2,400 learner sessions in mathematics, reading comprehension, and STEM problem-solving. CAEAF introduces the Cognitive Alignment Score (CAS) measuring how closely the system maintains learners in their optimal cognitive engagement zone. Key results: multimodal cognitive monitoring (gaze + response latency + error patterns) achieves CAS = 0.904; adaptive pacing with worked-example fading reduces cognitive overload incidents by 42.4%; metacognitive prompting improves self-regulation scores by 28.6%. The framework provides design principles for cognitive-aware learner-centric educational AI.