Enhancing deep learning in AI-enhanced education: a dual mediation model of cognitive load and learning motivation through interaction quality

Abstract

This research develops and validates a dual mediation framework examining the pathways through which interaction quality in artificial intelligence educational systems is positively associated with deep learning outcomes via cognitive load reduction and motivational enhancement. Utilizing covariance-based structural equation modeling (CB-SEM), we analyzed survey data from 570 university teachers engaged with AI-powered learning platforms. Findings demonstrate that high-quality human-AI interaction significantly reduces cognitive burden, which in turn is positively related to learning motivation and shows pathways to deep learning approaches. Bootstrap procedures confirmed robust sequential mediation effects, with this pathway accounting for 53% of the total variance. The model achieved excellent fit indices and explained 31.5% of variance in deep learning outcomes. By synthesizing Cognitive Load Theory with Self-Determination Theory, this study contributes to educational technology scholarship by elucidating the psychological mechanisms linking interface design to learning depth. The empirical evidence provides actionable insights for developing AI educational systems that strategically minimize cognitive demands, foster motivational engagement, and support meaningful learning experiences.