A Generative AI-Enhanced Mobile Learning Framework for Adaptive Human–Machine Collaboration

Abstract

Ubiquitous mobile learning has emerged as a central modality in higher education. While generative artificial intelligence offers significant potential for personalized learning, its deployment is constrained by two fundamental bottlenecks: the hard resource limitations of end-device computing and the inadequate adaptability of human–machine instructional roles. Existing systems predominantly rely on cloud-end binary scheduling and static role partitioning, which fail to simultaneously guarantee operational performance on mobile terminals and collaborative teaching effectiveness. To address these challenges, a three-tier progressive agent architecture was proposed. A cognition-driven progressive routing protocol was introduced to dynamically align computational resource levels with the cognitive load imposed by learning tasks. Local clustering pruning and cache gain mechanisms were further incorporated to optimize retrieval efficiency on the end-device side. Building upon this infrastructure, a cognition–collaboration dual-loop nested dynamic weighting model was formulated, in which technical architectural stability was incorporated as a decision variable within the framework. Smooth switching of the human–machine instructional roles was achieved through continuous weight computation, and an iterative closed loop with bidirectional coupling was established. This study provides a technical paradigm for the scalable deployment of generative artificial intelligence in mobile learning scenarios, addressing both engineering practicality and instructional adaptability.