Cognitive Symbiosis or Cognitive Dependence: The Inherent Tensions and Governance Pathways of Generative Artificial Intelligence in Reshaping Education

Generative artificial intelligence is profoundly reshaping the cognitive ecology of education. Yet much mainstream research remains confined to a framework of "tool-based augmentation", treating it as an external means of improving efficiency and therefore struggling to explain deeper changes in human-machine relations. This paper argues for moving beyond instrumentalism: generative AI should be reconceived not as a teaching tool external to the learner, but as a "cognitive mediator" that participates in the learner's construction of meaning. It then uses the paired categories of "cognitive symbiosis" and "cognitive dependence" to characterise two possible trajectories of human-machine collaborative learning. The analysis shows that the educational use of generative AI contains three interrelated tensions: cognitive augmentation versus degradation at the individual level, efficiency gains versus widening divides at the social level, and personalisation versus homogenisation at the level of knowledge. These tensions share a common source: when machine output replaces human cognitive labour without preserving human agency, symbiosis degenerates into dependence. On this basis, the paper proposes "responsible cognitive symbiosis" as a governance orientation, to be pursued through coordinated action in AI literacy, instructional and assessment design, and institutional safeguards. It concludes that the fundamental task of education in the age of AI is not to compete with machines, but to protect those human capacities that cannot be replaced by algorithms, including judgement, meaning-making, and care.