AI Personalized Learning and the Risk of Epistemic Consolidation

Abstract Many researchers and industry leaders hold that AI personalized learning systems are about to unleash a better, more flexible, and richly diverse form of instructional support than was experienced by any previous generation of learners. In this essay I grant that AI personalized learning systems may diversify curriculum, instruction, and assessment experiences in a typical classroom. Nevertheless, I argue that efforts to personalize learning by shifting diverse human teachers' work into a smaller number of scalable AI instructional systems can result in a transfer and concentration of human teachers’ previously diffuse and limited pedagogical power over a population’s instructional experiences. I call this process of concentrating power over a population’s instructional experiences the epistemic consolidation of teaching. Highly scalable epistemically consolidated AI teaching systems—capable of shaping (or mishaping) what the public thinks, does, and feels in highly personalized ways—raise new and unique social, political, and epistemic risks. Among these risks, I argue, is a threat to the epistemic independence and, consequently, the epistemic agency of the public. To avoid this risk, I argue that policy and design strategies that guide AI personalized learning approaches towards non-dominating decentralized democratic control of education are of critical importance.