The ongoing and decades-long digitalisation of educational practices is largely motivated by the assumption that learning is best understood in terms of information processing. AI tutors and other educational AI tools leverage and accelerate this shift by presenting abstract, symbolic interactions as enhancing learning. Yet, while such AI implementations present as human-like, they are detached from physical, social, and ethical contexts and in fact herald a further shift towards disembodied lea…
Read moreThe ongoing and decades-long digitalisation of educational practices is largely motivated by the assumption that learning is best understood in terms of information processing. AI tutors and other educational AI tools leverage and accelerate this shift by presenting abstract, symbolic interactions as enhancing learning. Yet, while such AI implementations present as human-like, they are detached from physical, social, and ethical contexts and in fact herald a further shift towards disembodied learning. In this critical review, we develop a fine-grained conceptual framework through which to examine AI in education and suggest ways to avoid further disembodying learning. To do so, we first examine the proliferation of AI in education. Second, we situate four interaction frames that AI affords through being assigned different roles, like with AI math tutor Khanmigo. Third, we present four aspects that educational AI operates on: sensorimotor, interpersonal, epistemic, and moral frictions for learning. Fourth, we advance a human-centred approach that aims to re-embody AI in education: namely, by outlining research and policy agenda's that enable educators and learners to flourish through meaningful AI interventions on the educational frictions examined. Finally, we conclude on the wider societal implications of these developments for the role of Big Tech, future research, and the development of AI literacy courses as required by the EU AI Act.
