Tools or Crutches? Budgeting human and machine autonomy when introducing GenAI in education

Scholars have defined education as an endeavour to foster autonomy — empowering individuals to think and act independently. This positioning of autonomy at the centre of the learning process comes from several educational approaches, from cognitive to philosophical and political. The pivotal role of autonomy has also been emphasised and extensively studied in psychological approaches to learning and education, becoming an integral value in frameworks such as Self-Determination Theory. The prominent introduction of generative AI (GenAI), however, creates an urgency to (re)assess the role of autonomy in the educational process. While GenAI promises to enhance learning efficiency and accessibility, here we ask: do these tools align or conflict with the overarching goal of cultivating autonomous subjects? In this paper, we lay down a theoretical framework aimed at first, answering that question and second, informing the design of safe educational environments and practices. Our proposal is based on two key assumptions: (i) human educational processes should be understood through a distributed, socio-technical approach in which technologies are active participants; (ii) autonomy in education must be understood from a multidisciplinary perspective in which cognitive, epistemic, and political autonomies of humans interact with the different levels of machine autonomy. These assumptions inform our main contribution: the `autonomy budget', which enables teachers, curriculum designers, and institutional decision-makers to analyze the autonomy trade-offs arising from integrating GenAI in education. We conceptualize the autonomy budget as a zero-sum game of resource allocation, providing a theoretical lens to examine the educational impacts of GenAI adoption and to outline key policy elements for its use in specific pedagogical contexts.