Reframing Explainability Through Agency: The Agency Allocation Framework for Educational AI

As AI increasingly shapes what learners practice, how they progress, and which opportunities become available to them, learner agency is becoming both more important and more difficult to design for. Might explainable AI support meaningful human oversight and control by making the ways in which AI shapes learners visible and actionable? In this talk, I introduce the Agency Allocation Framework, which reframes learner agency as the distribution of decision-making authority across learners, educators, institutions, and AI systems. The framework examines who can initiate, influence, veto, or override consequential decisions, how available options and supporting evidence are structured, and over what time horizons those decisions matter. I illustrate this perspective through work on explainable learning analytics and intelligent tutoring systems, including studies with middle school students exploring how mastery visualizations and what-if explanations influence practice strategies, as well as simulation studies that illuminate the implications of different hypotheses about learner decision-making by examining how learner task-selection behaviors interact with system constraints in mastery learning. Across these examples, explainability and agency are inseparable design concerns. I conclude by discussing what it would mean to design explainability not simply to make AI systems understandable, but to create AI-mediated learning experiences that preserve and expand learners’ capacity to direct their own learning.