Epistemic Injustice in AI-Mediated Learning: A Learner-Centred Conceptual Framework

Generative artificial intelligence (AI) is rapidly becoming a primary source of explanation, feedback and knowledge for learners. Philosophical accounts of epistemic injustice in generative AI have so far addressed the information ecosystem at large, while educational accounts have examined the formative wrong of delegating educative tasks, the quality of learners' dependence on AI, or particular forms of harm. What is still missing is an account that connects these wrongs, states when they are injustices and not merely harms, and asks how they are distributed across learners who are positioned unequally. This conceptual paper argues that AI-mediated learning produces forms of epistemic injustice that are specific to education. Drawing on Fricker's account of testimonial and hermeneutical injustice, on Nikolaidis's conception of formative epistemic injustice and on the philosophy of epistemic dependence, I propose a learner-centred framework with four dimensions: (1) credibility displacement, (2) hermeneutical narrowing, (3) access asymmetry and (4) formative injustice. I specify criteria that distinguish an injustice from a mere harm, argue that all four dimensions are amplified when learners depend on AI out of necessity rather than choice, and show that the dimensions interact in ways that make single-point interventions insufficient. I situate the framework against existing accounts, identify who bears responsibility for addressing these wrongs, answer four objections, derive six falsifiable propositions with indicative operationalisations, and set out implications for pedagogy, assessment, teacher education, curriculum and policy. The framework reframes the educational question about generative AI from whether learners should use it to whether AI-mediated learning treats all learners as equal knowers in the making.