From Diagnosis to Remedy: Epistemic Justice as an Aim of AI-Mediated Learning

A companion paper (Basal, 2026) diagnosed four forms of epistemic injustice in AI-mediated learning: credibility displacement, hermeneutical narrowing, access asymmetry and formative injustice. A diagnosis, however, does not yet say what justice would require. This conceptual paper takes up that second task. I argue that epistemic justice should be treated as an explicit aim of AI-mediated learning, not only as the absence of injustice, and I develop its content through three contributions that the companion paper did not contain. First, I confront a problem the diagnosis left open, which I call the necessity paradox: for learners who depend on AI because they have no alternative, restricting AI may itself deepen injustice, yet unrestricted reliance exposes them to every dimension at once. I resolve the paradox by locating the injustice in the absence of alternatives rather than in the use of AI, and by distinguishing a comparative from an absolute standard of justice. Second, I refine the account of credibility displacement by recasting its self-directed form as an erosion of epistemic self-trust, which separates it from testimonial injustice proper and clarifies when it is unjust. Third, I propose four correctives matched to the four dimensions: credibility calibration, hermeneutical widening, comparative transparency and formative protection. I then specify these correctives as virtues of teachers, as virtues of institutions and as obligations of system designers, following Fricker's virtue of testimonial justice and Anderson's account of epistemic justice as a virtue of social institutions. The resulting account gives educators and policymakers a positive standard: AI-mediated learning is epistemically just when it leaves every learner better placed to know, to interpret and to judge for themselves than the realistic alternatives would.