The Epistemic Substitution Error: Why Accurate Algorithmic Decisions Can Still Be Illegitimate

Institutions increasingly rely on algorithmic systems to assess people, allocate opportunities, and justify consequential decisions. Existing debates often ask whether these systems are accurate, fair, transparent, or explainable. This article identifies a distinct philosophical problem: an institution may reach a factually correct outcome while illegitimately replacing a person's standing to offer relevant reasons with an institutional prediction of what those reasons must be. Through analytical reconstruction, contrastive thought experiments, and normative stress tests, the article develops the concept of epistemic substitution, distinguishes predictive evidence from participatory justification, and proposes a non-substitution principle for decisions affecting persons as accountable agents. It demonstrates that prediction, explanation, and even accurate individualized classification do not by themselves establish that affected persons have been appropriately heard. The resulting framework specifies five conditions for legitimate institutional uptake: access, admissibility, responsiveness, traceability, and revisability. It also distinguishes bounded circumstances in which substitution is permissible from those in which it is not, and derives researchable expectations for public administration, employment, education, and automated peer review. The central conclusion is neither that algorithms should never decide nor that personal testimony should automatically prevail. Rather, decision legitimacy requires institutions to preserve the possibility that an affected person can change the reasons on which a decision is made.