Epistemic Justice and the Moral Limits of Algorithmic Authority: How AI Systems Challenge the Knowledge Foundations of Organizational Decision-Making

Abstract

The deployment of artificial intelligence systems in organizational decision-making has introduced a form of epistemic injustice that has received insufficient scholarly attention: the systematic undermining of human beings' capacity to contribute to, challenge, and shape organizational knowledge through the algorithmic delegation of epistemic authority. Drawing on Miranda Fricker's concept of epistemic injustice and its recent extensions to AI contexts, this paper argues that when organizations rely on AI systems to generate, validate, and distribute knowledge—including knowledge about their own employees, customers, and operations—they risk committing distinctive epistemic wrongs against the human agents whose understanding, experience, and judgment are displaced or devalued. The paper is grounded in three foundational references: Wang's MIRROR benchmark revealing systematic metacognitive miscalibration in large language models, Wang's Verification Tax analysis exposing the structural resource constraints that prevent adequate AI auditing, and Bei et al.'s framework for strategic human resource analytics with explainable AI. These references illuminate the intersection between AI's technical limitations and its moral consequences, demonstrating that the calibration failures and auditing gaps documented in the technical literature translate directly into forms of epistemic harm. The analysis identifies two primary mechanisms of AI-driven epistemic injustice—testimonial injustice through algorithmic credibility discounting and hermeneutical injustice through the imposition of computational meaning structures—and examines their operation across organizational domains including human resource management, strategic decision-making, and knowledge production. The paper concludes that addressing epistemic injustice requires not only technical improvements in AI explainability and calibration but also a fundamental reorientation of organizational AI governance toward epistemic justice as a core design principle.