Resolving value conflicts in public AI governance: A procedural justice framework
本文提出一个程序正义框架，区分衍生信任标准与基本民主价值冲突，通过瑞典就业服务AI案例展示如何平衡效率、可解释性与合法性，为公共机构提供结构化治理方法。
This paper addresses the challenge of resolving value conflicts in the public governance of artificial intelligence (AI). While existing AI ethics and regulatory frameworks emphasize a range of normative criteria—such as accuracy, transparency, fairness, and accountability—many of these values are in tension and, in some cases, incommensurable. I propose a procedural justice framework that distinguishes between conflicts among derivative trustworthiness criteria and those involving fundamental democratic values. For the former, I apply analytical tools such as the Dominance Principle, Supervaluationism, and Maximality to eliminate clearly inferior alternatives. For the latter, I argue that justifiable decision-making requires procedurally fair deliberation grounded in widely endorsed principles such as publicity, inclusion, relevance, and appeal. I demonstrate the applicability of this framework through an indepth analysis of an AI-based decision support system used by the Swedish Public Employment Service (PES), showing how institutional decision-makers can navigate complex trade-offs between efficiency, explainability, and legality. The framework provides public institutions with a structured method for addressing normative conflicts in AI implementation, moving beyond technical optimization toward democratically legitimate governance. • Proposes a framework for resolving value conflicts in public AI governance. • Distinguishes between derivative and fundamental value conflicts in AI. • Applies procedural justice principles to incommensurable value trade-offs. • Demonstrates framework through a case study of the Swedish PES AI system. • Supports democratic legitimacy via structured deliberation and auditability.