AI for public good: reorientating fairness, accountability and uncertainty in public sector governance

Abstract

AI is no longer confined to laboratories or private enterprise; it is increasingly embedded within the institutions of public governance. From predictive systems in welfare and justice to large language models shaping policy processes, AI now mediates decisions that affect lives and democratic legitimacy alike. Yet these technologies enter public institutions carrying assumptions, design priorities, and normative trade-offs rooted in their internal logics. This raises urgent questions about how AI can be reconciled with the interpretative norms, procedural fairness, and epistemic commitments that structure public governance. This thesis addresses these questions through an interdisciplinary inquiry into three interconnected domains: fairness, accountability, and uncertainty. While all three are examined, accountability serves as the central analytical axis, with fairness and uncertainty operating as critical supporting perspectives. The contributions of this thesis are threefold. First, it develops conceptual clarity by mapping how dominant AI paradigms embed implicit value commitments that frequently conflict with governance norms. Second, it provides empirical insight into how such systems operate in practice, drawing on case studies and technical experiments involving fairness-aware algorithms and generative models. Third, it advances an institutional analysis demonstrating that alignment with public values must extend beyond predictive performance to encompass the procedural and normative conditions that sustain legitimacy. The central argument advanced is that achieving “AI for public good” requires more than technical refinement. It demands a reorientation of both AI system design and the governance frameworks that oversee their deployment. This includes developing context-sensitive evaluation methodologies, embedding accountability architectures, and recognising uncertainty not as a technical failure but as an intrinsic feature of democratic decision-making. By integrating technical, conceptual, and institutional analysis, this thesis demonstrates that interdisciplinary engagement is indispensable to shaping AI’s role in public life, one that strengthens, rather than undermines, the legitimacy of public institutions.