Whose Fairness? Epistemic Injustice as an Intersectional Framework to assess Bias in Government Information Systems

Abstract

Artificial intelligence is increasingly used by governmental organisations to support decision-making processes which have substantial consequences on people’s lives. This highlights the importance of ensuring that such systems are fair and do not perpetuate existing inequalities. While many methods have been proposed to assess the fairness of AI systems, there is no single, universally accepted notion of fairness, nor a definitive way to measure it. This raises the question of whose notion of fairness we operationalize. Drawing on insights from epistemic injustice and and intersectionality, this thesis examines how technical approaches to AI development and fairness assessment can overlook systematic forms of inequality. This analysis is grounded in an empirical analysis of three case studies of algorithmic systems used in governmental contexts. The findings demonstrate that approaches which do not adopt an intersectional perspective risk overlooking those situated at the margins of different axes of disadvantage, thereby contributing to the perpetuation of epistemic injustices. The thesis argues that fairness should not be understood as a single, static property of a system, but as a complex and iterative process that requires critical questions throughout its design, deployment, and evaluation. Based on these insights, the thesis provides a guiding framework and a set of concrete recommendations for AI fairness practice.