Algorithmic discrimination in public service provision: Understanding citizens’ attribution of responsibility for human versus algorithmic discriminatory outcomes

Abstract

As public bodies increasingly adopt AI technologies in their work, there is simultaneously growing attention to the risk that the reliance on the technology may introduce biases and produce discriminatory administrative outcomes, as demonstrated by multiple real-world cases. Our contribution addresses a core theoretical puzzle: with AI algorithms being increasingly embedded across public services, we lack crucial knowledge about how citizens assign responsibility to public organizations for algorithmic failures and discrimination in public services compared to human discrimination. This speaks to key questions as to whether organizational responsibility attribution mechanisms and public demand for consequences fundamentally change in the context of algorithmic governance. Addressing this gap, we examine whether individual citizens are less likely to attribute responsibility for algorithmic versus human discrimination in public service provision. Building on psychology literature, we further theorize potential mechanisms that underlie these effects and shape citizens’ responses. We investigate these research questions through a pre-registered survey experiment conducted in the Netherlands (N = 2,483). Our findings indicate that public organizations are not held to a lower responsibility standard for algorithmic compared to human discrimination. Technological delegation to AI does not allow public bodies to bypass responsibility for discriminatory outcomes. However, we find that citizens assign more responsibility for algorithmic discrimination when the algorithm is developed in-house rather than externally. This could lead to the emergence of accountability deficits pertaining to technological outsourcing.