Abstract This article examines the challenges of designing AI-enabled welfare allocation systems in public administration, with a focus on the ontological issues that arise in a co-design process. Based on ethnographic fieldwork in a Danish municipality, the study explores how social workers and IT designers struggled to align their respective understandings of a welfare benefit called ‘loss of earnings’. While IT designers sought to structure case work as a predictable, rule-based process suitable for symbolic AI modelling, social workers emphasised the need for discretionary freedom in terms of not only case outcomes but also work processes. This mismatch led to ontological trouble, where fundamental differences in what welfare allocation is—and ought to be—persistently shaped the co-design process. The article highlights that digital transformation in public administration is not just a technical challenge but may also be an ontological one, requiring digital design to navigate multiple, conflicting realities. Second, it demonstrates that co-design does not always lead to mutual learning but may instead expose persistent friction between stakeholders' worldviews. Third, it argues that rather than resolving ontological differences, co-design should create spaces for articulating and negotiating them, helping to safeguard values such as professional discretion and holistic decision-making in AI-enabled welfare services. The findings underscore the need for CSCW scholars and practitioners to engage with the ontological dimensions of AI design critically, ensuring that digital transformation respects the complexities of public sector work.
Christensen et al. (Mon,) studied this question.