This thesis contributes to human-computer interaction (HCI) by exploring how various stakeholders in Swedish public organisations make sense of ethical considerations and negotiate ethical responsibility in the development and use of artificial intelligence (AI)-based systems. 

While high-level ethical frameworks (e.g., guidelines that emphasise principles such as fairness, transparency, and accountability) are intended to guide AI ethics application, prior research reveals that practitioners frequently struggle to translate abstract frameworks into concrete actions within design and use contexts. Responding to calls in HCI for situated, empirical approaches to studying AI ethics in practice, this thesis investigates how stakeholders engage in ethical reasoning through three interconnected dimensions: how they reflect and make sense of ethical considerations, the ethical tensions they encounter when working with AI-based systems, and how ethical responsibility is described and negotiated across AI-based systems’ life cycles.

Drawing on two qualitative case studies combining semi-structured interviews and a multi-stakeholder focus group, the thesis develops an empirically grounded account of stakeholders' ethical reasoning processes. The analysis draws attention to three cross-study themes. First, stakeholders make sense of ethical considerations in situ, shaped by organisational roles, institutional demands, and technological constraints, rather than direct application of abstract frameworks. Second, ethical tensions are not simply obstacles but catalysts that prompt ethical reasoning, surfacing hidden assumptions and conflicts that require stakeholders to renegotiate responsibilities. Third, the negotiation of responsibility is made and remade among actors, shifting across the AI-based system’s life cycle in response to tensions and contextual constraints.

Together, these findings show that ethical reasoning in public sector AI work is best understood as contextual, relational, and evolving – taking shape through the interplay of sense-making, handling tension, and doing responsibilities. In doing so, this thesis invites more reflective (embracing tensions as triggers for ethical reflection), relational (attuned to the shared and negotiated nature of responsibility), and practice-oriented (grounded in the situated ways stakeholders make sense of ethical considerations in everyday work) approaches to Responsible AI.