Public sector adoption of AI is frequently framed as a step toward more efficient and effective service provision. Commonly framed as decision-support tools, these systems do more than assist officials. They influence how decisions are prepared and structured before any formal judgement is made. This paper argues that such pre-structuring alters the architecture of administrative decision-making in ways that current governance and ethics frameworks fail to capture, with direct implications for democratic legitimacy.
To analyse this shift, the paper introduces the concept of delegated autonomy. Delegated autonomy describes situations in which public officials retain formal authority and responsibility while algorithmic systems shape significant elements of the decision process by preselecting available options. Under these conditions, those who shape outcomes are no longer identical with those who are held accountable for them. This misalignment poses challenges for transparency and for citizens’ ability to understand and contest administrative decisions.
The paper further argues that the way officials rely on AI recommendations cannot remain a matter of individual judgement. Instead, institutional arrangements must define how algorithmic inputs are integrated into decision processes, when they may be overridden, and how such actions must be justified. This framework helps explain how the gradual erosion of human decision space challenges democratic legitimacy in AI-supported administration.