When officials disagree with AI: public perceptions of bureaucratic discretion in algorithmic governance

While existing research has examined how officials interact with AI systems, less is known about how citizens evaluate the legitimacy of bureaucratic discretion in relation to AI recommendations. This study develops a theoretical framework that contrasts two competing sources of authority, bureaucratic professionalism and algorithmic rationality, and examines the moderating role of AI explainability. Using a vignette-based survey experiment in China within a public recruitment setting (N = 1,150), we find a pronounced asymmetry in public evaluations. Divergence from AI recommendations significantly reduces perceived legitimacy, whereas adherence has no significant effect. AI explainability modestly enhances legitimacy when officials follow AI recommendations. Exploratory analysis further suggests that greater familiarity with AI may strengthen the tendency to privilege algorithmic outputs. The findings suggest that AI establishes a new benchmark for legitimacy judgements, creating new challenges for the professional authority of public officials in the context of algorithmic governance.