Artificial Intelligence, Bureaucratic Discretion, and Democratic Administration

Since COVID-19, governments have accelerated efforts to replace civil servants with AI. This article argues that this rapid substitution risks undermining a core commitment of democratic administration: accountable, reason-giving bureaucratic discretion. Drawing on cases of automated eligibility, sanctioning, and standardization, it shows that algorithmic tools govern through statistical regularities, which can obscure responsibility, mishandle atypical cases, and weaken citizens’ claims to individualized consideration. Building on classic public administration debates and newer work on system-level bureaucracy, the article reframes AI adoption as a constitutional, not merely technical, question: which administrative decisions may be standardized, and which must remain under human control because they implicate fairness, equity, or rights? The article concludes with a selective ``centaur’’ model in which AI supports but does not displace human officials, preserving explanation, accountability, and appeal.