Algorithm or Assembly? An Experimental Study of Recommendations by AI Versus Deliberative Mini-Publics in Democratic Decision-Making

Recent advances in artificial intelligence (AI) renewed debates about who—or what—can legitimately contribute to democratic processes. While both AI and deliberative mini-publics can supplement citizens’ efforts to form judgments about issues, little is known about how citizens evaluate policy advice from these sources. In an online survey experiment with U.S. adults (N = 502), participants read a mock news story reporting a policy recommendation, attributed either to an AI system or a deliberative citizens’ assembly. Opinion climate was orthogonally varied to assess contextual influences. Results were split: citizens’ assemblies were rated fairer and more satisfying, but AI was perceived to be more useful. Opinion climate cues had no moderating effects. Further analysis indicated that acceptance of AI versus citizens’ assembly recommendations depends on individual factors, such as prior attitudes toward the issue and trust in AI. Findings offer theoretical and practical implications for designing democratic decision support systems.