The Question of AI and Democracy: Four Categories of AI Governance

This paper examines the integration of Artificial Intelligence (AI) into democratic governance, focusing on the tension between democracy’s epistemic shortcomings—often manifested as voter ignorance—and AI’s capacity to improve decision-making. Building on the Input-Process-Output (IPO) model, the paper distinguishes AI applications into four categories based on the democratic source of their inputs (i.e., whether they originate from the citizenry) and the binding nature of their outputs (i.e., whether AI decisions carry legal or authoritative weight). Each category—democratic binding AI, undemocratic binding AI, democratic unbinding AI, and undemocratic unbinding AI—is then evaluated against core democratic elements: inclusive and equal participation, quality of decisions, deliberation, and the autonomy of citizens to set the political agenda. While some undemocratic binding AI risks centralizing power into the hands of a few, certain forms of AI, such as AI advisers, AI delegates with deliberative consent, and AI nudger, can enhance democratic processes by helping citizens overcome epistemic barriers, refine their political views, and participate more effectively in governance. The paper concludes that carefully implemented AI has the potential to enhance democratic governance while preserving its core ideals.