You Can't Solve Democracy in Q3: A Critical Analysis of Initiatives to Democratize AI

As AI systems encounter public criticism, technology companies seek ways to increase the perceived legitimacy of their decision-making through methods for public input, such as citizen assemblies, alignment assemblies, and constitutional AI. However, while these methods mimic the language of democratic forms, such as deliberative democracy, we argue that these initiatives are best understood as efforts to obtain legitimacy for AI products without changing the organizing logics of AI: i.e., centralizing power at scale. Through an examination of multiple strands of political philosophy of democracy, we consider the extent to which industry efforts to democratize AI realize these goals, as well as their persistent gaps and tensions. We close by identifying opportunities to deepen democratic governance in AI development.