Beyond Individual Accountability: (Re-)Asserting Democratic Control of AI

AI control mechanisms like accountability procedures or technical standards are usually subpolitical: decisions are primarily debated and made within circumscribed subsystems of experts or interest groups, like the professional community of data scientists. However, AI systems are more deeply intertwined with a wider sense of politics than these mechanisms contemplate. In Winner's dual senses, they are incidentally political as they settle disputes within political communities through their design, invention, and arrangement, and inherently political as they reciprocally contribute to and are sustained by patterning of economic, social, and political orders. This work, therefore, draws upon political theory to argue for democratically controlled AI beyond individual notions of accountability. In its weaker form, it demands substantive, rule-bound oversight of state actors' use of AI systems, seeking to remedy historical tendencies toward extra-legal surveillance and strengthen accountability beyond individuals. Conversely, the stronger form advocates for comprehensive democratic control over all facets of AI, even by questioning the permissibility of AI within particular socio-economic spheres, as these systems are becoming fundamental parts of our collective life. I sketch the necessary institutional frameworks to operationalize these two forms of democratic control: first, for the "weak" form through the concept of a "control" power separate from the executive from Sun Yat-Sen's political thought, and second, participatory institutions such as citizens' assemblies. Finally, I discuss actions data scientists can take without legal frameworks for control: furthering new social imaginaries of AI that foreground the possibility of control and involving affected communities in decision-making around AI systems. The concept of democratic control is then both a measuring stick for existing standards and legislation and a clarion call for future advocacy.