Constraining Unaccountable AI with the Militant Pragmatism of Everyday Institutions

I assess Ayşe Aslı Bozdağ’s account of AI as political infrastructure, situating it at the intersection of Bratton’s Stack theory and Dussel’s ethics of exteriority. Bratton supplies a systems-level diagnosis of how AI governs through layered forms of power, while Dussel provides a normative standpoint to judge this governance, as the lived experience of persons excluded or harmed by dominant systems. Bozdağ’s originality lies in translating this dual framework into a political ethics of contestation through “valves” that interrupt system-wide closures by enforcing opacity, hesitation, redistribution, and proximity, thereby designing AI to remain deliberately partial and uncompleted. I argue that this approach under-specifies mechanisms of implementation, agents of change, and existing sites of resistance. In response, I propose a militant pragmatism that leverages adversarial litigation, prescriptive regulation, antitrust intervention, and labor organizing to constrain unaccountable AI through existing institutions. Such institutional counter-power is a necessary precondition for making Bozdağ’s more radical design-based interventions politically feasible at scale.