Keith S. Brown has posted The Citizen in the Room: AI and the Architecture of Freedom on SSRN. Here is the abstract:
AI-assisted government action represents the largest expansion of state decisional capacity in history, with substantial potential for state abuse. Prior expansions of American governmental power — the administrative state’s exercise of arbitrary authority, the surveillance state’s systematic violations of privacy and free speech — generated harms that existing institutions could not fully check. AI-assisted decision-making is different in kind, not merely degree: it operates at speeds and scales that make retrospective oversight structurally inadequate. No congressional committee, inspector general, or court can be present at every consequential AI-assisted decision. The AI system itself can. AI systems possess a structural advantage no prior oversight institution has held: for the first time, the oversight function can be embedded within the decision itself. An AI system is simultaneously the instrument of state power and the automatic recorder of government abuse — generating a tamper-proof signal at the moment of every consequential decision. This changes the accountability calculus fundamentally. Prior oversight institutions detected violations after the fact. Embedded oversight deters them before. Yet harnessing this advantage requires solving a problem existing institutions have not addressed: the government that deploys the AI also controls who trains it, and controls whether the signal it sends can be suppressed. This paper proposes an architecture that solves that problem and meets the central challenge of AI constitutional design.