Unowned Reasons: Generative AI And the Erosion of Public Answerability in Democratic Judgment

Abstract

This thesis argues that the central democratic threat posed by generative AI is not misinformation but the erosion of public answerability, the structural condition that warrant-bearing reasons in binding institutional decisions have an identifiable agent obligated to defend, revise, or withdraw them. Where the dominant policy and scholarly responses ask whether AI-generated content is accurate, this thesis argues that accuracy is structurally insufficient. A perfectly accurate AI-generated policy rationale that enters a binding decision without an answerable owner is democratically deficient. The scope is restricted to high-stakes institutional discourse, including government agencies, regulatory bodies, legislative staff, and policy-adjacent ecosystems where reasons function as warrants in binding decisions.

The argument develops across four chapters. Chapter 2 extracts diagnostic instruments from Dewey (the public, inquiry, and the eclipse), from Anscombe and Dennett (the interpersonal grammar of reasons that names the action-theoretic conditions under which one person can be asked to defend their reasoning), and from Arendt (action, plurality, the space of appearance, and worldliness). Chapter 3 maps the traversal pipeline through which social media moves unowned reasons from low-stakes circulation into institutional proximity. Chapter 4 diagnoses how generative AI industrializes warrant-form at scale, producing both structurally unownable reasons and answerability theater, the condition in which signatures and procedures persist while the substantive capacity to defend is absent. Chapter 5 walks a five-link normative chain from democratic legitimacy to reason-ownership and derives an institutional repair architecture: authorization-first responsibility, five accountability rails operating at the point of uptake, and an exclusion rule that applies where capacity for substantive answerability cannot be supplied.

The thesis offers three contributions: a conceptual reframing that locates the democratic threat in answerability rather than accuracy; a diagnostic toolkit of nine interrelated concepts, an uptake ladder, and four capacity criteria for identifying where institutional ownership fails; and a normative architecture derived from democratic first principles through a chain with no skipped gears.