Against Disembodied AI Authority: A Counter-Argument in Defense of Human Testimony, Origin, and Epistemic Justice

A growing assumption underlies modern AI deployment: that probabilistic, machine-generated outputs may be treated as authoritative even when severed from human origin, lived context, and examinable testimony. This position paper rejects that assumption as epistemically invalid and institutionally dangerous. Drawing on foundational legal principles—including the Confrontation Clause (Crawford v. Washington), expert testimony standards (Daubert), and administrative law requirements for reasoned decision-making—this paper argues that disembodied AI outputs violate the basic requirements of legitimate authority: origin, accountability, and responsibility. The paper demonstrates that AI outputs used in sentencing, parole decisions, administrative proceedings, and judicial contexts function as testimonial statements yet cannot be cross-examined. It argues that courts and institutions must refuse to grant evidentiary weight to outputs that lack examinable origin, and that the Wisconsin Supreme Court's decision in State v. Loomis (upholding COMPAS risk assessment) represents a dangerous acceptance of disembodied authority. This is not an argument against AI. It is a principled defence of the foundational standards that make judgment legitimate: no origin → no weight; no witness → no authority. The paper provides clear standards for the proper role of AI outputs (advisory, not determinative) and establishes that human authority is not a performance problem to be optimized away—it is the foundation of justice.