Testimony Without a Testifier: Epistemic Value and Commitment Ownership in Institutional AI

Abstract

Machine-mediated output can inform, justify belief, guide inquiry, and sometimes acquire the practical force of institutional testimony. Yet the epistemic question whether a hearer may learn from such output is distinct from the speaker-side question of who owns the communicative commitment that the output presents. This paper develops that distinction through the concept of Testimony Without a Testifier: the institutional case in which output circulates with testimonial force while no bearer at the apparent point of address is answerable for the commitment in the relevant speaker-side sense. The argument is compatible with both permissive and restrictive accounts of artificial testimony. A permissive view may allow knowledge from machine output without thereby establishing model-level commitment ownership; a restrictive view may deny that current systems literally testify without implying that their output is epistemically worthless. Building on the Answerability and Empty Occupancy framework in Cott (2026), the paper distinguishes information, evidence, testimonial force, and owned testimony; explains why reliability, fluency, provenance, and conversational continuity do not by themselves identify a testifier; and analyzes the transition from advisory evidence to normatively operative institutional speech. The resulting account is substrate-neutral: a future artificial candidate could qualify as a testifier if the relevant commitments remained attributable to it under challenge, repair, and consequence. The present problem is not that machines are non-human. It is that institutional authority can attach to an apparent speaker before commitment ownership has been secured.