That sentence: trust is the reason a person agrees to learn from another, and it has always depended on the conversant being accountable to something beyond the exchange — a community, a body of prior commitments, a record that precedes the conversation and survives it. Lankes presents this as the field’s next open problem. It is not open. It is a problem with a hundred-year literature attached, the warrant tradition.

Warrant entered our field as a design principle. Hulme argued in 1911 that classification should stand not on a philosophical map of knowledge but on the actual literature to be organized, literary warrant. Beghtol showed that every knowledge organization system answers, implicitly, the question on what authority do these categories stand? in the literature, in scientific consensus, in use, in culture. Structurally, warrant is precisely Lankes’s trust formula applied to systems: an accountability to something beyond the system itself. And for trust in persons rather than documents, Patrick Wilson’s cognitive authority did the same work, for track record, institutional position, answerability to a community of assessors. Lankes says interrogating an exchange is like interrogating a person, “a very different act.” Quite so. But the field has a theory of that act. What it lacks is an extension to nonhuman interlocutors, institutional warrant, where trust is grounded not in item or pseudo-person but in the arrangements under which a system is trained, deployed, and held answerable. I develop this argument at length in a manuscript now under submission to the Journal of Documentation.

The second amendment goes deeper, to the metaphysics under the shock. Lankes’s structure requires a before (documents were stable, inspectable, returnable-to) and after (the exchange dissolves that unit and leaves nothing). But the document never possessed its stability intrinsically. Here I borrow an instrument from Buddhist philosophy that I have long argued belongs in the philosophy of information: nothing exists with self-standing essence; whatever exists, exists dependently, held in place by the durable web of conditions that sustains it. The document’s stability was always of this kind, conferred by legal deposit, publication, cataloging, citation, the archive, the shelf. Briet’s antelope, via Buckland, becomes a document when it is captured, tagged, and placed in a zoo where someone can return tomorrow and find the same animal. What confers the returnability? The zoo, as institutional warrant materialized. So AI has not let the antelope out of the zoo, and it has not destroyed an intrinsically stable unit, because none ever existed. It has produced a new animal. The difference matters because it is the difference between grief and planning. If stability was a natural property now lost, the field can only mourn and retreat defensively to bluebooks, writing rooms, and walled-off archives. If stability was always an institutional achievement, then we build the institutions that confer stability, at a new level, on a new object, in a framework of dependencies not threatened by the discovery that knowledge is transactionally constituted. For example, Lankes’s own use of Conversation Theory already says knowing arises between conversants.

Thirdly, the claim that the exchange leaves “no unit to catalog, retrieve, cite, or verify” is true at one level of abstraction and false at others, where the new tools live. At the item level, Lankes is right: there is no datable source, no author to check. Item-level provenance fails, and the University of Virginia archival AI protocol he analyzes is item-level warrant fighting for its life. But ascend a level and observables reappear. At the system level are training provenance practices, alignment commitments, evaluation records, behavioral regularities auditable in aggregate even though no single output repeats. A stochastic system is not an inscrutable one, since its reliability is a statistical and institutional property. We already warrant clinical trials and weather models without offense. At the next, institutional level, we find who deploys the system, under what governance, answerable to whom, and with what enforceable commitments. Institutions, unlike outputs, can hold to a prior word. This answers Lankes’s own question about what succeeds information retrieval: warrant-conferring instruments, with registries of systems and commitments rather than documents. Audit regimes make regularity inspectable the way the catalog made holdings inspectable. Now we have accreditation of conversants, as we accredit the professionals and institutions whose epistemic goods were always transactional. Reference itself was always an unrepeatable, undocumented, co-constructed exchange, warranted not item-by-item but by training, institutional standing, and community oversight. So the transactional mode is not alien territory, but where librarianship began.

And so the final amendment is on identity. Lankes says the machine has taken the mediating seat the field held since Alexandria. I would say the field never truly held that seat. The document did not know for anyone, and on our own best theory the reference librarian did not transmit knowledge but arranged the conditions under which an inquirer built it. What the field has always held is the warranting seat: the accountable arrangement of conditions for trustworthy coming-to-know. That seat is not lost, for the reason Lankes himself gives: the system answers to no community and cannot warrant itself. Something must stand behind the exchange, accountable for its conditions, and it is our seat.

One might ask who warrants the warranters, when the dominant institutions of the transactional mode are the AI providers themselves. A registry maintained by the registrant is a non-starter. The answer is plurality. Warrant has never been self-certification. Rather it has been conferred by scholarly ecologies, with publishers checked by libraries checked by professional bodies checked by law. The urgent danger is not that warrant is impossible for the exchange but that the warranting ecology is being enclosed along with the commons. Building warranting institutions independent of the systems they warrant is the field’s advocacy program, not its afterthought.

Thus the trust Lankes reaches for has a name in our own literature, and a genealogy built for grounding reliance when the item cannot be interrogated. The stability he mourns was institutionally conferred, which means it can be institutionally re-conferred. As the document dissolves, its dependence remains, and warrant is the craft of making dependence trustworthy. For the working information professional, that craft is already actionable: advocate for audit and disclosure regimes, help build registries and accreditation independent of the providers, and treat the vetting of conversants as reference’s successor. It has been the work since the first tag went on the first antelope. The exchange is simply the next animal.

Ken Herold is Head of Technology and Digital Strategies and library faculty at California State University, Los Angeles. His manuscript on institutional warrant in AI-mediated information environments is under submission to the Journal of Documentation.

Author’s note on AI use: The author used Claude Fable 5 (Anthropic) for drafting and editorial assistance in developing this response from the author’s existing published and unpublished frameworks, including condensing a longer draft to this length. The author evaluated, edited, and accepted or rejected all suggestions; the argument, positions, and conclusions are the author’s own, and the author takes sole responsibility for the content.