Cybernetic and Epistemic: A Missing Vocabulary for Trustworthy Agentic Delegation
Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 1058-1067
DOI: 10.1609/aaaiss.v10i1.43527 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Jérémie Lumbroso.
Abstract (Originalsprache)
As code generation is increasingly delegated to AI systems, the bottleneck in software production is shifting from writing code to supervising the systems that write it — a shift computing-education researchers have already begun to name. We argue this shift exposes a vocabulary gap with direct consequences for trustworthy-agentic-systems research: the field asks for "human oversight" without a working distinction between the two things language does in a delegation channel — coordinate action (cybernetic: the words succeed when the world comes to match them) and coordinate understanding (epistemic: they succeed when they answer to the world and a hearer can check that they do). The failure this names is not cybernetic language but epistemic-form language doing cybernetic work: explanation-shaped output calibrated for approval rather than truth. Oversight that checks only whether an output was approved is satisfiable by rubber-stamping and cannot see this failure; oversight that holds an agent accountable requires that the reasoning behind its work be retrievable and checkable, not merely assertable. We present three delegation episodes — illustrations, not controlled evidence — in which epistemic engagement proved practicable at working pace while remaining auditable, one public record in which a recommendation was withdrawn on its own stated terms, and one failure case illustrating the limits of oversight that requires no reasons for the discretionary choices it counts, spanning both directions of a delegation channel. We propose a criterion for agentic-system governance, alongside — not instead of — existing technical trust properties: that every consequential choice in a delegation carry the condition under which it would have gone otherwise, in a form a third party can test. A decision that carries no such condition, stated or reconstructible, gives a third party no way to tell it from a rubber stamp. We give the criterion an operational form — a two-part reconstruction test that scores a delegation record by whether a second reader can predict from it what the agent does under a perturbation — and a deliberation-recording convention, ORRCF, that makes the condition a required component of every recorded choice.
Zitieren
APA
Lumbroso, J. (2026). Cybernetic and Epistemic: A Missing Vocabulary for Trustworthy Agentic Delegation. Proceedings of the AAAI Symposium Series, 10(1), 1058-1067. https://doi.org/10.1609/aaaiss.v10i1.43527
Vancouver
Lumbroso J. Cybernetic and Epistemic: A Missing Vocabulary for Trustworthy Agentic Delegation. Proceedings of the AAAI Symposium Series. 2026;10(1):1058-1067. doi:10.1609/aaaiss.v10i1.43527
BibTeX
@article{Lumbroso2026,
author = {Lumbroso, Jérémie},
title = {Cybernetic and Epistemic: A Missing Vocabulary for Trustworthy Agentic Delegation},
journal = {Proceedings of the AAAI Symposium Series},
year = {2026},
volume = {10},
number = {1},
pages = {1058--1067},
doi = {10.1609/aaaiss.v10i1.43527}
}
EUVIMED