The Decision No One Authored
The Answerability Gap in Generative AI
The answerability gap names cases where humans remain connectable to an AI-shaped outcome, but no one exercised answerable judgment over the evaluative frame that made the outcome what it was.
Abstract
A person can hold full power to override an AI-shaped decision, be expected to exercise it, and remain answerable for the result, yet never exercise judgment over the evaluative frame the decision turns on. This paper argues that the gap this opens, not the question of machine consciousness, is where responsibility for machine-mediated action is won or lost, and that the norm which closes it is human authorship: the answerable exercise of judgment over a decision's ends, standards, verification, acceptance, and final form.
In plain terms
As AI systems grow more fluent, the people nominally in charge of their outputs face a quiet trap: you can keep full power to override a system, be expected to use it, and remain answerable for the result, while never actually exercising judgment over the frame the decision rests on. The paper names this the answerability gap, and separates it from the more familiar problem of tracing an outcome back to some responsible human. It argues the norm that closes the gap is human authorship: answerable judgment exercised at the five junctures where it is easiest to skip, the ends a system serves, the standards its outputs must meet, their verification, their acceptance into action, and the final form someone stands answerable for. The risk, as systems improve, is a "so-so automation of judgment" that displaces evaluative work without putting anything answerable in its place.
Cite this paper when…
- You need to distinguish being attributable for an AI-shaped outcome from having authored the evaluative frame behind it.
- You're arguing the limits of meaningful human control, human-in-the-loop oversight, or automation bias in generative systems.
- You're analyzing cases where AI preserves formal human approval while hollowing out responsible judgment.
- You're treating answer-engine optimization or data poisoning as a responsibility problem, not just a security one.
Cite
@misc{walton2026answerability,
author = {Walton, Luke F.},
title = {The Decision No One Authored: The Answerability Gap in Generative {AI}},
year = {2026},
month = jun,
howpublished = {PhilArchive},
note = {Preprint, version 1.0},
url = {https://philpapers.org/rec/WALTDN}
}