Skip to main content
sociable systems.
Episode 273 · 2026-09-30

The Right to Be Answered Back

The whirlwind addresses Job and never answers him. A sociable system stays addressable after it explains, so that 'that does not answer what I asked' reaches someone obliged to reply.

Cover art for episode 273: The Right to Be Answered Back
Account ArcContestabilityGrievance Mechanisms
Episode 273: The Right to Be Answered Back

The voice arrived in wind and fire and filled the house from floor to beam. It asked him where he'd been at dawn. It never asked him what he'd seen.

After thirty-seven chapters of demanding a hearing, Job gets one. "Then the LORD answered Job out of the whirlwind," the King James rendering begins, and what follows is among the most magnificent poetry in the language. Where wast thou when I laid the foundations of the earth? Canst thou bind the sweet influences of Pleiades?

It is overwhelming and beautiful, and it does not address a single thing Job asked.

Job had asked why this had happened to him, and on what charge. The whirlwind replies with the scale of creation and the smallness of the questioner. Job is addressed, at length, by the highest authority available. He is never answered.

Addressed and answered

The distinction matters because institutions have become rather good at the first and remain reluctant about the second.

A claimant can now receive a personalised letter, a portal notification, an automatically generated explanation pitched to the right reading age, and a satisfaction survey. Each of these addresses them. Each is a one-way transmission. The moment the claimant writes back to say "that doesn't answer what I asked," the architecture runs out. There is a form for appeals and a form for complaints. There is rarely any route for saying that the explanation itself missed the point, and almost never a person obliged to receive that reply and respond to it.

The account, in other words, is treated as a delivery. It is sent, and then it is closed.

Hirschman from the other end

What Voice Costs, during the Exit arc, made the strongest available case for Albert Hirschman's voice and found it quietly depended on a listener. Voice works when someone on the other side hears it and can act.

The claimant in October has already exercised voice. They appealed, and the institution listened (four times, truthfully). What they lack is the next turn in the conversation: a respondent who is obliged to reply to the reply. Voice needs a listener to function at all. An account needs one further step, a respondent who stays in the room after explaining, long enough to be told the explanation fell short.

That second turn is where the name of this practice comes home. Sociable, as the site's note on the name puts it, refers to mutuality and to audit: systems that remain answerable to the people they act upon. A sociable system emits an explanation and then stays addressable, so that "no, that's not what I asked" reaches someone with the authority to try again.

What the rulebooks say

The EU AI Act comes close. Article 86 gives a person affected by a decision taken on the basis of certain high-risk AI systems the right to obtain from the deployer "clear and meaningful explanations of the role of the AI system in the decision-making procedure and the main elements of the decision taken." It is a real right, and it attaches to the deployer, which is to say to the institution rather than the tool.

It is also, as drafted, a right to receive. It says nothing about what happens when the explanation arrives and does not explain.

The extractive industries, oddly enough, have a more developed standard for this than most AI law. Principle 31 of the UN Guiding Principles on Business and Human Rights sets out the effectiveness criteria for grievance mechanisms, and alongside legitimacy, accessibility, predictability and the rest, it asks that operational-level mechanisms be "based on engagement and dialogue." A grievance mechanism that only issues determinations fails that test. The standard assumes the complainant will talk back, and that the company will be there when they do.

Anyone who has run a grievance log knows what this looks like in practice. The first response rarely closes the matter. The second conversation, the one where the complainant explains what the first response missed, is usually where the grievance is actually understood.

Designing the second turn

A system built for the second turn needs some unglamorous things. A channel through which the claimant can contest the account itself, separate from contesting the outcome. A named person who receives that contest, with authority to reopen the reasoning. A record showing that the reply was read, and what was done with it. And a time limit, because an institution that answers back eventually has not answered back.

None of this requires the model to do anything differently. It requires the institution to stay in the conversation after the model has left it.

The whirlwind, having finished, departs. Job is left with an experience of overwhelming authority and his question intact. He responds with humility (the text is ambiguous on whether he is persuaded or merely outmatched), and there is no one left to hear whether that was the end of it.

Companions


These notes come out of Sociable Systems, a practice that reads AI-shaped documents the way a hostile reviewer will, before a lender or a court finds the gap. The argument has an operational form: the Interim Protocol sets out four rules for AI use in environmental and social deliverables, covering disclosure at touch-point grain, evidence custody, the phrases no automated screening may settle, and a hostile read before anything ships. Free, and written to be cited or retired once institutional guidance arrives.