The decision nobody made

We worry about the wrong failure. When people imagine automated systems going wrong, they picture a bad decision — the model that denies the deserving applicant, the algorithm that flags the innocent traveler, the score that misjudges the careful borrower. Those failures are real, and they are correctable, because a bad decision still has the shape of a decision: there was a choice, it can be examined, and it can be reversed. The failure that should worry us more is quieter and harder to fix. It is the decision that no one made — the consequential outcome that emerged from a chain of systems without any identifiable person having decided it.

Consider how a modern decision actually gets produced. A signal is generated by one system, scored by a second, thresholded by a third, routed by a fourth, and acted upon by a fifth, each operating correctly within its own narrow mandate. No single component decided the outcome; each merely passed its output downstream. The applicant is rejected, the account is frozen, the shipment is held — and if you ask who decided that, you will find that the honest answer is no one. Every person in the chain can truthfully say they did not make the call. The call was made by the chain, and a chain cannot be held to account, because a chain cannot be summoned, questioned, or asked to explain itself.

This is the condition I want to name precisely: not the unjust decision, but the unauthored one. An unjust decision has an author who can be wrong and corrected. An unauthored decision has no author at all, and that absence is not a small administrative gap. It is the dissolution of the entire structure by which we hold power responsible. Responsibility presupposes an agent. Remove the agent and you have not made the decision more objective; you have made it unanswerable.

An outcome that no one decided is not a neutral outcome. It is a decision with the accountability quietly engineered out of it.

Diffusion is not neutrality

The seductive thing about a diffused decision is that it feels fairer than a personal one. No biased official, no arbitrary gatekeeper, no human caprice — just systems doing what systems do. But this confuses the absence of a visible decider with the absence of a decision. The decision still happened; someone still bears its consequences. What has changed is only that the authorship has been spread so thin across the chain that it has effectively vanished. And a decision whose authorship has vanished is not a fairer decision. It is one whose author has been hidden, often without anyone intending to hide anything.

That is the part worth sitting with. Unauthored outcomes are rarely the product of bad faith. No one in the chain set out to escape accountability; each just built or operated a component that did its job. The evaporation of responsibility is an emergent property of the architecture, not a conspiracy within it. Which makes it more dangerous, not less — there is no villain to remove, no single bad actor whose dismissal fixes the problem. The accountability gap is structural, and structural problems do not respond to indignation. They respond only to design.

Restoring the author

So the work is not to slow automation down or to insist a human personally make every call — that is neither possible nor, in most cases, desirable. The work is to ensure that every consequential outcome, however many systems it passed through, can be traced back to a person or an office that authorized it and can be made to answer for it. Not a person who pulled each lever by hand, but a person who stands behind the result — who put the chain into operation knowing what it would do, and who therefore owns what it does.

This is what a signature once accomplished, before automation quietly removed the signature line. A signature was never a claim that one person performed every step; it was a claim that one person would answer for the whole. Who signs is not an antiquarian question. It is the load-bearing question of the entire enterprise. A decision that carries no signature carries no author, and a decision with no author is, by construction, a decision no one can be held to account for.

The instrument that closes the gap is a record that binds the outcome to its origin — one that carries not only what was decided and on what evidence, but under whose authority, against which rules, and traceable through which chain. When an automated outcome comes with a record that names the authorizing party and reconstructs the path that produced it, the decision has an author again. The person it affected can find the someone who must answer; the someone who must answer cannot dissolve into the architecture. The chain becomes legible, and a legible chain has a beginning, and a beginning has a name.

The phrase "the system decided" was, for a long time, a sentence that ended conversations — an answer that no further question could get past. It should be the opposite: the moment the real questioning begins. Because no system decides anything. Systems execute decisions that people authorized, and the only question that ever matters when something goes wrong is which person, behind which authority, will answer for what the system did. If that question has no answer, the failure is not in the model or the data or the threshold. It is in a structure that produced a consequence and arranged, however innocently, for no one to have made it.

— Dispatches · Summit Cognitive