Standing: who gets to contest a machine

Every organization that has ever made an automated decision has, at some point, been told it should be accountable for it. The usual response is to build a record. Logs are kept, audit trails are wired up, a compliance team is given a dashboard, and the institution can now say, truthfully, that every decision is documented. And yet nothing has changed for the person on the receiving end of those decisions — because a record that only the institution can read, interpret, and act on is not accountability. It is bookkeeping. The decision is just as final, and just as unquestionable, as it was before the logging existed.

What is missing is a contestant. A record exists to be challenged by someone, and if there is no one with both the right and the means to challenge it, the record is set dressing — proof that the institution takes itself seriously, addressed to itself. The law has a precise word for the thing that turns a grievance into a case: standing. You cannot bring a complaint to court merely because something offends you. You must show that you were affected, and that the court can do something about it. Standing is the hinge between caring about an outcome and being able to do anything about it, and it is exactly the concept automated accountability has been missing.

Who counts as having standing

Standing to contest a machine belongs, first and most obviously, to the person the machine acted upon. The applicant who was denied. The traveler who was rebooked. The account holder who was flagged. The worker whose performance was rated. These are not stakeholders in the diffuse sense the word has come to mean; they are the affected party in the strict legal one — the decision changed their circumstances, and so the decision is, in a real sense, partly theirs to interrogate.

But naming who should have standing is the easy half. Standing is not a sentiment; it is a capacity. Borrowed from law, it has two components that have to be present together. The first is the right to contest — a recognized claim that this decision is yours to question. The second is the practical means to do so — actual access to what the decision was based on, in a form you can use. A right without means is the more common failure, and the more corrosive, because it looks like accountability while delivering none of it. You are told you may appeal, and then handed nothing you could appeal with.

A right to contest without the means to contest is not a right. It is a permission slip for a door that stays locked.

This is what standing demands of the record itself, and it is a harder demand than "auditable." A record can be perfectly auditable by the institution that built it — every field populated, every event timestamped — and still be useless to the person it concerns. Auditability serves the auditor. Standing requires that the record be legible and challengeable by a non-engineer: the affected party, reading it cold, must be able to see what evidence was in front of the decision, which rule connected that evidence to this outcome, and where they think it went wrong. A record only the builder can read keeps the builder in sole possession of the truth. That is the opposite of standing.

The asymmetry, and how it flips

The reason standing has been so rare in practice is an asymmetry of effort. Today, contesting an automated decision is expensive for the weakest party in the exchange and nearly free to refuse for the strongest. The affected person has to notice the decision, formulate an objection, find the channel, file the request, wait, escalate, and assemble — usually from nothing — some account of what they think happened. The institution has only to decline, or to delay until the person gives up. The cost of the question falls on whoever can least afford it, and the cost of stonewalling falls on no one.

A self-describing decision record flips that. When a decision carries its own account — the evidence it weighed, the rule it applied, enough state that the result can be reconstructed and checked — the affected party no longer has to manufacture the case against it. The material is already there, produced as a byproduct of having made the decision properly. The expensive part of contesting, the part that priced out everyone without a lawyer, has been pre-paid by the institution at the moment of decision rather than extracted from the complainant afterward. Standing stops being something you have to fight to establish and starts being something the record confers.

And the refusal flips too. An institution that has nothing to produce when asked no longer looks discreet; it looks evasive, because everyone now knows the record was possible to make. The asymmetry does not vanish — the institution will always be larger than the person — but the burden of having an answer shifts onto the party that made the decision, which is exactly where, on any honest account of fairness, it belonged all along.

The hollow version

There is a way this goes wrong that is worth naming plainly, because it is the path of least resistance and it will be taken often. Contestability can be hollowed out into a complaint form. The institution adds a button. The affected party clicks it, types into a box, and receives, some days later, a courteous message confirming that the original decision stands. Standing has been simulated: there is a channel, there is a record of the objection, there is even a response. What there is not is any mechanism by which the challenge could actually move the outcome. The contestant has been given a stage and no script.

Real standing means a challenge can change the result. Not always in the challenger's favor — a well-founded decision should survive being contested, and surviving scrutiny is itself part of what makes it legitimate. But the channel has to be live. There must be some reachable state of the world in which the affected party is right, presents the evidence, and the decision is reversed, corrected, or reopened on the merits. If no such state exists — if the outcome is fixed before the contest begins — then the contest is theater, and we are back where we started, with a more elaborate set. Accountability is measured not by whether a record exists, nor by whether an objection can be filed, but by whether the right person, holding the right evidence, can actually make the machine answer.

— Dispatches · Summit Cognitive