The verdict without a trial

Something is done to a person — an account frozen, a benefit stopped, an application denied, a name added to a list that closes doors. It happens instantly, cleanly, without warning, and by the time the person learns of it the matter is already settled. No one told them it was coming. No one asked what they had to say. There was no moment at which they could put their side of it, present a fact the system did not have, or argue that the rule should not apply to them. The outcome simply arrived, complete, like a verdict handed down by a court that never convened. And the strangest part is that the outcome might be entirely correct. That is precisely what makes it worth thinking hard about — because a verdict without a trial is troubling even when it is right, and understanding why is the whole point.

We have a deep intuition, older than any particular legal system, that being subjected to a consequential adverse action entitles you to a proceeding — a chance to be heard before the thing is done, or at minimum a real chance to contest it after. This is not a technicality. It reflects a belief about what it means to treat someone as a person rather than as an object to be sorted: that they are owed the dignity of participating in a decision that falls on them, of being a party to it rather than merely its target. Automated adverse action, at scale and at speed, tends to strip this out entirely. The decision is made, the action is taken, and the person's role has been reduced from participant to recipient — from someone the process happens with to someone it happens to.

The efficiency argument for skipping the proceeding is seductive precisely because the proceeding looks like pure friction. Why convene anything, why wait, why solicit the person's side, when the system already has the inputs and can produce the correct output directly? But this reasoning mistakes the output for the whole of what a decision owes. It treats the proceeding as a slow and error-prone way of computing an answer the machine can compute faster — when the proceeding was never only about computing the answer. It was about who gets to participate in a judgment that will land on them, and that function does not become unnecessary just because the answer can now be reached without anyone's participation. Speed did not make the proceeding obsolete. It made it easy to skip and easy to forget it was ever owed.

A correct outcome imposed with no chance to be heard is not a small procedural shortcut. It is a verdict handed down by a proceeding that refused to convene.

Why the right answer is not enough

Suppose the automated decision is right — the person really does match the rule, the adverse action really is warranted. Even then, something essential is missing, and naming it precisely matters. What is missing is that the person had no opportunity to make it not be their case: to supply the fact the system lacked, to explain the circumstance that made the rule a poor fit, to contest the inputs that drove the result. The system decided on the evidence it happened to have, and the person never got to add to that evidence or challenge it. That the system's conclusion was correct on the inputs it had tells you nothing about whether it would have been correct on the fuller record a proceeding would have produced. Skipping the proceeding does not merely skip a formality. It forecloses the person's only route to changing the inputs, and then declares victory on inputs they were never allowed to touch.

And this is the case where the automated decision was accurate. The far larger danger is that it was not — that the input was wrong, the identity confused, the rule misapplied, the proxy misfired — and that the person, subjected to the consequence, has no proceeding in which to say so. Without a proceeding, an error is not something the affected person can surface; it is something they simply endure, indistinguishable from a correct decision because both arrive the same way: silently, finished, unappealable. A system that offers no way to contest cannot tell its own right answers from its wrong ones, because the mechanism by which wrongness would surface — the objection of the person harmed — has been designed out. The absence of a trial does not just deny justice in the hard cases. It removes the institution's own ability to know which cases were hard.

So the injustice of a verdict without a trial is not located in the correctness of the verdict. It is located in the missing trial — in the fact that a consequential judgment fell on a person who was never a party to it, could not be heard before it, and in many arrangements cannot meaningfully contest it after. You can get the answer right and still owe the person the proceeding, because the proceeding was never a means to the answer. It was a good the person was owed in its own right, and no amount of accuracy discharges a debt that was never about accuracy.

What an adverse decision owes

The remedy is not to slow every automated decision to the pace of a hearing, which would collapse under its own weight and help no one. The remedy is to build contestability into the adverse action itself — to treat the ability to challenge the decision as a required feature of imposing it, not an optional courtesy extended afterward. That begins with notice: the person must actually learn what was done and on what basis, because you cannot contest what you were never told. It requires disclosure of the grounds: a Decision Receipt that hands the affected person the evidence and reasoning behind the action, so their challenge can be aimed at something specific rather than shouted into a void. And it requires a real forum — a path by which the person's objection is not merely received but can actually change the outcome, before someone with the authority to reverse it.

The measure of such a system is not how rarely it is challenged but how well it survives challenge — whether, when a person does contest, the record is there to be examined, the reasoning is there to be tested, and a wrong decision can actually be undone. This is what turns an adverse action from a verdict handed down into a decision a person can answer. The point is not to make automated systems decide less. It is to make sure that when they decide against someone, the someone retains the standing to say no, here is why you are wrong — and to be heard by something that can act on the answer.

So the question to ask of any automated adverse action is not is it accurate but if it fell on me wrongly, what could I do about it — would I be told, would I know the grounds, could I contest it before something that could reverse it. If the honest answer is nothing, then what you are looking at is not a fast decision. It is a verdict without a trial — and a verdict without a trial does not become justice by being right. Justice was never only the outcome. It was the outcome plus the proceeding, and a system that keeps the first while quietly discarding the second has kept the easy half and thrown away the half that was the point.

— Dispatches · Summit Cognitive