Faster than the law

Technology has always run ahead of the law, and the law has always known it. The printing press, the railroad, the telephone, the spreadsheet — each arrived before the rules that would govern it, and in each case the law did what the law does, which is to arrive late, generalize from the harms that occurred, and write a settlement that the next generation inherits as ordinary. The lag was real, and it cost people, but it was survivable, because the technology and the law moved at speeds that were different but not incommensurable. The law could see what it was chasing.

Automated decision-making breaks that arrangement, not by changing its direction but by changing its scale. A legislature, a court, a regulator works at the pace of deliberation — the pace at which human beings can read, argue, and agree on the text of a rule. That pace has a floor, and the floor is measured in months and years. Meanwhile a single automated system makes consequential decisions at a pace measured in milliseconds, millions of them, each one affecting a real person, all of them happening in the interval before a single sentence of governing rule can be drafted, debated, and passed. The gap was always a lag. Now it is a chasm, and you cannot legislate across a chasm by writing the same way, only faster.

The race that cannot be won on speed

The instinctive response is to demand that the law accelerate — to write rules faster, to stand up agile regulators, to legislate at something closer to the speed of the systems being legislated. This is well-intentioned and it will not work, because the constraint on the law's speed is not a defect to be engineered away. It is the deliberation itself. A rule that is written as fast as a machine decides is not a faster rule; it is an undeliberated one, which is to say it is not really law at all, only a reflex with a statute number. The very thing that makes the law worth obeying — that it was argued over, that it represents a settlement rather than an impulse — is the thing that makes it slow. You cannot keep the legitimacy and discard the pace. They are the same property.

So the law cannot win the race on the terms the race is offered. It will always be too slow to govern automated decisions as they happen, and any attempt to make it fast enough will hollow out the thing that made it law. This sounds like a counsel of despair, and it would be, if governing decisions as they happen were the only form accountability could take. It is not. There is another form, and it is the law's oldest move: to govern not at the moment of the act but afterward, when the dispute arises and the case is brought.

The law has never needed to be present at the decision. It has only ever needed the decision to still be there when the law arrives.

The record buys the time

Almost all law works retrospectively. The rule may exist in advance, but its force is felt later — when a harm is alleged, a case is filed, and a tribunal reaches back to a past act and judges it. This retrospection is not a weakness of law; it is how law accommodates the fact that it cannot anticipate everything. It does not need to keep pace with events. It needs events to remain examinable until it gets to them. And that is precisely the capacity that automated systems, left to their own instincts, destroy. They decide and move on. They overwrite. By the time the law arrives — and it will arrive late, because that is the only speed it has — the decision it wants to judge is gone, leaving nothing to reach back to.

This locates the one thing the law can usefully require now, before it knows what the substantive rules should be, before the chasm has been measured, before the deliberation is anywhere near done. It can require that the decisions be preserved. Not governed, not yet pre-approved, not slowed to the law's own pace — merely kept, in a form that a future tribunal will be able to examine. A preserved decision record is the bridge across the chasm. It does not let the law move faster. It lets the decision wait, holding its evidence and its rule and its replayable state intact, so that when the law finally arrives the act is still there to be judged.

This is why the record is the right first demand, ahead of the substantive rules everyone is still arguing about. The substantive rules will take years, and they should; that is deliberation working as designed. But the requirement to preserve does not depend on knowing what the rules will be. It only requires the recognition that some rule, someday, will want to reach back — and that it will find nothing to hold unless we keep the record now. A Decision Receipt is, in this light, an instrument of legal time. It lets accountability arrive late, as it always has, and still find something to act on.

The chasm between the law and the machine cannot be closed by making the law quick. It can only be bridged by making the decision durable. The law buys back the time it has lost not by running faster but by insisting that the record outlast the decision — so that when judgment comes, however late, the act it judges has not already vanished into the speed that made it.

— Dispatches · Summit Cognitive