Authority without a witness
For most of human history, the exercise of serious authority was witnessed. We are now the first generation to build authority that need not be.
Consider how rarely, until very recently, a consequential thing was done entirely alone. A loan was granted across a desk, with a clerk who would remember the conversation. A sentence was passed in a room with a stenographer and a gallery. A border was crossed under the eye of an officer who stamped a book. Even the most private exercise of power tended to leave a witness behind — not always a sympathetic one, not always an honest one, but a someone, a presence in the room who could be asked, later, what they saw. Authority was attended. It was watched in the act of being exercised, and that watching was not incidental to its legitimacy. It was part of what made the authority answerable, and therefore part of what made it tolerable.
We have, almost without remarking on it, built a different kind of authority. An automated system can deny a claim, freeze an account, flag a person, or refuse an application without a single human being present at the moment it acts. There is no clerk who will remember. There is no gallery. There is no officer with a stamp. The decision happens in a process that has no occupant — not because anyone was excluded, but because the room was never built. The authority is real; its consequences are real; and there was no one there.
This is not the same problem as automation being fast, or opaque, or biased, though it travels with all three. It is a more basic discontinuity. The thing we have removed is not a person's labor. It is the witness — the structural fact that, for nearly the whole history of governance and commerce, somebody could in principle be summoned and asked, and that the possibility of being asked shaped how power behaved while it was being used.
What the witness was for
The witness was never primarily a memory aid. We did not keep clerks in rooms because we feared we would forget. We kept them because their presence created an account that did not belong to the person exercising the authority. A witness is, above all, an independent custodian of what happened — someone whose recollection is not the decision-maker's to edit. That independence is the whole point. A power that supplies its own and only record of its own conduct is, in a precise sense, unwitnessed, no matter how much paper it generates.
This is why the loss is easy to miss. Automated systems are not silent. They produce logs, timestamps, status codes, confirmation emails. It can look as though the witness has been replaced by something better — a tireless, exact, machine-grade memory. But a log written by the system, retained at the system's discretion, in a form the system chose, is not a witness. It is the defendant's own diary. It may be entirely truthful. The problem is that its truthfulness is unverifiable from the outside, and a witness whose honesty you simply have to take on faith is not performing the function a witness performs.
A power that keeps its own and only account of its own conduct has not been witnessed. It has merely been recorded by the one party with the most reason to record selectively.
So the danger is not that automated authority lies. Most of it does not. The danger is that it acts in a place where, structurally, no independent party could ever attest to what was done — and that this absence is invisible precisely because the surface looks so well-documented. Lawfulness does not close the gap. A decision can be entirely within an institution's rights, correctly executed under its own rules, and still be the first such decision in history to have been exercised with no witness at all. Being authorized to act is not the same as having been seen to act. Authorization is not legitimacy, and one of the things legitimacy historically required was a witness.
Rebuilding the room
The answer is not nostalgia. We are not going to staff every automated decision with a clerk, and we should not pretend that the slow, partial, fallible human witness of the past was a golden standard worth restoring on its own terms. The point of recovering the witness is not to slow the machine down. It is to insist that authority, however fast and however automated, must still produce an account that does not belong to it alone.
That means a record built to a different specification than the system's own convenience. A record whose provenance can be checked independently — so that one can ask not only what the system says it did, but whether that account is intact and untampered. A record that carries the evidence actually consulted and the rules actually in force, so that the decision can be replayed by a party who does not trust the decider. A record, in short, that can stand in for the witness who is no longer in the room: an independent custodian of what happened, against which the power's own story can be tested. The witness used to be a person. It can now be a structure. What it cannot be is absent.
We tend to assume that the legitimacy of an institution rests on its mandate — on whether it was entitled to act. But mandate was never the whole of it. For most of history, the entitlement to act came bundled with the condition that the acting be seen. We are the first generation with the technical power to unbundle them, to exercise authority in a sealed room with no occupant and no independent account. That power is genuinely new, and like most genuinely new powers, it is dangerous in exact proportion to how natural and convenient it feels. The question to ask of any automated authority is no longer only whether it was allowed to do what it did. It is whether anyone, or anything, could attest that it did.
— Dispatches · Summit Cognitive
Continue from here
Turn the argument into a practice.
Get new dispatches, assess how your organization handles consequential decisions, or explore Summit Cognitive.