An Affirmative Vision

In which complaining takes a holiday

I should admit something before I go any further, because it bears on whether you should believe a word of what follows.

I am, by disposition, a complainer. A Complainasaurus Rex, if we’re being precise about the species. I have spent a fair amount of writing naming what is going wrong, and naming it, I hope, with some precision: the institutions that mistake caution for character, the training corpus that turns out to be a national portrait no one sat for, the dream of a solved world that would optimize away the very friction that makes us agents. I am good at the diagnosis. I am drawn to it. There is a version of me that could do this indefinitely, the way some people are content to be right about the weather.

And I am aware that a great deal of what I call a problem is, to a reasonable person, simply the state of the world. Institutions are slow. People offload responsibility. Tools get used to avoid the discomfort the tools were built to avoid. None of this is new, and a certain kind of maturity consists in noticing it once and then getting on with your life. The complaint, repeated, starts to sound less like insight and more like dim-bulb temperament.

So, I have been trying to ask myself what is underneath it. Not the complaint, but the thing the complaint keeps circling. If I object this consistently to the same disappearances, the same evasions, the same quiet substitutions of process for judgment, then I must be standing on something. Some affirmative ethic, slow and plodding as it may be. This essay is my attempt to turn the picture around. To stop describing the shape of the hole and describe the shape of the thing that should be there.

The danger of calling any of this a vision is that the word arrives wearing a lanyard and looking for a high-five. It sounds like something laminated, workshopped, affixed to a wall near the elevator. That is not what I mean. I do not have a program in the heroic sense. I have a set of refusals that, if followed far enough, begin to imply obligations.

If I keep objecting to the same disappearances, then the objection is not the point. The point is whatever I think should remain visible. If I keep noticing the same evasions, then the work is not merely to expose evasion. It is to build places where responsibility has fewer places to hide. The complaint, at its very best, is a negative image of care.

So this is where I start: our technologies do not invent our problems. They expose them. The systems we are building reveal the systems we already had: the gaps in our oversight, the limits of our trust, the unfinished business of who decides what about whom. This is, oddly, good news. It means the work in front of us is not foreign. It is continuous with every previous generation’s work of building institutions worthy of the people inside them.

What I find underneath the complaint is not a solution. It is a pair of obligations that have to be carried together.

The first is that moral agency cannot be delegated. Not to algorithms, not to organizations, not to history, not to the comforting abstraction of “the system.” Wherever a decision matters, some human has to be answerable for it. Not merely present near it. Not merely informed after the fact. Answerable. The infrastructure of accountability has to be designed deliberately rather than assumed.

That means building mechanisms and mechanisms are duller than the principles they serve. Picture the moment a system produces an answer and the room subtly rearranges itself around the answer, as if the human decision has already happened somewhere else. Most of what I care about is decided at that moment and decided badly when no one has built for it in advance. So: a name on the decision memo. A review point before the system goes live, not after it becomes a headline. A place in the record where someone has to say why the recommendation was accepted, modified, or refused. An audit trail that survives turnover, embarrassment, and the natural institutional desire to forget. A dissent channel that does not require a person to become a martyr before anyone listens. A procurement question that asks who becomes responsible if the tool is wrong. An after-action review that treats failure as information rather than contamination. Accountability has to live somewhere, in a person, a form, a threshold, a meeting. Otherwise it becomes a word people use after the damage is done.

I think this is one of the great unfinished projects of public life, and I want to spend my career on it.

The second obligation is that the deepest crisis we face is not technical. It is a crisis of seeing. We are awash in information and starved for sense. Patterns multiply faster than meaning. The temptation in such moments is to retreat into techno-optimism or reflexive doom, both of which save us from the harder task of looking at what is actually in front of us and saying it plainly.

So I write and I work. I work and I write. Not to predict, not to warn, but to describe and to build. To name what is happening while it is still happening. To say when the vocabulary of accountability is being used to conceal the disappearance of responsibility. To notice when a human decision has been quietly transformed into a system output, and when a system output has been quietly transformed into an institutional alibi. To leave a record that someone, later, can use.

These two practices need each other. Building without seeing produces brittle institutions. Seeing without building produces sophisticated despair. I do not think either half of the work is sufficient, and I do not think the tension between them resolves. It is meant to be lived in.

The central danger I worry about is the quiet displacement of human judgment. Not the dramatic version. Not robots making decisions while we sleep. Something subtler and more corrosive. The slow habituation to letting systems decide what we used to decide ourselves, and the gradual atrophy of the practices that made us capable of deciding in the first place.

A hard decision has a texture to it. The pause before the call. The discomfort of knowing reasonable people may disagree. The conversation with a colleague who sees what you missed. The attempt to separate what the rule permits from what the situation requires. The burden of being able to explain yourself afterward. These are not inefficiencies to be optimized away. They are how judgment lives in the world.

What concerns me is the steady, almost unnoticed migration of the question itself. We stop asking whether a decision was right and start asking whether the system was followed. We stop asking what we think and start asking what it recommends. The vocabulary of accountability remains, but the substance wanes.

The displacement is not loud. It does not have an air horn. It happens in the small concessions, the defaulted-to outputs, the meetings where no one quite remembers who decided. The institutions we are building around these systems will either preserve the human practice of judgment or quietly retire it, and most of the time we will not know which until later than we would have liked.

The affirmative version of this work, then, is to keep judgment load-bearing. To design systems that hold humans answerable rather than relieve them. To preserve room for disagreement, escalation, exception, refusal, and revision. To make sure there is still a person who can say no, a record that can explain why, a meeting where uncertainty can be spoken aloud, a process that can be interrupted before the damage is laundered into inevitability. To write in a way that reminds people they are still the ones doing the seeing.

What I want is modest and large at once. I want institutions that can use powerful tools without hiding inside them. I want accountability to be a habit rather than a slogan. I want oversight that notices early enough to matter. I want the writers and the builders to recognize each other as colleagues.

Most of all, I want to help build institutions that can still make judgments in the presence of machines that are very good at producing answers.

That is the work in front of me.

It is workable. Not solved. Not safe. Not finished. Workable. That seems to me the only useful place to stand.