The First Work Was Naming

The first work ever given to a human being, in the oldest story we have, was not building. It was not ruling, or organizing, or defending anything. It was naming.

The scene is strange enough that it is easy to read past. Creatures are formed and brought to the man — as the World English Bible renders it, "Out of the ground Yahweh God formed every animal of the field, and every bird of the sky, and brought them to the man to see what he would call them. Whatever the man called each living creature became its name."

To see what he would call them. I have read that line most of my life and only recently noticed what is being described. Not an inventory. Not a taxonomy handed down for memorization. An open question, with a real answer that was not predetermined, given to someone who would then have to live in a world shaped by his own answer.

Naming and labeling are not the same act

I have spent forty years building systems that classify. In payments, that is most of the job — this transaction is fraudulent, this merchant is high-risk, this borrower is creditworthy, this account should be frozen pending review. I have written the rules and, later, trained the models. I am not standing outside this problem pointing at it. I am inside it.

And what I have come to believe is that classification, which I do professionally, and naming, which that old story describes, are different acts that we have started to confuse. Three differences seem to matter most.

A name is given in the presence of the named. The animals were brought to him. He had to look. Labeling is done at a distance, from a record — and the whole point of the record is that it lets you decide without being present. That is not incidental to how classification scales. It is the mechanism by which it scales. You cannot process ten million cases if you have to look at ten million faces.

A name can be contested by the one who carries it. If someone calls you something wrong to your face, you can say so, and the exchange happens between two parties who both have standing. A label assigned by a system is difficult to appeal precisely because it was never addressed to you. It was addressed to another part of the machine. You are not in the conversation; you are the subject of it. Anyone who has ever tried to argue with a credit decision has felt the peculiar vertigo of trying to talk back to a sentence that was not spoken to them.

A name binds the namer. This is the one I keep returning to. The man who names the animals still has to live among them. He is not evaluated on naming accuracy and then reassigned. He stays. A classifier is accountable to a metric, and a metric has no memory of any individual it was wrong about. It reports ninety-four percent and moves on, and the six percent are not an error the system experiences — they are a rounding cost the system was designed to accept.

The scale is new; the temptation is not

I want to be careful not to pretend this is a problem that machines invented. Human beings have always labeled each other carelessly and at a distance, and we have used those labels to justify appalling things for as long as there have been records to keep them in. A ledger is a labeling machine. So is a census, a caste, a rumor.

What is new is volume, speed, and the disappearance of the person who could be asked to answer. When a human being labels you unfairly, there is at least someone to confront. When a system does it, the honest answer to "who decided this about me?" is often that nobody did — the decision emerged from a model trained on records of decisions made by people who are no longer at the company, encoding an environment that no longer exists.

That is the part I find genuinely troubling, and I do not think it is solved by better models. A more accurate classifier is still a classifier. Accuracy is a good property and it is not the property in question here. The question in question is whether anyone stayed.

What I think this asks of the people who build

I do not read the naming scene as an argument against classification. Modern life is not available without it, and I intend to keep building systems that do it. Refusing to classify is not humility; it is usually just handing the job to someone with less care.

What I take from it is narrower and harder. If naming was the first work given to us, then the people who build classifying systems have inherited an old responsibility that is not written anywhere in the job description.

Practically, for me, that has come to mean a few things. That every consequential label a system applies should have a human being who can be asked about it and who has the authority to change it — not a queue, a person. That the appeal path deserves as much design attention as the decision path, which in most systems I have seen it emphatically does not get. That when we choose a threshold, we are choosing which people fall on the wrong side of it, and we should say that out loud in the meeting rather than let the number do the saying for us.

And underneath all of it, a conviction I hold on faith rather than from engineering: that a person is not the sum of what a system has recorded about them. The record is real. It is often useful. It is never the whole creature. Whatever a model has concluded from a person's history, that conclusion is a description of the data, and the data was never the point.

To see what he would call them

What stays with me most is the openness of that phrase. God does not dictate the names. He brings the creatures and waits to see. There is something almost vulnerable in it — a genuine handing-over, with the outcome not fixed in advance.

That is a startling amount of trust to place in a creature who, a chapter later, will get something enormously more important badly wrong. And yet the naming stands. The names were his.

We are doing the same work now, at a scale that would have been unimaginable, with tools of extraordinary power, on subjects who are not animals but each other. The work is the same work. What we should not lose is the posture the story assumes: that the one who names is present, is answerable, and has to stay.