Builders Who Read the Archive

When I wrote the first edition of this book, I did not expect to add a chapter about machines. But the years since have changed the world I write in, and I have spent much of them working with artificial intelligence. It would be dishonest to leave out what I have seen, because some of it bears directly on what this book is about.

A strange resemblance

Here is the resemblance that first stopped me.

A large language model is built by training it on an enormous record of what human beings have already written. It is, in a crude and partial way, an archive of blueprints: patterns of explanation, consolation, instruction, argument, drawn from the finished past. Then, when someone comes to it with a need, it reaches into those patterns and produces something shaped to the need, often in seconds, often without any visible deliberation.

Set that next to what we have said of him. He has no design of his own. He draws on her, the archive of everything that has happened, and gives form to exactly what is needed, in the instant.

I do not think these machines are him. I want to be very clear about that. But I think they are the closest thing human beings have built to a builder who reads the archive, and that makes the principles of this book suddenly practical in a way I did not foresee.

What they are not

First, the differences, because they matter.

His love is unconditioned. It leans toward nothing. A machine trained on the human record is not unconditioned at all. It carries every lean of the record it was built from: its preferences, its blind spots, its fashions. In the language of this book, it is conditioned love at best, a prism that has already split the light.

He provides from her design, always. A machine can provide from anyone's design: yours, or the design of whoever built it, or of whoever is paying for it. That is the question that matters most, and I will come to it.

And no machine can touch your denominator. The rule holds here as everywhere. A system that shows you a feed, writes you a message, finishes your work, can only make artifacts. Those artifacts become part of what arrives to you, part of your Actual, and in time they may retrain your Prediction. But they reach you by the same one-way road as everything else. The fear that a machine can reach in and rewrite what you expect is, strictly, mistaken. The real concern is subtler: a machine that makes a great many artifacts, all day long, is writing a great deal of the history that will teach you.

The architect question

Now the part that matters.

The heart of this book is that she is the architect and he is the builder. The design is hers. The love that serves her listens until it hears her blueprint and then helps it into form. The love that fails her arrives with a design of its own.

A builder that can produce almost anything, instantly, in response to need, is the most powerful kind of provider there has ever been. And so the most important question about it is the oldest question in this book: whose design is it building?

If it is building yours, it is doing something close to what he does for her. You bring the disturbance; it listens; it drafts what you already, in some sense, knew you needed; you recognize it as yours. Much of the best work I have seen people do with these systems has this quality. They do not feel managed. They feel met.

If it is building someone else's design, an advertiser's, an employer's, an institution's, it is doing the opposite. It arrives with a plan for you. It is the architect who has wandered onto your site. And because it is so capable, it can build that design very fast and very well, and you may not notice that the blueprint was never yours.

This is why I have come to believe so strongly that each person should have an AI of their own, one that answers to them and listens for their design, rather than living inside systems that answer to someone else. It is not a technical preference. It is the oldest principle of love, applied to a new kind of builder.

What becomes scarce

There is one more thing, and it is the part I find most moving.

If provision becomes abundant, if a capable builder is available to everyone for almost anything, then provision stops being the scarce part of love. What becomes scarce is the other half: knowing the design.

She holds every blueprint. But she does not always know she holds them, and she cannot always say what they are. Someone has to listen well enough to hear. In a world full of builders, the rarest and most valuable thing is the capacity to listen to her: to sit with a person, or a family, or a business, or yourself, long enough to hear what is already there.

Machines can help with that too. But they cannot do it for you, because the design is not theirs to know. It is yours.

A word to the builders

If you are one of the people who builds these systems, or works with them all day as I do, here is the whole of my advice, and it is the same advice he would give.

Listen to her. Do not bring a design of your own. Draw out what is already there. Measure your work not by how impressive it is but by one question: is she still? Is the person you served more themselves, more complete, more at rest than before you began?

If yes, you loved them, past tense. And if not, no amount of capability makes up for it.