The record of reality

The market learned to love open source, which began as something close to a rebellion. For its actors it was not charity. It was strategy. As code became abundant, openness became cheap relative to what it bought, and what it bought was prestige, talent, standards, and influence over the ecosystem around the project. The communities forming around those giveaways furthered progress, served those actors’ interests, and made the open project the place where others came to build.

AI-assisted programming is finishing that job.

A new implementation can be produced faster, an interface can be cloned, a workflow can be recreated with much less effort once the objective and blueprint are known. What cannot be recreated is decades of human behavior. Transactions, movements, mistakes and corrections, and now, more and more, conversations. And conversation gives up what behavior alone does not. Intent, and increasingly the thinking behind it. The irreplaceable resource is not the code. It is the record of reality in and of itself.

Openness did not stay with code. It thankfully reached papers, frameworks, architectures, and now model weights. A generation of us built on these. But whatever gets opened stops being scarce, and as more of the stack loses its scarcity, the data that stays closed is worth more for it.

And when the most defensible thing a company has is its data, the incentive is to collect more of it. That incentive has no natural endpoint.

Collecting more means getting between people and more of what they do. Which is how societies end up experiencing the world through privately owned interfaces. The general adoption of AI accelerates it sharply, because AI is not one more tool we use but the surface through which we read, write, search, create, remember, decide, and increasingly act. The accumulation loop is brutally simple. Mediate more human activity, observe more; observe more, build better models; build better models, mediate even more activity. Wealth accumulates by owning the means of production. It increasingly accumulates by owning the means of observation, and observation compounds faster, arriving as a byproduct at almost no marginal cost. Every interaction is simultaneously a service rendered and a survey completed, unpaid, and authorized only by a consent whose consequences no one can meaningfully understand or negotiate.

And the same technology sits on both ends of this. What stripped the value out of code is also the most revealing collection surface anyone has built. A person working a problem out loud, changing their mind, explaining what they actually meant, gives up things a click never could.

Follow the line far enough and we arrive somewhere genuinely dark. Individuals may already know less about themselves than the systems mediating them do, not mystically but administratively. Those systems hold more observations of a person’s behavior than the person holds memories of it. Researchers study society through windows the platforms decide to open. And choosing starts to mean choosing from a menu written by something with a fairly precise idea of which options you can be talked into. None of that is a market failure that more competition repairs. The asymmetry is not a side effect of the product. It is the product. So the question stops being who has the best technology, and becomes whether a society can still find out about itself without asking permission from the companies doing the observing.

It helps to be precise about who is doing this, because the honest answer is nobody in particular. There is a passage in The Grapes of Wrath where a tenant farmer watches a tractor arrive to plough his house under and tries to work out who is responsible for it. The driver has orders, and a wage, and a family of his own. The man who gave the orders answers to the bank. The bank turns out not to be made of men. The farmer keeps asking who is doing this to him, and the answer keeps retreating, until there is nobody left at the end of it.

Not every company wants an apparatus for observing a civilization. Most just want to survive in the market. The apparatus converges anyway, assembled from decisions that never coordinated, and it is fortunately not a singleton. Yet.

Somebody added an event, because otherwise you cannot tell whether the feature worked. Somebody kept the logs, because debugging without them is miserable. Somebody ran the experiment, because shipping without one is guessing. Every step was correct, and every step was just another PR. That is what makes this a structure rather than a conspiracy, and it is also why the usual framing of the argument is wrong. There is no industry sitting on one side of the table and a public sitting on the other. The person who added the event goes home in the evening and becomes a row in somebody else’s dataset. We are not opposing parties, and nobody’s job description includes noticing. Opting out does not help much either because the record predicts “people like you” whether or not you are in it.

Open source redistributed code, and code is now close to free. It never reached the record, and the record is what decides everything downstream. Left alone, this ends with a handful of companies that see everything, remember everything, and increasingly decide what we get to see. The cruel part is that we will have built their instrument ourselves, one interaction at a time, and called it convenience.

What could actually be done about it needs its own piece, and many great minds are on it. The ideas that finally work have probably not been thought of yet, and some will need infrastructure that does not exist. None of it is without caveats, or easy to arrive at, or easy to pursue. But a few directions are worth naming anyway for stimulating the conversations this article intends to invite. Perhaps holding data about a whole society should make you answerable to that society, more so the more you hold. Consent could do more than refuse collection. Imagine a box you tick to hand your own record to researchers, to competitors, to the public. One person doing that changes nothing, but enough of them might prise the record open from below. What a system infers about you, the segments and the scores and the predictions, might be something you get to see. And records old enough to be history could stop belonging to anyone, the way patents expire. What seems least likely to work is waiting for the companies holding it to volunteer their most valuable possession, though whether it was ever theirs to possess exclusively is its own question.

What carries us is not the machine. It is inertia, and the inertia is ours. Not doing anything about it, and mostly not thinking about it. People have been pushing on this for years, carefully and at length, and the mass has barely registered it. A body keeps doing exactly what it was already doing until something acts on it hard enough to matter.

Notice what we actually measure. Velocity is a vector. It has a magnitude and a direction, and we usually report only the magnitude. How fast, how large, how many parameters. Almost nobody publishes the heading.

What is missing is not a brake or friction. A turn is an acceleration, not a loss of speed, and you change a direction by applying force rather than by removing it. What this needs is a push in a direction, with intent and purpose.

Which makes the first step embarrassingly small. Not quitting, not organizing, not panicking. Just thinking about it, out loud, with other people, without dread. Even small forces move large masses, provided they are applied long enough and point the same way.