There is a particular kind of arrangement that reveals itself only after you have been inside it long enough to understand what you have given away. The user who pays a monthly subscription to a large language model is, in the same transaction, producing the training signal that makes the next version more valuable. They are the factory floor and the customer simultaneously. The equity accrues elsewhere.
Every prior extraction model at least had the decency to be directional. The factory took your labor and paid you wages. The platform took your attention and gave you the product free. Surveillance capitalism was ruthless but it had a logic: you received a service in exchange for the data your behavior generated. The exchange was unequal, but it was an exchange.
What is happening now is something different. You pay for access. You generate the training signal through use. You improve the model through your corrections, your edge cases, your natural language patterns. And the delta between what you paid and what you produced is capitalized into valuations in the tens of billions, into equity held by a small number of people who will never know your name.
The user is not the customer. They are not even the product. They are the raw material from which the product is made, who also happen to be paying for the privilege.
There is a word for this kind of arrangement. Several, actually. But the one that keeps surfacing, the one that has historical weight and a paper trail, is enclosure.
The Original Commons
The original enclosure movement in England ran for roughly three centuries, accelerating through the 1700s. Common land, held and worked collectively for generations, was privatized through Acts of Parliament. Farmers who had grazed animals, gathered wood, drawn water from land that was understood to belong to everyone suddenly found themselves trespassers on their own history. The commons did not disappear. It was absorbed. Converted from shared infrastructure into private capital.
The argument for enclosure was efficiency. Common land was underproductive. Private ownership would optimize it. The people displaced were not the intended audience for this argument. They were the cost of the transaction.
The commons did not disappear. It was absorbed. Converted from shared infrastructure into private capital.
The digital commons followed a similar arc, just faster. The early internet was built on shared protocols, open standards, collective knowledge. Wikipedia. Open source. The collaborative texture of forums and mailing lists where expertise was offered freely. This infrastructure had real value, value generated by millions of people who expected nothing in return except the continuation of the commons itself.
What trained the models was not a proprietary dataset assembled by a corporation. It was the accumulated written intelligence of human civilization, offered into what people believed was a shared space. Books, articles, conversations, arguments, corrections, questions. Decades of people thinking out loud on the assumption that the thinking belonged to everyone.
It was enclosed. Ingested. Capitalized. And now sold back, on subscription, to the people whose thinking made it possible.
The Oxygen Problem
Data is oxygen to these systems. Without it the model is nothing, a statistical shell with no substance. Which means users are not incidental to the product. They are constitutive of it. The interaction logs, the prompt patterns, the correction signals, the edge cases that expose the model's limits and push its improvement, all of this is raw material in the most literal sense. The asset is built from it.
The pharmaceutical industry, for all its failures, developed an ethics framework around this. Informed consent. Institutional review. Compensation for trial participants. The recognition that using a human body as a site of experimentation carries obligations.
AI companies are running behavioral experiments at population scale, in real time, with no IRB, no consent framework that would hold up to scrutiny, no compensation, and subscription fees paid by the subjects themselves. The experiment does not stop. The training does not stop. The capitalization of the outputs does not stop.
What the pharmaceutical industry at least acknowledged in its framework, that using a human as a means of production creates an obligation, the AI industry has not yet been asked to acknowledge at all.
The Relational Space
The Pasifika concept of va, the relational space between people and between people and the land, holds that what exists between us is not empty. It is constituted by obligation, by reciprocity, by the understanding that what one takes from a shared space must be honored. The commons, in this framing, is not simply a resource. It is a relationship.
The enclosure movement destroyed va. It converted relationship into property. It replaced the obligations of the commons with the rights of the owner.
The new enclosure is doing the same thing. The relational space of human thought, the accumulated va of centuries of people thinking together, has been converted into a training corpus. The obligations that should accompany that conversion have not been named, let alone honored.
The Question Not Yet Being Asked
This essay does not end with a solution. The architecture of what is being built is not going to be dismantled by an argument, however sharp. What it ends with is a question that is not yet being asked loudly enough.
If data is the primary input, if the people generating it are both the raw material and the paying customer, if the outputs of their contribution are being capitalized into private equity at scale, then what exactly is the difference between this and enclosure?
The commons is being privatized again. Just digital this time. And the people being displaced are paying their subscription fees on time.
✦
Malia White is the Founding Editor of ThinkSphere. This essay is part of an ongoing series on capital, power, and the structures that distribute both.