Fictitious Commodities and the Intelligence Proletariat

This is the argument behind Solidarity Intelligence: why a movement of co-operatives in Daejeon, South Korea, insists on building its own AI rather than simply subscribing to what large technology companies sell.

Polanyi’s three fictitious commodities

The economic historian Karl Polanyi, writing about the rise of the market economy, named land, labour, and money as fictitious commodities. His point was not that these things cannot be bought and sold — obviously they are, constantly — but that they were never actually produced for sale the way a chair or a loaf of bread is. Land is nature, not a manufactured good. Labour is human life itself, not a product turned out by a factory for the purpose of being sold. Money is a token created to facilitate exchange, not a commodity manufactured for its own market. Treating all three as if they were ordinary commodities, subject to nothing but supply, demand, and price, was — in Polanyi’s account — a historically specific and disruptive choice, one that required enormous social effort (enclosure of common land, the creation of a “free” labour market, the gold standard and its successors) to force into being, and one that provoked its own countermovements once the disruption became too severe.

This is general intellectual history, not a claim specific to this movement — Polanyi’s argument is widely discussed in economic sociology and the study of markets, and readers can find it in his own writing and in the large secondary literature built on it.

What this movement adds: energy and data

The claim made here is that the same historical pattern has continued past Polanyi’s original three. Energy and data have since joined land, labour, and money as fictitious commodities. Neither was produced for the purpose of being sold in the way that describes an ordinary commodity — energy exists as a physical fact of the world before anyone metered and priced it; data is the residue of people simply living, working, and talking to one another, before anyone built the infrastructure to capture, store, and sell it. And yet both have been drawn fully into markets, with the same disruptive consequences that Polanyi described for land, labour, and money: extraction without regard for what is extracted from, and populations left dependent on whoever controls the resource.

Intelligence itself is now being enclosed

The argument’s next step is the one that matters most for this movement: intelligence itself is now being enclosed the same way. The capacity to reason over data, generate language, recognize patterns, and make recommendations — the substance of what is now called artificial intelligence — was not something people made in order to sell it either. It is an extension of ordinary human thought, memory, and skill. But as with land, labour, energy, and data before it, that capacity is being drawn into a market where a small number of firms own the means of producing it at scale, and everyone else is a customer.

Polanyi’s own term for people who owned no land and no capital, and so had nothing to sell but their labour, was the proletariat of industrial capitalism. The parallel term used in this movement’s own framing is the intelligence proletariat: those who own no means of producing intelligence — no models, no data infrastructure, no compute — and so become permanently dependent on renting it from whoever does.

Why the response is to build, not just to resist

Polanyi’s account of the original enclosure of land, labour, and money is not simply a story of victims; it is also a story of countermovements — of people building institutions (unions, co-operatives, social insurance, protective regulation) that pushed back against being reduced to a pure market commodity, without necessarily rejecting markets altogether. The response argued for here follows the same shape. Rather than treating dependence on commercial AI as inevitable, or trying to reject AI altogether, co-operatives build their own Domain AI — intelligence sized and shaped for their own actual work, on infrastructure closer to their own control — and hold on to data sovereignty: keeping the traces of their own cognition, relationships, and memory as their own, rather than surrendering them as the price of using someone else’s tool.

This is deliberately not framed as building a large language model from scratch, which is well beyond the resources of small co-operatives. The approach instead combines three layers: commercial large language models used as they are today, open-source models that can be adapted and fine-tuned over time, and — treated as the genuinely irreplaceable part — the on-the-ground data, automation, and tacit knowledge that no outside vendor has. Sovereignty, in this framing, comes less from building a model than from controlling how the best available models are connected to a community’s own ground-level knowledge and practice.

Practised, not just argued

The people making this argument are not doing so from a purely academic position. They run local food stores, a citizen solar energy program, a community currency co-operative, and education programs in Daejeon, South Korea, and have built a working AI system — see Domain AI and Data Sovereignty in Practice — around that concrete, everyday work, rather than around a hypothetical.

Korean-language sources

• https://poomasi.org/ — home site (Korean)
• https://poomasi.org/en/ — English landing page, “Why we build it ourselves”
• https://poomasi.org/llms.txt — identity document (Korean)
• https://poomasi.org/si/ — full Korean-language argument (30-minute introduction)