Primitive Accumulation in the AI Age

In the AI Story

The enclosure operates at civilizational scale. The datasets training frontier AI models contain trillions of tokens drawn from millions of creators — novelists, poets, programmers, journalists, academics, artists, musicians, bloggers, forum participants. This material was produced over decades of labor. It was shared with the expectation that it would circulate within the commons of human knowledge. It was scraped without consent in a period measured in months. The speed and scope of the appropriation have no historical parallel.

The transformation is as significant as the seizure. The English enclosures did not merely transfer land ownership; they destroyed the communal governance structures, shared ecological knowledge, and mutual aid networks that the commons had sustained. The AI enclosure does not merely copy creative work; it transforms work that was collective, attributed, and embedded in communities of practice into statistical patterns in proprietary models. The knowledge that was common becomes the raw material for products that compete with the workers whose knowledge they embody. The dispossession is epistemic as well as economic.

David Harvey's 'accumulation by dispossession' describes how neoliberal capital expands through the ongoing appropriation of public goods — privatization of utilities, financialization of housing, commodification of knowledge. The AI enclosure extends this mechanism into the creative and intellectual commons. What makes the AI case distinctive is the technological mediation: the appropriation is performed not through legal privatization or military seizure but through automated scraping and computational transformation. The technology does not create the logic of enclosure. It enables enclosure at a scale and speed that previous technologies could not achieve.

The training data enclosure eliminates the reciprocity that had governed the creative commons. Writers shared their work expecting readers. Programmers shared code expecting users and collaborators. Artists shared images expecting viewers. The sharing sustained communities of practice, enabled learning, and built the collective knowledge on which all creative work depends. AI models trained on this commons do not participate in these reciprocal relationships. They extract knowledge without contributing to the community that produced it, creating a one-way transfer that degrades the commons while enriching the enclosers.

Origin

Marx introduced the concept of primitive accumulation in Volume 1 of Capital (1867) to describe the violent origins of the wage relation — the enclosures, the colonial conquests, the forced separations of producers from their means of subsistence that created the preconditions for capitalist production. Federici's Caliban and the Witch (2004) extended the concept by demonstrating that primitive accumulation included the gendered violence of witch hunts, the disciplining of women's bodies, and the destruction of women's economic autonomy. The application to AI training data draws on both Marx's original insight and Federici's extension, identifying the appropriation of collective knowledge as a contemporary instance of the same structural process.

Key Ideas

The creative commons was produced by labor. The training data is not a natural resource but the accumulated product of millions of creators' work — labor that was appropriated without consent or compensation.

Enclosure transforms common into private. The knowledge that was shared, attributed, and governed by communities of practice becomes statistical patterns in proprietary models owned by corporations.

Dispossession is epistemic and economic. Creators lose not only market position but sovereignty over their own creative knowledge, which is absorbed and recombined without their participation.

The enclosure is ongoing, not historical. Each new model trained, each new dataset scraped, each new expansion of proprietary AI capabilities continues the process of primitive accumulation at computational speed.