The Fragment on Machines

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI reads the present through the eyes of thinkers who saw it coming from unexpected angles. Marx saw it from 1857, and what he saw was the enclosure question: not “what can the machine do?” but “who owns it, and whose knowledge is inside it?” The Fragment on Machines names the precise condition of AI training data: Wikipedia articles, Stack Overflow threads, arXiv preprints, GitHub repositories, novels, poems, scientific papers—the collective cognitive product of millions of human beings who contributed their work to a commons and whose contributions have been converted, without consent or compensation, into the training corpus of a privately owned productive system.

The Fragment also contains Marx’s most explicit prediction about the structural consequences of this transition. When the general intellect becomes the primary productive force, the basis on which value is distributed—labor time—ceases to correspond to the basis on which it is produced. The worker who contributes one hour of labor to a process dominated by AI contributes a smaller and smaller fraction of the total productive capacity; yet the distribution of output—wages for labor, returns for capital—continues to be organized as if labor time were the primary productive factor. This contradiction, Marx argued, is not stable. The specific form it takes in the AI age—productivity gains captured by capital while creative and knowledge workers face structural displacement—is what the cycle asks its readers to see clearly.

The Fragment sits in the cycle alongside dead labor analysis and the concept of enclosure of the intelligence commons, as the three Marxian instruments the cycle deploys to make the political economy of AI legible. Of the three, the Fragment is the most forward-looking: it does not describe what capitalism does to knowledge but what knowledge does to capitalism when it becomes the dominant productive force.

Origin

The Grundrisse was written in 1857-8 as Marx’s preparatory work for Capital, which he was not to begin publishing until 1867. The manuscript was never finished and was not published until 1939-41, in German, in an edition by the Marx-Engels Institute in Moscow. The first English translation appeared in 1973. The “Fragment on Machines” is not a formal section heading but the name given by translators and commentators to a passage of roughly twenty pages in the notebook section of the Grundrisse.

The passage gained its current prominence through the Italian operaismo (workerism) movement of the 1960s and 1970s, which used it to argue that the primary site of productive activity was shifting from the factory floor to society as a whole—that “social knowledge” had become the primary productive force and that this required a new political analysis. The operaisti (including Romano Alquati, Mario Tronti, and Antonio Negri) developed the Fragment’s analysis of the general intellect into a framework for understanding post-Fordist capitalism that anticipated many features of the knowledge economy before the knowledge economy had a name.

In the AI context, the Fragment has been taken up by a range of thinkers including Nick Dyer-Witheford, Matteo Pasquinelli, and Aaron Bastani, each of whom draws on it to analyze the enclosure of the cognitive commons represented by AI training data and the specific political economy of platforms that return collective knowledge as a proprietary service.

Key Ideas

The general intellect as productive force. Marx’s most original claim in the Fragment is that collective human knowledge—science, technology, social intelligence, accumulated craft—can become the primary productive force of a society, displacing individual labor as the source of productive capacity. When this happens, the standard measure of value (labor time) loses its purchase on the reality of production. The Fragment is Marx’s acknowledgment that his own labor theory of value has a limit condition.

The contradiction of collective production and private ownership. The general intellect is produced collectively, across generations, through processes that include education, scientific publication, open-source development, and cultural transmission. Its appropriation as private capital—the training corpus converted into proprietary weights—is a structural contradiction: value is socially produced but privately captured. This is not new to capitalism, but the scale of the appropriation in the AI case is unprecedented.

The end of labor-time as the measure of value. If the general intellect is the primary productive force, then the amount of individual labor time that goes into a product is no longer a reliable measure of its value. This creates instability in the wage relation: workers are paid for their time, but the value produced is a function of the knowledge crystallized in the machinery, not the time spent operating it. The automation vs. augmentation debate is, in Marx’s terms, a debate about who captures the value produced when the general intellect becomes the primary productive force.

Knowledge as the new enclosure frontier. The Fragment implies that the enclosure movement of the AI age is an enclosure of cognitive commons: the conversion of collectively produced knowledge into proprietary productive capacity. Primitive accumulation in the AI age is not the appropriation of land but of training data—and its victims are not peasants but writers, artists, scientists, and software developers whose contributions to the cognitive commons have been absorbed without consent into commercial productive machinery.