You On AI Field Guide · Dead Labor The You On AI Field Guide Home
TxtLowMedHigh
CONCEPT

Dead Labor

Marx’s term for the past human labor crystallized in machinery—which “vampire-like, only lives by sucking living labour”—and the most exact vocabulary available for naming what an AI model contains, who produced it, and why the value it generates flows to the owner of the weights rather than the worker who prompts them.
Dead labor is Karl Marx’s term for the past human effort crystallized in machinery and infrastructure—the labor that has been expended, transformed into a productive instrument, and now operates to generate value without the continuous expenditure of effort that originally produced it. Marx’s image for it was deliberately gothic: capital, he wrote in Capital, is “dead labour that, vampire-like, only lives by sucking living labour, and lives the more, the more labour it sucks.” The living labor of current workers combines with the dead labor embedded in the machine to produce output; the value that output generates accrues to the owner of the machine, not the worker who operates it. Applied to large language models, the concept achieves a precision that borders on the uncanny. A foundation model is the largest crystallization of dead labor in human history: billions of hours of human writing, coding, artistic production, and scientific reasoning, crystallized as model weights through a training process funded by capital. The living labor of every user who prompts the model combines with this dead labor to produce outputs whose value is captured by the owner of the weights. Marx did not anticipate AI, but he described its political economy with uncomfortable accuracy a century and a half early.
Dead Labor
Dead Labor

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI asks what the orange pill sees that is obscured by hype. One thing it sees is the labor structure behind the smooth interface. The AI product presents as a neutral, helpful service; the dead labor visible from behind the interface includes the writers whose prose was scraped without consent, the annotators in Nairobi and Manila who labeled data for wages below subsistence, the open-source developers whose code was absorbed into training corpora, and the creative professionals whose entire portfolios became, without their knowledge, part of a generative system now competing against them.

The concept also illuminates the specific dynamic of human-AI collaboration. When a practitioner uses an AI to produce output, they are combining their living labor—the prompt, the judgment, the contextual knowledge that makes the output useful—with the dead labor crystallized in the model. The ratio of living to dead labor shifts continuously as models improve: the same output that required significant living labor in 2023 requires less in 2025. Marx’s analysis of mechanization predicts this trajectory exactly: as the dead labor component increases relative to the living labor component, the value contributed by each unit of living labor falls, which is simultaneously a productivity gain (more output per hour) and a structural pressure on the wages of living laborers.

The Machinery Question
The Machinery Question

The cycle does not propose that practitioners refuse this dynamic. It proposes that they understand it—that seeing the dead labor in the machine is part of what taking the orange pill means. The practitioner who understands that they are combining their living labor with an enormous crystallization of other people’s past labor is better positioned to ask what that combination is worth, to whom, and on what terms.

Origin

Marx developed the distinction between dead and living labor in the first volume of Capital (1867), as part of his analysis of how machinery functions in the capitalist production process. The distinction rests on his labor theory of value: value is created by the expenditure of human labor power, measured in socially necessary labor time. Machinery, like all capital, does not create new value; it transfers value that was already embodied in it (the labor of the workers who built it) to the commodities produced with it, while living labor creates the surplus value that accumulates as profit.

The gothic imagery—the vampire, the undead—was deliberate. Marx chose it to name the specific dynamic by which past labor, once converted into capital, acquires the power to govern and extract value from present labor: the product of human activity becomes an alien power over human activity. This inversion—what he called the fetishism of capital, the personification of things and the reification of persons—is the specific feature of capitalist production that he wanted the category of dead labor to make visible.

The Fragment on Machines in the Grundrisse (1857-8) extends the analysis to the point at which the general intellect—the accumulated scientific and social knowledge of humanity—becomes the primary productive force, directly embodied in the machinery. At this point, Marx argued, the basis of production shifts from labor time to knowledge, and the contradiction between the social production of knowledge and its private appropriation as capital becomes the central antagonism of the age. The transition to AI is, on this analysis, not a new development but the culmination of a trajectory Marx described in the 1850s.

Key Ideas

The crystallization of labor in machines. Dead labor is not the absence of labor but its crystallized form: human effort that has been transformed into a productive capacity and is now deployed without requiring new expenditure of effort. The dead labor in a steam engine is the labor of the workers who built it; the dead labor in an AI model is the cognitive labor of the millions of humans whose writing, code, and creative production was absorbed into the training data.

The ratio of dead to living labor and the pressure on wages. As machinery improves, the ratio of dead labor to living labor in each unit of output rises. This means that each hour of living labor produces more but also that each hour of living labor contributes a smaller proportion of the value in the final product. Marx’s analysis predicts that AI improvement will increase productivity while simultaneously reducing the bargaining power of human workers whose skills the AI can replicate.

The ownership question. Dead labor generates no value on its own; it requires living labor to animate it. But the value generated by the combination accrues to the owner of the dead labor—the owner of the means of production—not the worker whose living labor makes it productive. This is not an accident or an injustice in the moral sense; it is the defining structure of capitalist production. Marx’s point is that naming the structure is the precondition for changing it.

Reanimation and accountability. The vampire image names a specific moral claim: the dead labor in capital was produced by actual human beings, most of whom received no benefit from its conversion into productive machinery. The writers and artists whose work trains AI models are in the position of the artisan whose tools were expropriated during primitive accumulation: their labor has been crystallized into an apparatus that now competes against them. Commodity fetishism makes this invisible; ideology critique makes it visible again.

Debates & Critiques

The central debate about the dead labor analysis as applied to AI concerns the labor theory of value on which it rests. Most contemporary economists regard the labor theory of value as untenable as a theory of price formation; there is no reliable mechanism by which prices converge on labor values in real markets. Defenders of Marx’s analysis argue that the theory is not a price theory but a social ontology: a claim about the social relations that constitute value in a capitalist society, not a claim about how prices are determined. On this reading, “whose labor is in this model?” is not a technical question but a political one, and the answer matters for accountability, consent, and the distribution of benefits. The practical policy debate concerns intellectual property: do writers and artists whose work was scraped to train a model have a claim on the value that model generates? Legal systems are actively grappling with this question; Marx’s framework offers the sharpest conceptual vocabulary for it, even for those who reject his political conclusions. Primitive accumulation analysis connects the dead labor concept to the historical parallel with the enclosure movement, where the conversion of common land to private property was a precondition for industrial capitalism, just as the conversion of the cognitive commons to proprietary training data is the precondition for AI capital.

Further Reading

  1. Karl Marx, Capital: A Critique of Political Economy, Volume I (1867; Penguin Classics edition, trans. Ben Fowkes, 1990) — especially Chapter 15, “Machinery and Large-Scale Industry”
  2. Karl Marx, Grundrisse (1857-8; Penguin Classics edition, trans. Martin Nicolaus, 1993) — the “Fragment on Machines,” pp. 690–712
  3. Matteo Pasquinelli, The Eye of the Master: A Social History of Artificial Intelligence (Verso, 2023)
  4. Nick Dyer-Witheford, Atle Mikkola Kjøsen & James Steinhoff, Inhuman Power: Artificial Intelligence and the Future of Capitalism (Pluto Press, 2019)
  5. Kate Crawford, Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence (Yale University Press, 2021)
Explore more
Browse the full You On AI Field Guide — over 8,500 entries
← Home0%
CONCEPTBook →