The Fragment on Machines: Marx Described Automation in 1858

I came to Karl Marx on machines sideways, not through politics but through a question about AI that I could not get out of my head: what exactly is a model trained on? The answer — the collective written, drawn, coded and photographed output of an enormous number of people, almost none of whom were asked — kept pointing at a shape I half-recognised. I found the shape described, in a rough and unfinished way, in a set of notebooks written in London in the winter of 1857–58. Marx never published them. They were not printed in German until 1939, in Moscow, with the second volume following in 1941, and a complete English translation did not appear until 1973. The notebooks are known as the Grundrisse, and buried inside them is a passage that later readers named the Fragment on Machines.

What the Fragment actually says

The Fragment sits in the last of the notebooks, in a long, digressive stretch about fixed capital — machinery, plant, the equipment a factory owns. Marx is trying to work out what happens to his own theory when machines get very good. His reasoning runs roughly like this.

In craft production, the tool is an extension of the worker. The skill lives in the human being; the tool obeys. In developed industry this inverts. The machine is the active thing, and the worker becomes the machine's accessory — feeding it, watching it, correcting it. Marx puts it in a line that survives the translation intact: the human being, he writes, “steps to the side of the production process instead of being its chief actor,” coming to relate to production “more as watchman and regulator.” That is verbatim from the standard English translation, and it is the sentence everyone quotes, for good reason. Anyone who has stood next to an automated line, or watched a colleague spend their day approving outputs that something else generated, recognises the posture.

But the more interesting move comes next. Marx asks where the machine's capability came from. Not from the machine. It came from accumulated human knowledge — generations of experiment, failure, apprenticeship, shared technique, science that was itself a social product. That knowledge has been poured into steel and gearing. His phrase for the machines is exact and worth sitting with: they are “organs of the human brain, created by the human hand; the power of knowledge, objectified.”

And then the conclusion that gives the passage its charge. The growth of fixed capital, he says, “indicates to what degree general social knowledge has become a direct force of production.” In the same stretch he uses the English phrase general intellect — writing in German, he reaches for English — to describe the collective knowledge and know-how of a society, now operating not through individual heads but through the machinery those heads collectively built.

The machine is not a rival intelligence. It is our own intelligence, accumulated over centuries, handed over — and then charged back to us as rent.

The general intellect, plainly

Strip the vocabulary and the idea is simple. Knowledge is social. Nobody invents a loom, a compiler, or a vaccine alone; every technical act stands on an inheritance that no individual paid for and no individual owns. When that inheritance gets embodied in a machine, something odd happens to who controls it. The knowledge was everyone's. The machine belongs to whoever bought it.

Marx thought this produced a contradiction that capitalism could not metabolise. A system that measures value by labour time keeps building machines that make labour time irrelevant to output. He saw two possible readings of that. One was liberation: if wealth no longer depends on hours worked, free time rather than work time becomes the real measure of what a society has, and that is an argument about abundance, not misery. The other was that the system would defend itself — keeping labour time as the measure even after it had stopped making sense, and forcing people to keep proving their worth in hours while machines did the producing. Anyone who has watched a company automate a function and then increase the surveillance on whoever remains knows that the second reading did not die out.

Why it maps onto AI so uncomfortably

A large language model is, functionally, a compression of an enormous quantity of human output. Books, forums, documentation, arguments, code, journalism, the accumulated habit of how people explain things to each other. That corpus is about as close to a literal general intellect as anything humans have built. And then it is trained, weighted, wrapped in an API, and priced.

The structural fit is uncanny. Collective knowledge, objectified in a privately owned machine, sold back to the people whose knowledge it is, with the returns flowing to whoever owns the machine. The worker does not disappear; they step to the side. They stop being the chief actor and become the watchman and regulator — reviewing, approving, cleaning up. Marx's sentence describes a loom operator in Lancashire and a person checking model output in 2026 with the same words, which is either a profound insight or a coincidence of abstraction. I think it is somewhere in between, and I want to be careful about which.

What the Fragment is not

Here is where the honest reading has to push back against the enthusiastic one.

• Marx was describing industrial machinery, not computation. He had steam, iron and the factory system in front of him. He was not gesturing at neural networks; he was working out the consequences of the spinning mule and the railway. Reading “he predicted AI” into the Fragment requires ignoring what he was actually looking at.
• It was a notebook, not an argument. The Grundrisse is a rough draft — Marx thinking on paper, for himself, at speed, without an editor or a reader in view. He did not publish it. Passages contradict each other. When he reworked this material for Capital, the machinery analysis that made it into print looks materially different, and some of the Fragment's boldest claims are simply absent.
• Serious scholars disagree about how much weight it bears. Michael Heinrich reads the Fragment as a misconception Marx later corrected — a line of thought abandoned rather than suppressed. Antonio Negri and the Italian autonomist tradition built a whole politics on it and treat it as more important than Capital. David Harvey cautions that Marx used the phrase “general intellect” essentially once in his entire body of work, which is thin ground for a theory. David McLellan called the Grundrisse the centrepiece of Marx's thought. These are not small differences of emphasis; they are a live argument about what the text is.
• Prediction is the wrong frame. The Fragment does not forecast. It identifies a pattern. Those are different intellectual products, and conflating them flatters Marx in a way that makes him easier to dismiss.

I would add one more caution, aimed at myself as much as anyone. The Fragment is unusually quotable, and unusually vague. That combination is dangerous — it lets you find in it whatever you brought. If a passage seems to confirm everything you already believed about your own decade, that is a reason to read it harder, not softer.

The part that survives the objections

So what is left after all that stripping? Something narrower than a prophecy and more useful than a slogan: a name for a recurring structure. Shared knowledge gets embodied in privately owned machines, and the ownership of the machine — not the origin of the knowledge — decides who captures the gain.

That structure is not unique to any technology, which is exactly why it is worth naming. It is the same shape I keep finding when I look at how technology gets captured: the capability is genuinely new, genuinely collective in origin, and the distribution of its benefits is settled by ownership arrangements that have nothing to do with who built the capability. It is the structure underneath the question of whether were the Luddites right — they were not opposing machinery as such; they were contesting who the machinery's gains belonged to, which is a question the Fragment takes seriously and most histories of them do not. It is the continuity running from from cotton mill to smartphone, where the same arrangement reappears in each new material. And it is one plausible explanation for the productivity paradox — why extraordinary technical capability keeps failing to show up as broadly shared prosperity, decade after decade, in ways that surprise economists each time.

The Fragment's value is not that it saw our machines coming. It is that it described the ownership question we keep forgetting to ask about every machine.

Reading it if you disagree with Marx

You can find Marx's economics wrong in its essentials — his theory of value, his historical stages, his politics — and still get something from this passage, because the observation it contains does not depend on the system around it. That observation is descriptive: technical capability has social origins, and its ownership is determined separately from those origins. You can accept that and conclude that private ownership is the right arrangement, because it funds the machines and rewards the risk. That is a real argument and a serious one. What you cannot easily do, once you have read the Fragment, is pretend the question does not exist — that a model's capability springs from nowhere, or that the corpus it learned from was nobody's.

The notebooks were never meant to be read. They are unfinished, inconsistent, and in places plainly superseded by what their author wrote later. But a rough sketch can still catch a shape that a finished argument misses. Marx sat down in 1858 to think about looms and ended up describing an arrangement — our knowledge, their machine, their returns — that we are now negotiating again, at a scale he could not have imagined, with the same question unresolved.