Talking with Marx through a Machine
When the knowledge of society becomes machinery.
Karl Marx once imagined a future where society’s accumulated knowledge would live inside machines, guiding production while humans stood alongside the system rather than inside it. When he wrote those words, the most advanced machines in the world were steam engines and textile mills.
Tonight I found myself discussing that idea with a machine trained on the writings of millions of people.
Somewhere between those two moments—the notebook in 1858 and the conversation on a glowing screen—knowledge became machinery.
Everyone keeps calling what’s happening with AI automation.
The word feels wrong. It feels old.
Factories automated. Payroll systems automated. Scripts automated repetitive tasks long before anyone used the term AI. The concept is simple: a human designs a set of rules, and the machine executes them faster and more reliably.
Automation replaces procedures.
But when you sit down and talk with a modern AI system, that description begins to strain. You are not giving it a predefined procedure. You are describing an intention, and the system produces something that looks suspiciously like reasoning.
Write a paragraph about this topic. Explain this concept. Code an app that does that.
There is no rulebook written by engineers that explicitly describes how to do those things. Instead, the system generates responses from patterns learned across enormous amounts of data.
The machine is not executing instructions. It is producing judgments.
That difference is subtle, but it matters. Automation removed human hands from repetitive work. What we may be beginning to see is the removal of certain judgment loops.
One useful way to understand the shift is economic rather than technical. Classical economics describes production as a combination of labor, capital, and land. Automation historically meant capital replacing labor. A machine replaced a worker performing a fixed routine. But systems like this blur that boundary. An AI coding agent does not behave like a piece of capital equipment. It behaves more like a worker. It writes, plans, interprets instructions, and iterates on solutions.
Some have started calling this phenomenon synthetic labor—a new kind of labor input produced by machines. The machine is no longer simply a tool inside the process. It is beginning to look like a participant in the process.
And the strange thing is that someone noticed this direction almost two centuries ago.
In 1858, long before computers existed, Marx wrote a set of notebooks now known as the Grundrisse. The notebooks were unpublished for decades and only became widely known in the twentieth century, which gave it an almost prophetic reputation when rediscovered. Buried inside them is a short section that later became famous under the name The Fragment on Machines.
Marx imagines a stage of industrial development where machines no longer simply amplify human labor. Instead, they begin to incorporate scientific knowledge itself. Engineering, mathematics, and technique become embedded directly into machinery.
Production increasingly runs through what Marx called an “automatic system of machinery.” Human workers are still present, but their role changes. They supervise, regulate, and maintain processes that largely run themselves. As Marx puts it, the worker becomes little more than a “watchman and regulator of the production process.”
Then Marx makes an observation that must have sounded bizarre in the middle of the 19th century: if machines embody the productive knowledge of society, labor time can no longer be the real measure of wealth.
“As soon as labour in the direct form has ceased to be the great well-spring of wealth, labour time ceases and must cease to be its measure.”
— Karl Marx, Grundrisse, “Fragment on Machines” (1858)
Capitalism measures value through hours worked. But if production depends primarily on knowledge embedded in machines rather than human labor, that measurement begins to lose meaning.
Marx believed this created a deep contradiction inside capitalism. The system measures value through labor time, yet technological development steadily reduces the role of labor in production. As knowledge embedded in machinery becomes the real productive force, the framework built around labor begins to strain against the reality it helped create. He even hints at the ultimate implication. As industrial development advances, he writes, “capital thus works towards its own dissolution as the form dominating production.”
As Marx wrote in the Grundrisse:
“The theft of alien labour time, on which the present wealth is based, appears a miserable foundation in face of this new one created by large-scale industry itself.”
Marx called this accumulated knowledge the general intellect—the scientific and technical intelligence of society as a whole.
For Marx, this was mostly a theoretical horizon. Steam engines and textile machines were impressive, but they were still mechanical devices. They did not contain anything resembling intelligence.
Yet Marx pushed the logic of industrial development forward. If machines kept absorbing more human skill and knowledge, he reasoned, eventually the real productive force of society would be knowledge operating through machinery.
Marx even described machines as “organs of the human brain… the power of knowledge, objectified.” In the nineteenth century he was thinking about industrial machinery. But the description feels strangely familiar today.
That passage sat quietly in a notebook for decades.
Today, it reads less like theory and more like a description.
“They are organs of the human brain… the power of knowledge, objectified.”
— Karl Marx, Grundrisse, “Fragment on Machines” (1858)
Large language models are trained on enormous datasets: books, research papers, code repositories, conversations, images, diagrams. The intellectual output of millions of people becomes the raw material.
During training, the model compresses that material into a vast network of statistical relationships. What emerges is not a database of stored answers but a system capable of generating new text, explanations, and solutions by recombining patterns from that corpus.
In other words, a machine trained directly on the accumulated cognitive labor of humanity1.
This is not consciousness. It is not understanding in the human sense. But it is something new: a system that can synthesize large fragments of society’s intellectual output on demand.
When you ask it to summarize a scientific idea or draft a piece of code, you are interacting with something that resembles a compressed layer of collective knowledge.
If Marx were writing today, this is probably what he would have pointed to as the general intellect made tangible. Instead of labor power being the primary productive input, accumulated knowledge encoded in machines begins to drive production.
For the first time, the accumulated knowledge of society is not just stored in libraries or institutions. It is embedded in systems that can respond dynamically.
You can talk to it.
When people try to explain this, they often emphasize that AI systems are “just probability machines.” They generate outputs from probability distributions learned during training.
That explanation is technically correct. But it raises an uncomfortable thought: what if human cognition works in a broadly similar way?
Modern neuroscience increasingly describes the brain as a prediction engine. Neural networks in the brain constantly update internal models based on incoming information. Perception, reasoning, and language may all emerge from networks learning patterns across experience.
Strip away the biological details and the comparison becomes difficult to ignore.
Both humans and machine learning systems are large networks adjusting internal parameters based on data.
The differences are real. Humans learn through sensory experience, embodiment, and motivation. We explore the world, generate our own data, and update our models continuously. Current AI systems mostly process static datasets and respond when prompted.
But those differences increasingly look like problems of capability rather than design principle.
The basic mechanism—learning patterns from data through large networks—may be shared.
If that turns out to be true, the mystery of intelligence might be less about biology than we once believed.
Seen from a wider perspective, something interesting has been happening across human history.
At first, intelligence existed only inside individual minds.
Then language appeared, and knowledge began to accumulate across generations. Culture became a distributed memory system shared by entire societies.
Writing expanded that system further. Knowledge could persist across centuries.
Printing accelerated the process again.
Now we are building systems trained directly on that accumulated cultural output.
Each step moves more intelligence outside the individual organism.
This is why interacting with modern AI systems can feel uncanny. You are engaging with a system that has absorbed enormous fragments of collective human knowledge.
Something like the general intellect is beginning to take form.
Perhaps this is why the word automation feels insufficient to me.
Automation describes machines executing predefined procedures. What we are beginning to see is machines participating in the cognitive layer of production, writing, analyzing, designing, coding, explaining.
Automation replaced human hands.
This new category of systems may replace or replicate certain judgment loops.
The vocabulary of the industrial era does not quite capture that shift. We still call it automation because we have not yet invented a better word.
But when a machine trained on the accumulated thoughts of humanity can participate in reasoning through language, something deeper than automation may be happening.
Marx had a phrase for it. He called it the general intellect.
I should remind myself to use this as another argument for Universal Basic Income.