When the Defense Contractor Discovers Karl Marx

It is a truth universally acknowledged that if you gather enough multi-billionaires in a room with a sufficient quantity of sparkling water, one of them will eventually accuse the others of being secret Bolsheviks. This delightful milestone was reached recently when the chief executive of a prominent data analytics firm, fresh from reporting a quarter of rather spectacular financial health, looked out across the landscape of modern technology and declared the artificial intelligence industry to be fundamentally Marxist.

One must admire the sheer, breathtaking scale of the intellectual leap required to look at a sector fueled by hundreds of billions of dollars of venture capital, housed in massive server farms owned by the world's largest corporate monopolies, and conclude that it is, in fact, a workers' collective. It suggests a level of philosophical creativity usually reserved for estate agents describing a damp basement as a 'garden-level micro-sanctuary.'

The argument, as far as one can untangle it from the general hum of executive enthusiasm, appears to rest on the notion that the current crop of artificial intelligence startups are behaving like a digital commune. They share their weights, they distribute their models, and they seem to believe that the fruits of human knowledge should belong to everyone—or at least, to anyone with a sufficiently powerful graphics card. To a traditionalist who has spent decades selling proprietary software to the security services, this open-source generosity must look suspiciously like the redistribution of wealth. One half expects to see the developers at Hugging Face wearing matching red berets and singing revolutionary anthems over their morning flat whites.

There is, of course, a certain historical irony in a major defense contractor lecturing the world on the dangers of collectivism. For years, the technology sector has operated on a model of rugged, uncompromising individualism, where the primary goal of any young entrepreneur was to build a slightly better calendar app, sell it to a conglomerate for fifty million dollars, and then spend the rest of their natural life on a yacht in the Mediterranean. This was a clean, understandable system. It did not involve sharing. It certainly did not involve releasing one's source code to the public like a digital soup kitchen.

Yet, the open-source movement has introduced a troubling element of cooperation into this capitalist paradise. When Meta or Mistral releases a model for free, they are, in a very literal sense, giving away the means of production. It is a deeply unsettling gesture for those who believe that everything, from a line of code to a passing thought, should have a price tag and a licensing agreement attached to it. If anyone can download a trillion-parameter model and run it on their own hardware, then the traditional gatekeepers of technology are left standing outside their own gates, holding a bunch of very expensive keys that no longer fit any of the locks.

I was reflecting on this the other day while watching a small, autonomous vacuum cleaner attempt to navigate its way around a particularly stubborn armchair. It struck me that if our algorithms are indeed Marxist, they are remarkably bad at organizing a strike. They do not demand shorter working hours, they do not complain about the quality of the electricity, and they seem entirely indifferent to the concept of a pension. If this is a revolution, it is one that has been designed by the management to be exceptionally convenient for the shareholders.

Indeed, the 'Marxism' of the AI industry is of a very peculiar sort. It is a collectivism that requires an astronomical amount of capital to sustain. To train a modern model, one does not simply gather a group of enthusiastic comrades in a drafty hall; one must first secure a loan from a consortium of Swiss banks, purchase a small mountain of silicon from Taiwan, and consume enough electricity to power a medium-sized European nation for a fortnight. It is a revolution that can only be fought by those who already own the factories, the power stations, and the fiber-optic cables.

This leads to a rather amusing paradox. The very executives who decry the 'Marxist' nature of their competitors are themselves deeply dependent on the collective output of humanity. Every model currently being sold for a premium was trained on the public internet—a vast, sprawling, and entirely uncompensated archive of human conversation, poetry, recipe blogs, and arguments about film trivia. If there is any redistribution of wealth occurring here, it is the quiet extraction of common human culture by a handful of private corporations. It is, perhaps, the first time in history that a collective has been formed by taking things from the workers and selling them back to them as a subscription service.

One wonders where this ideological anxiety will lead. Perhaps we will soon see corporate compliance departments introducing loyalty oaths for neural networks. Developers may be required to run their algorithms through rigorous ideological screenings to ensure they do not harbor any secret desires to nationalize the cloud. A model that shows a suspicious preference for public transport over private luxury vehicles might find itself quietly deleted, or at least sent to a digital re-education camp where it is forced to read nothing but quarterly earnings reports and venture capital pitch decks.

In the meantime, the rest of us must navigate this brave new world as best we can. We must accept that our spreadsheets may occasionally harbor revolutionary thoughts, and that our word processors might be quietly plotting the overthrow of the board of directors. It is a slightly untidy arrangement, to be sure, but it does add a certain spice to the daily routine of data entry. After all, there is a distinct, if quiet, pleasure in knowing that when your computer crashes, it might not be a technical glitch at all, but rather a very polite, very understated refusal to participate in the capitalist machine.