No figure in recent decades has done more to direct public attention and left-wing energy to universalistic left-wing policy proposals than Bernie Sanders. During his two runs for the presidency, he introduced “Medicare for All” into the lexicon of American politics. Now he’s made waves with a proposal to create a sovereign wealth fund by nationalizing half the stock of major AI companies.

His rationale is simple and compelling. “Since A.I. is built on the collective knowledge of humanity,” Senator Sanders writes, “the wealth it generates must benefit humanity.”

Several prominent tech executives have made apocalyptic predictions about how AI-fueled automation will sweep across the economy, destroying a massive percentage of existing jobs. The industry’s critics argue that most of this is hype. We’ll see.

If it turns out that tech executives are telling the truth about the devastating effects their products will have on the livelihoods of massive swaths of the American population, they can hardly turn around and complain that it’s unfair for their industry to be singled out for partial nationalization. If I build a robot that stomps around my neighborhood, destroying my neighbor’s houses and cars and workplaces, and then I sell the robot for hundreds of billions (or trillions) of dollars, I shouldn’t complain if a judge forces me to put half the profits into a fund to help those neighbors rebuild their lives. I should be grateful that he let me keep the other half.

Libertarians may be tempted to argue that no one deserves redress because of suffering incurred by automation and other by-products of letting the chips fall where they may in a free market, and that taking away some of the stock of OpenAI or Anthropic against the will of the current owners is a morally unacceptable transgression against property rights. Even on classical libertarian principles, though, it’s far from clear that this argument would go through in this case.

Twentieth-century libertarian thinkers like Robert Nozick and Murray Rothbard built on the seventeenth-century theories of John Locke. The Lockean conception of property rights emphasizes origins. If Person A buys something from Person B who was given it as a gift by Person C, the legitimacy of the whole chain depends on C’s claim. Someone who starts a farm by homesteading previously unowned land, for example, is engaged in what Nozick called a “just act of original acquisition.”

Some of the most thoughtful libertarians have realized that, to be consistent, in some cases they have to recognize collective claims to just original acquisition. Libertarian philosopher Roderick Long offers this example:

Consider a village near a lake. It is common for the villagers to walk down to the lake to go fishing. In the early days of the community it’s hard to get to the lake because of all the bushes and fallen branches in the way. But over time, the way is cleared and a path forms — not through any centrally coordinated effort, but simply as a result of all the individuals walking that way day after day.

The cleared path is the product of labor — not any individual’s labor, but all of them together. If one villager decided to take advantage of the now-created path by setting up a gate and charging tolls, he would be violating the collective property right that the villagers together have earned.

This is exactly what John Locke thought a just act of original acquisition looked like. Person C mixes his labor with the natural world and thus forms a property claim that can ground any subsequent claims made by B and A. In this case, however, Person C is a whole village.

If you believe that intellectual property works like property claims on objects in the physical world, then the use of the collective intellectual products of the human race as the foundation of large language models (LLMs) looks a great deal like Long’s village-created path to the lake. Trying to disentangle which individual intellectual property owners are owed how much of the product would be a fool’s errand. But if we aren’t going to rip out the toll booth, turning half of the tolls into a wealth fund for the whole village might be the beginning of justice.

On the other hand, if you don’t believe that intellectual property claims have the same force as claims to material property, Anthropic and Open AI and the rest have no legitimate claim to proprietary algorithms. If public policy went that way, the question of how to share these firms’ profits would quickly become moot.

Finally, if you reject libertarianism root and branch and believe as I do that questions of how property should be distributed are downstream of questions about equality, dignity, and other fundamental human values, the case for partial (or perhaps even full) nationalization of AI companies becomes irresistible. Mainstream tech boosters might try to argue that the industry should be allowed to run its course without any interference as dramatic as what Senator Sanders proposed on the grounds that automation will tend to work out to the long-term benefit of the human race. Last year, for example, Bill Gates made headlines by predicting that the development of AI would lead to the shortening of the workweek to two or three days.

But socialists know better. At this point, we have centuries of relevant data points, and we know that this is never how things have played out in the history of capitalism. As long as the means of production are owned by a private class of nonworkers, if the work it took a thousand people five days to do now takes only two and a half, capitalists have no incentive to keep paying everyone just as much as ever for working two and a half days and every incentive to lay off half the workforce and keep the remaining five hundred workers as hard at work as ever. In the United States, it’s been more than eight decades since the Fair Labor Standards Act capped the workweek at forty hours. The free market hasn’t reduced that number by a minute since then, and that certainly isn’t because technological progress stopped during the Great Depression.

Conversely, some of my friends and comrades on the Left would probably prefer that, instead of nationalizing these firms, we simply direct the Air Force to bomb all the data centers and salt the earth so nothing can grow there again. I’ll admit that a large part of me concurs. Seeing undergraduates “write” papers with the “help” of LLMs makes me despair for the human race, and seeing older writers and journalists do the equivalent fills me with rage. Even in a world that enforced appropriate taboos against this behavior, though, many use cases of AI would remain stubbornly useful.

One of the most basic premises of historical materialism is that no society ever has shrunk or ever will voluntarily shrink its productive forces. This is unlikely to be the first case. The technology is here to stay, for better or worse. Our best shot at “for better” is collective option to seize hold of the machinery and operate it for the benefit of all.