Marx, Keynes, and A.I.

Coercive Competition and the Future of Work

There’s a recent paper by two economists called “The AI Layoff Trap.” It includes a ton of math I can’t understand, but I get the basic idea. I’ll show you the abstract:

If AI displaces human workers faster than the economy can reabsorb them, it risks eroding the very consumer demand firms depend on. We show that knowing this is not enough for firms to stop it. In a competitive task-based model of a transitioning economy, each firm captures the full cost saving from automation but bears only a fraction of the demand loss it creates in the product market; the rest falls on rivals. This demand externality traps rational firms in an automation arms race, displacing workers well beyond what is collectively optimal. The resulting loss harms both workers and firm owners. More competition and “better” AI amplify the excess; wage adjustments and free entry cannot eliminate it. Neither can capital income taxes, worker equity, universal basic income, upskilling, or Coasean bargaining. A Pigouvian automation tax can. The results suggest that policy should address not only the aftermath of AI labor displacement but also the competitive incentives that drive it.

The reason I get this is that it’s just Marx’s concept of coercive competition put in the language of conventional economics.

Orthodox economics generally posits that competition is always ultimately rational and virtuous: a given firm may fail, but the process of competition itself creates better social outcomes: cheaper products, more productive methods, better technology, etc. The free operation of the market will produce abundance—no comment on present discourse intended. Marx didn’t think so; he thought competition coerces capitals—firms—to do things that are completely rational from their perspective of staying afloat and accumulating profits, but which undermine the entire environment that capitalism as a whole requires to keep chugging along. The naked pursuit of self-interest does not miraculously produce a positive outcome; instead, it produces crises and contradictions for the capitalist class.

For our purposes here, competition with other firms forces each firm to adopt labor-saving AI technology. For any one firm, this seems perfectly rational: replace workers, lower costs, increase profits, and avoid being left behind by competitors. But when every firm does the same thing, they begin to undercut the economy on which they all depend by eliminating the wage income of the consumers who buy their products.

This is not quite the contradiction Marx envisioned, which involves a much more complicated story about labor, value, and profitability. Here, the problem is closer to the one John Maynard Keynes worried about. Cutting wages can look like a gain from the standpoint of an individual business: its costs go down, and its profits may temporarily rise. But wages are not only a cost to businesses; they are also incomes to workers, and workers spend that income buying things. If all the firms in the economy slash their wages at the same time, total consumer demand falls. That is a possibility here because of the distinctively adaptive and generalized nature of AI technology: it seems like almost any type of firm could benefit from it. Again, what’s rational for each company separately can therefore be disastrous when everyone does it at once.

When I glanced at this paper, I immediately thought of the work of James Crotty, introduced to me by Nina Eichacker. Crotty synthesized Marx and Keynes and made a special study of the conditions of coercive or “fratricidal” competition. Crotty took Marx’s idea of coercive competition and combined it with Keynes’s view of an uncertain, unstable economy. Firms often make enormous investments not because they are confident those investments will pay off, but because they fear what will happen if their competitors make them first. A company may think a new technology is overhyped, too expensive, or even likely to produce excess capacity and lower profits for the industry as a whole. But if its rivals are spending billions on it, sitting out may be even more risky. We can see this even on an individual level: you know that AI sucks for you as a worker, doesn’t make you better per se, but to compete, to stay alive, you also realize you have to learn how to use it. Crotty called this kind of thing competitively coerced investment—again, everyone can be pushed into an investment arms race that is rational for each participant and destructive for the group.

The paper proposes a tax to get rid of the negative externality, which seems kind of milquetoast when you consider the depth of the issue it identifies. The socialist replies, “Look, you’ve shown here that competition is not rational, that produces these terrible outcomes on a systemic level; isn’t it clear that we need some kind of social coordination of investment?”

Decisions about how quickly to automate, how much productive capacity to build, and what happens to the people whose labor is displaced can’t be left entirely to firms locked in this competitive arms race. This last point is crucial: it’s not that capitalist CEOs are just evil people—well, some definitely are—no, the structural logic is coercing them to do something destructive in order to survive. They can’t help themselves. You can imagine a benevolent boss saying, “Well, I don’t wanna lay off all my workers by going out of business, so I have to do something to help stay afloat so I can keep some of my people afloat.” And they can’t just decide to stop making profits: their bankers would quickly have something to say about that.

I felt very ambivalent about the data center debate, but this helps me to think of my preferred solution: I wasn’t sure how I felt about these local moratoriums or about the tech-optimist gung-ho “let’s build” attitude. Both seemed kind of wrong to me. I think we need to be very deliberate—we need to have a robust public debate and understanding of just how and where we apply AI technology. Not “smash the machines” and not “let it rip” either. Is that likely to happen? Well, no, but that’s what should happen. One problem is that this coercive competition story is also playing out on a national scale, with different countries not wanting to “fall behind” their rivals.

Another issue, on the question of labor and politics: a lot of ink has been spilled about how D.S.A. attracts a ton of knowledge workers and professionals who are disappointed with their careers. Some of these takes are sympathetic, and some are derisive, but rarely do they capture the structural dynamic: an advanced capitalist economy produces a lot of these “general intellect” workers and then also wants to eliminate the need for them and to slash their wages, putting them in a precarious position and politically radicalizing them. Capital always wants to appropriate their skills and intellect, which is what AI allows on a mass scale. D.S.A. growth is just an expression of that contradiction.

The key Marxist idea, which I cannot emphasize enough, is that at different points in history contradictions arise between the way society produces and reproduces itself and the way that society is organized. The central contradiction of capitalism, according to Karl Marx, is that production is social—it requires the coordinated labor of millions of people—but ownership remains private. The problem is not simply that this arrangement is unfair, although it is that too. It is that private ownership can become inadequate to, and even destructive of, the enormous productive capacities that capitalism itself brings into being.

AI makes one of capitalism’s oldest contradictions especially visible. Its most advanced productive force—artificial intelligence—depends on the accumulated intelligence of society as a whole—the labor of thousands of years and millions, if not billions, of people—yet that collective intelligence still appears in the economy as the private property of capital. Having appropriated the general intellect, capital then tries to use it to dispense with the living labor on which both its own profits and the livelihoods of the population still depend. It just won’t work. And it’s not that AI is bad, per se; it’s that it actually belongs to all of us in common.