One absurd thing with so-called “Leftists for AI” is that beyond pretending that the hype-market for LLMs and their capital-conducive nature that is shaped rather than being an immutable property of language models, “is” metaphysically what “AI” is, and often using this to justify putting out slop “art” to promote their stance, they act as if we’re reaching a higher phase of communism overnight.
Rather than evaluating intellectual property in terms of its actual results and potential results under current concrete conditions, they metaphysically characterize it as inherently containing a sort of “bourgeois contamination” based on what its absence could give us under ideal circumstances. Big tech being inconsistently absolved of copyright infringement when smaller players and harmless people aren’t, isn’t surprising. But this does not mean — especially when you have no other alternative to provide beyond a promise of what it would be like in a far-future socialism — that the strategy should be pretending IP does not exist, and only so when the “victim” is specifically not big tech.
The purpose of a system is what it does, and laws do not exist in a vacuum. If IP could be weilded against the usage usually perpetrated by the status quo, its purpose changes. Pretending otherwise while not providing any meaningful alternative, beyond promising “collective ownership of AI”, which, under capitalism, would not prevent the work of artists and writers from being used in ways that go against their interests and their own craft, is not really revolutionary. It’s closer to thinking that having enough worker-owned cooperatives under capitalism will lead to socialism.
Many “Leftists for AI” take capitalists’ own ideology of “resources”, where what is produced is entirely separated from who produces it and how it is produced. They might, for instance, think that LLMs (regardless of who owns it) scraping Wikipedia and presenting “summaries” of it is some form of “good”, when in reality, Wikipedia is a digital commons sustained by voluntary editors, who are a subset of readers, which such a “summary service” will take away from. Such a “service” has a contradiction that is rather characteristic of almost any kind of capitalist pursuit which requires enclosing the commons without completely doing away with them: in taking “resources” while thwarting what lead to their production and maintenance, they damage the very ecosystem they need in order to be useful.
This is not unlike when these same “Leftists for AI” make/write “AI”-generated memes, “art”, banners, posters, and even articles. The LLMs learn from and rely on real artists, writers, etc, and their focus on the output of LLMs while trying to supersede what makes it possible is a part of the pattern where the “resource” is taken and the “production/maintenance system” that made it possible is left on a spectrum from ignored to damaged.
This is not a good parallel to the industrial revolution or mere “innovation” in terms of tools. Cars can replace horses in that cars collectively don’t directly rely on horse riders and horses in order to develop. Even if all horses went instinct and horse riders ceased to be a thing, cars can keep on developing in new ways. This is not true of LLMs, the output of which is known to degrade when fed their own input. Cars superseded horses, but LLMs still need real human labor.
This pattern of parasitism would be true even if all LLMs were suddenly to be “ethically-trained” and distributed under “ethical” licenses, yet another “bourgeois tool”, without changing how they are actually produced, presented, given interfaces, contextualized, etc. But even that is a fantasy right now. Big tech still dominates the “AI” industry, and a hand-wave at “collective ownership” (under capitalism) without a substantial strategy and promising practice that reaches a comparable scale is not something that someone who actually cares about their work will wait a million years for. It is in the interest of an artist for real people to see their work and respond and engage in a process of adaptation and learning, as it is in the interest of writers, and contributors to the digital commons in maintaining a relatively stable system that allows the “resources” that said commons produce to be produced at all and maintained for use.
You could say they could just use LLMs differently, separating it from its hype, marketing and default interfaces, and reducing resource usage in ways proportional to the importance of the task at hand, and only when LLMs are actually even needed for that. And I agree! I am not exactly “anti-AI” or even “anti-LLM”, but “anti-LLM-as-AI”. And my problem with “Leftists for AI” is exactly this. Only a few of them understand the mathematics underlying LLMs, and even then, the prevalent way in which LLMs are presented and given interfaces, and are justified being given massive, disproportionate resources over meeting basic needs, is something they often seemingly accept wholesale.
Sure, you might nominally say you don’t accept all this, but if you still use ChatGPT’s chatbot interface to promote your “Leftists for AI” stance, I won’t believe you.
You could do all sorts of things for yourself by self-hosting even “ethically-trained” models, and it might feel “liberating” and “democratic” to you in that it gives you the idea that you can make it act like your employee without having to have existing capital (which is ironic, since tokens are currently heavily subsidized and will likely not stay this way forever), but if you still rely on model training that requires building data centers in places where the portion of drinkable water is low and they put a strain on the amount of drinkable water available (you can’t drink vaporized water that later joins the river, nor can you drink seawater), which your “for AI” stance in this context would require to some degree, which disproportionately affects the impoverished working class, I won’t believe you are actually concerned about the emancipation of the working class (e.g. see Shamsuddin Illius’s DIGITAL DREAMS, PARCHED REALITY: The hidden cost of Bangladesh’s data industry gold rush written for The Climate Watch).
Rather than actually challenging the producer-consumer relationship that is so suitable for capital in mainstream software design (which vibe-coded software do not automatically challenge, since the actual software will almost certainly still have such a relationship intact, as that is the kind of data LLMs are trained on), and engineering software under democratic processes and with universality in mind, using a mere broader consumption of software as a form of “radical praxis” is exactly what benefits capitalists. It is not a break from catering to their interests.
You could say you only use open-source models, but that only demonstrates political illiteracy regarding open-sourcing. Open-source and free software, both in the sense of the Open Source Initiative (OSI) and Free Software Foundation (FSF), and beyond, are “open” and “free” in the libertarian sense. Not just nominally, but in practice.
The OSI’s definition of open-source, often peddled by “right-libertarian” open-source purists, which also dominates how open-source is broadly done, prohibits “discrimination against persons and groups” (fifth criteria) and “discrimination against fields of endeavor” (sixth criteria). So, if your field of endeavor is Zionism, the license prevents someone fitting the OSI’s definition from “discriminating against” your usage and field of endeavor. If your field of endeavor is taking a piece of software meant for collective use and development and turning it into a product, the survival of which relies on parasitism on the original collective system, the OSI says that’s still a part of what open source is for. It is “free” as in the neoliberal “free” market.
The FSF’s more “restrictive” GNU (A)GPL licenses may prevent some of the latter, especially if the commercialized product does not propagate the same license, but Richard Stallman didn’t found FSF and didn’t birth these licenses for “collective emancipation”. Even his “anti-capitalist” writing forefronts individual freedom from corporations, rather than collective counter-power against corporations and capitalists more broadly. Thus, while failing the OSI’s test for “openness”, the FSF’s stated purpose seemingly aligns much more with what it actually does. I could distribute my software under AGPL, and might possibly even manage to get legal protection if someone were to violate the license, but it has no explicit clause preventing, say, redistribution centering individual consumption in ways that represent the same kind of parasitism, whether said form of parasitism directly makes the redistributor themself money or not. As Richard Stallman would say, this is “free” as in “free speech”, which, in this context, means exactly that kind of individual freedom.
“This will accelerate technological development and deepen the contradictions of capitalism, thus leading us closer to socialism”, won’t work as a cop-out either. Because there’s no sustained socialism without collective agency, and if you can overlook some portion of the working class not even having safe drinking water and dying off of preventable disease at disproportionate rates, let alone contribute to said agency, your “us” is a chauvinist one that might as well leave the outgroup behind while forever promising eventual prosperity.
In spite of superficially changing the ownership model, or at least nominally claiming to, or at least promising it in a revolution they themselves are too comfortable to fight for, “Leftists for AI” manage to hoard in exactly how capitalism treats “resources” and “democratization” to sustain itself. It’s using the language of collective liberation, possibly even intending it, while supporting, and, to some degree, contributing, to a furthering enclosure of the digital commons by big tech capital. It’s a short-sighted view that will bite themselves in the back in the long-run, and, as privatizing the benefits and socializing the costs goes, they likely won’t be the ones paying the larger share of the price.