Two fairly annoying events have heralded in the New Year. The first of these is that the Library Genesis project, an online repository of copies of books and academic articles, has been seeing major downtime, amidst rumours that they may shut down permanently. The second is the news that Meta, in training their Llama language model, had been using LibGen as a training corpus: a decision apparently approved by Zuckerberg himself. The idea of one of the wealthiest corporations in the world accessing literary, academic and scientific works for free, to feed into a sphere of technology that is already very divisive, has understandably upset a lot of people. What are we to make of this?

I shall begin by describing a phenomenon that has characterised the world economy since 2008—that of economic crisis. There are many theories of crisis within the broad spectrum of Marxian thought, as well as many intersections between them. For instance, some Marxists view the present crisis as a consequence of the tendency in the rate of profit to fall — which posits that as human labour comes to be replaced with machinery, a crisis of surplus value (and therefore profit) begins to emerge. Others disagree, viewing the lowered rate of profit as a symptom of overproduction, rather than the direct cause. Through this reading, the contradictions between the capitalist imperatives to produce commodities and to keep the costs of labour down end up cementing in a dynamic where production grows faster than consumption, leading to the production of commodities that cannot find buyers. These debates are as academically tedious as they are extensive; and some theorists, even when disagreeing about cause and effect, seem to agree about the visibility of multiple modes of crisis today. 

What does any of this have to do with generative AI? Regardless of what theory of crisis one subscribes to, our present reality is at least somewhat easier to characterise — capital today is simply unable (or unwilling) to find productive and profitable outlets for investment. The US and the EU, responding to the crisis of 2008 by opting for quantitative easing as monetary policy, have been begging capitalists to reverse this, to find some way out of crisis. Consequently, both venture capital and Big Tech have benefited immensely from the free money released into the financial system. They have put this money to good use by pumping it into tech platforms that act as unproductive, perennially loss-making middlemen; into outright scams like NFTs; and into the latest such iteration, generative AI. 

The abysmal profitability of many of these investment avenues has made it particularly urgent for capitalists to modify legislation, in order to cheapen the labour and resources that drive accumulation. Labour is thus rendered more “flexible” through being thrust out of the formal wage relationship; nature is transformed into fuel for production; social interaction is torn apart and reconstituted in ways that best suit the accumulation of data. Nowhere has this been more true than with generative AI, with its reliance on the Internet as a cheap source of mass training data for generative models; on low-wage annotators to align said models; and on immense amounts of energy, required both to train and deploy these models. 

Enter LibGen, or Library Genesis. LibGen, an online archive of literary and academic works—photocopied, scanned, manually transcribed—has been a source of much comfort and utility to the millions of users that frequent it for material. LibGen, as  this excellent history points out, emerged from a very particular milieu that existed in the Soviet Union. Driven by the desire for culture to be as accessible as possible, to enable “the cultural development of the masses”, books and literary works were kept very cheap; copyrights were often not recognised, particularly when authorising translations. At the same time, the Soviet bureaucracy’s widespread censorship of literary material led to the emergence of underground libraries, where readers would share, proliferate, and collect censored material. The obsession with collection, with hoarding, with the hungry archival of knowledge, all emerged from the knowledge that the state could render entire bodies of literature inaccessible. 

Given that the Internet has made LibGen a shade more accessible than Xerox-driven archiving in dingy Moscow cellars, it has seen multiple attacks over the years, including multiple lawsuits from publishers, academic or otherwise. But the Internet is rather hard to regulate; it isn’t hard to host mirrors for LibGen’s database, which isn’t even obscenely large—at around 30 terabytes, given a rough storage cost of €15/TB, the entire database would cost less than €500 to store. Based in Russia, LibGen is also rather good at hiding the individuals who run it, making individual retribution unlikely. This hasn’t stopped publishers from trying, and lawsuits and outages have been frequent; but the website continues to thrive, with frequent updates with new books, new URLs, and active peer-to-peer sharing.

Projects like LibGen tend to be somewhat controversial, especially amongst those of us who are (or who have friends who are) struggling writers or artists or musicians. In a world where this sort of piracy was the norm, the argument goes, there would be no way for these workers to ever make a living. We should be paying writers for their work; not doing so is theft, and what Meta are resorting to right now is precisely this—theft on a grand scale. How do we protect writers from this theft? The best method we have under the current system: copyright law. 

This argument is however rather defeatist, as well as detached from actually-existing copyright law. First, getting the obvious out of the way—piracy cannot be stopped, not without deeply authoritarian internet firewalls. Duplicating and sharing digital material is trivial, and attempts to slap on increasingly complex DRM instead end up becoming a nuisance to paying customers. With that out of the way, it is important to remember that capitalism is rather successful at creating schisms between workers. Copyright law achieves precisely this, by claiming to act as the last line of defence against the impoverishment of writers who are thrust into a dependence upon the invisible hand of the free market. It reifies writers and artists as property owners, and reifies their material interests as directly opposed to the cultural interests of the masses. This process has only grown stronger in the neoliberal period, which promises both greater immiseration and potentially greater payoffs, relying upon its ideological trappings to try and convince writers and artists that they are all temporarily embarrassed millionaires, rather than workers who deserve a living wage. We should be clear, however, that this is far from the only feasible way to ensure the reproduction of writers. Alternative models that work for both readers and writers are not impossible to imagine or achieve, even under capitalism. The academic model of state-guaranteed wages for open-access literary output, for instance, is one such actually-existing, semi-functional alternative.

Relying on copyright law, as it turns out, ends up being actively detrimental not just to consumers, but also to writers. As Cory Doctorow lucidly points out, by reinforcing the position of writers as businesses rather than as workers, we atomise them, giving them rights that can later trivially be bargained away from them. The function of publishing houses, the prime beneficiaries of this arrangement, has traditionally been to provide the fixed capital necessary for book printing, marketing, distribution, etc. In doing so, they have successfully cemented their position as mediators, giving them immense power over writers in a system that writers have little choice but to buy into. Publishers thus gain the ability to squeeze writers’ wages below their domestic minimum wage, or to outright ask writers to pay to publish. Nowhere has this parasitic tendency been more visible than in academia, where academics see nary a cent of the vast amounts of money that flow into the academic publishing houses that build enclosures around vast stores of knowledge that they played no role in creating. Above all, the construction and reinforcement of this regime of property rights serves only to enable the stunting of the cultural and material development of the masses—transforming the act of piracy into genuine political praxis. 

The birth of generative AI has highlighted the incredibly precarious nature of intellectual property rights, and how they have always served the accumulation of capital—with the interests of writers and academics themselves remaining an inconvenient afterthought. The law giveth and the law taketh away; and what we are witnessing today is precisely the law taking away, as these established relations see renegotiation in search of profit. Publishers, record labels, film studios and their ilk have been champing at the bit to sell their vast stores of media to generative AI firms (or, potentially, to train their own models). Meta’s turn to LibGen is rather amusing in this context. Had the countless petty lawsuits against LibGen actually succeeded at shutting it down, Meta could simply have paid publishing houses (and maybe an indirect pittance to writers), and achieved much the same result anyway. But they are fully aware that today, when capital is firmly fixated upon generative AI as the harbinger of a renewed capitalism, they needn’t even bother negotiating with yesterday’s capitalists. They will face at most a slap on their wrist for these very publicly acknowledged copyright violations. 

People’s annoyances with generative AI aren’t hard to understand. It is one of the most striking examples of the deskilling of labour in recent history, affecting hitherto well-compensated workers. The milieu in which it has emerged—the painfully cringe, Revenge of the Nerds-esque Leitkultur that permeates Silicon Valley—only rubs salt into this wound. But the substitution of writers with AI-generated slop is yet another step in the long history of the global devastation of skilled labour, and must be engaged with as such, rather than by getting lost in futile battles over property. Very few institutions are as deeply mutually imbricated with the capitalist mode of production as property rights are; they exist to serve the accumulation of capital, and can see rapid renegotiation when the site of accumulation shifts. This is precisely what is happening now, allowing Big Tech to ride roughshod over the interests of both writers and publishers. 

We must also remember that our concerns with these models lie not in how they are trained, but in how they are deployed. Academics have been working with human language data for decades. This is a good thing; the open availability of a myriad genres of language data have contributed immensely to research into linguistics, cognitive science, sociology, and so on. The open availability of literary and scientific works to the general public is also a good thing, unless we actively desire that the masses remain culturally, scientifically, and intellectually stunted.The use of machinery to deskill, discipline, and beat down labour is precisely that—an issue with how machinery is used. This process is fundamental to the capitalist mode of production, and must be framed as the assault on labour that it is, rather than relying on appeals to property rights. The use of generative AI to shed workers en masse must be fought through a principled Luddism; through an opposition to capital’s use of machinery to discipline and replace workers; and through building solidarities both with other workers and with the consumers of written text. More importantly, we must reject the cynical fiction that property rights are a necessary evil—and what better moment than now, when the Faustian bargain between artists and publishers has revealed itself to be entirely one-sided?