Undead Digital Labor and the General Intellect – A Conversation on AI between Tiziana Terranova and Daniël de Zeeuw

Daniël de Zeeuw: A combination of genuine concerns and moral panics over AI (often fueled by the tech moguls themselves) has reached a new high in recent years, particularly with the introduction of ChatGPT by OpenAI. For some it represents the specter of the singularity hanging like an evolutionary cloud over humanity; while others sense the threat to their jobs, from journalists to designers, programmers, and actors; and yet others worry over their students using the technology to write their essays. Beyond the concern for employment and the invisible-precarious labor that goes into classifying and curating AI, there hasn’t been much attention to the ‘political economy’ side of these developments, specifically the free digital labor that produces the public texts and images on which large language models (LLMs) are typically trained.

As Julie E. Cohen shows in Between Truth and Power: The Legal Constructions of Informational Capitalism (2019), the notion of the web as a public domain already played a key legitimating and constituent role in the data-extractivist practices of platform capitalism, legally framing personal data ‘as a pool of materials that may be freely appropriated as inputs to economic production’ (48-9), and now extends to the appropriation of training data from the public web for AI tools like ChatGPT. She refers to this capitalist rendering of the public sphere as the ‘biopolitical public domain’. It is also at this intersection between labor, capital, and digital technologies that Terranova has conducted her research spanning more than two decades. For this conversation I asked her about the potentially renewed relevance of (post)workerist debates on immaterial labor, the social factory, and the general intellect for grasping current developments in generative AI and other new ‘synthetic media’.

You started writing about the Internet long before it evolved (or devolved) into, as you refer to it in your recent bundle of essays, the ‘Corporate Platform Complex’ (After the Internet: Digital Networks between Capital and the Common, published by Semiotext(e) in 2022). I wanted to start with the main title of the book, which suggests that it no longer makes sense to think about the online environments we inhabit today as ‘the Internet’ – and perhaps it never was. In any case, if the ‘good old’ Internet imaginary is relegated to the status of a historical relic or curiosity for early-web scholars – including all the emancipatory energies invested in it, like your vision of a digital commons beyond capitalist extraction – what is it exactly that we are currently confronting, and where have these energies gone today?

Tiziana Terranova: Of course, to say that we live “after the Internet” is a provocation. Internet protocols are still there doing the important work of making it possible for different networks and applications to connect to each other. Internet protocols are still the standards that make it possible most of the time to act as if there was only one giant global network rather than a patchwork of networks and machines. The Internet however was more than a set of protocols. It also stood for the possibility to overcome centralization, for unprecedented potential of many-to-many communication, for the democratization of cultural production – and even for a while for a post-capitalist economy. This kind of Internet existed in political discussions and aspirations, and that is the kind of Internet in whose aftermath we are living. It persists as an aspiration and a possibility, but we have to take into account the fact that the political economy of the medium has drastically changed – with enormous concentrations of power.

At the same time, in the 1980s and 1990s the Internet was actually accessed by a small segment of the world population which was overall also socially and (geo)politically quite homogeneous. One of the most important lessons that I have learned from my encounter with post-workerist Marxism, however, is that there is no ideal form of political organization or thinking that one must strive to get back to, but that all changes are the result of struggles and that the terrain is always changing. So yes, the 20th century Internet was mostly characterized by a certain trust into the emancipatory potential of digital technologies. Very few people, however, had the skills and technologies that were needed to participate. Today, we know that around five billion people use the Internet. So massive concentration of power and unprecedented distribution of access to distributed digital communication (one-to-one; one-to-many; and many-to-many). I think all the energies right now are being consumed by understanding what it means to deal with this paradox. Nick Dyer-Witheford and other scholars involved in the riot platform research project (PPPR) make a good point. Protests, riots, dissent are at an historical high, but not all riots are emancipatory, some others are quite reactionary too. Control capacities have also increased exponentially, but they are not increasing stability.

I do not think that we should stop being inspired by the common(s) either. There is no way out of the multiple crises of the present that does not involve a change in regimes of property, relation to natural resources, forms of government and the mode of production of wealth. The common(s) is still essential as a concept to me.

DZ: My earlier question is somewhat connected to the larger question of what (supposedly) comes ‘after capitalism’, as explored by McKenzie Wark in Capital is Dead: Is this Something Worse? (2020) and more recently Yanis Varoufakis in his book Techno-Feudalism: What Killed Capitalism (2023). In the early 2000s you published pioneering work on the Internet as harboring a new form of digital (often free) labor. The category of digital labor of course very much still operates within a Marxist analysis of capitalism, whereas Varoufakis for example prefers the label of ‘cloud serfs’. Is whatever comes after the Internet still capitalist (and digital interaction still ‘labor’) in any meaningful sense.

TT: Yes, I think it is capitalism, but maybe it is no longer liberal or neoliberal capitalism, but something that could be called technoliberalism. Liberal, neoliberal, and technoliberal express different variations with relations to the political rationality that supports, enables and governs a capitalist economy. In Black Marxism, Cedric Robinson points out how while the logic of capital persists (accumulation by means of dispossession and exploitation), there is no such thing as a capitalist class that changes with time while maintaining continuity. Different types of capitalists arise at different points and they become hegemonic – that is they become the cultural and political leaders of the capitalist classes (the owners of capital).

It is true that capitalists today form highly connected and dense clusters, which makes it easier for them to cooperate rather than compete when it is in the interests of their class. But these individuals are likely not of the same kind that ruled the world thirty years ago. I use the term that Neda Atanasoski and Kalindi Vora use in Surrogate Humanity: technoliberalism. Adam Fish also uses this term in Technoliberalism and the end of participatory culture in the United States (2016.) Barbrook and Andy Cameron started to describe it in the 1990s when they talked about the Californian ideology. It is much more developed now and it is much clearer how it also perpetuates a certain understanding of the Human and Freedom that are deeply racialized and gendered. Technoliberalists exists even in nations such as China, where a form of non-liberal state capitalism is at work. And there is also still labor, clearly. There is a technical composition of the labor force according to its position in the global and social division of labor. There is also a widespread reliance of the post-digital economy on what I called free labor and which is always a form of social cooperation. And yet labor no longer constitutes the kind of unifying political category that it was – at least for some time (the 19th and the 20th century) somewhere (mostly the Global North) and for some people (waged workers in industrial jobs). There is a profound crisis of modern political thinking – not so much I think with relation to the concept of capital, but that of labor.

Capitalist realism still means that you believe that capitalism exists. What you no longer believe is that there are workers that can potentially unite to change the course of things. Rather than being a pure economic and rational force, capitalism has proven itself to have relied structurally on processes of racialization and intersexual domination, colonial dispossession and patriarchal oppression that deeply divide workers and set some above others. This troubles a concept of labor that was based in the idea of unity – objectively given by the dialectical relation to capital and politically achieved through the development of revolutionary oppositional consciousness.

DZ: Beyond the larger question of capitalism, it is clear that social media platforms are extracting value from our acts of social, affective, and cultural production, and of users freely adding and sharing content for others to engage with: from writing film reviews to live streaming and posting selfies. In other words, a large part of online textual and (audio)visual content online is still user-generated. You analyzed these new forms of digital labor using the post-workerist Marxist framework of the ‘social factory’ and the ‘general intellect’ originally developed by the Italian autonomist movement, including people like Antonio Negri, Paolo Virno, and Maurizio Lazzarato. Now, as we became acutely aware of over the last two years, it is no longer just platforms extracting value from user-generated content but also machine-learning algorithms and tools based on LLMs like ChatGPT that are trained on this freely available, public content. That is, they are trained on the aggregate corpus of scraped web data generated by the sort of free digital labor described in your work.

Do AI tools like ChatGPT represent some kind of next phase in the exploitation of immaterial labor on a higher level of abstraction and immanence, namely as training data that fuels the recombinant powers of AI? In other words, could we understand this new enclosure of the commons as a sort of ‘plagiarism of the general intellect’? And in addition, to the extent that generative AI capitalizes on the potentialities or virtualities of social, cultural and intellectual production, exploiting as of yet-unrealized combinations of linguistic or visual expression, could we say, using its own technical vocabulary, that it is in fact colonizing humanity’s ‘latent space’? Are we seeing the real technological subsumption of what Virno referred to in A Grammar of the Multitude (2004) as ‘the subjective spirit of invention, invention-power’ (21), the potential to act, create, write, and so on?

TT: I think you said it very well. It is quite evident really. There would be no ChatGPT without the massive amounts of digitized textual and linguistic output that is out there in the public domain. It is data captured, stored, ‘cleaned’ by workers, and used to train the models. I think that the new models are trying to minimize human labor but it still there somewhere in the cycle. It is an important new technology and it is concentrated in the hands of the technoliberal capitalist class who have benefited from the kind of terms of services that users subscribe to when they open an account with a platform – when they are not freely harvested from the Web. And they are using it to increase return on investment, to further their hegemonic position in the global capitalist class, and selling it to armies, corporations, states, and other institutions that will use it to their purposes. The Israeli army has been using AI to increase its target production capabilities – in ways that have accelerated the destruction of the Gaza strip, for example.

But is it plagiarism? Is AI just copying? The Critical Computation Bureau (the group I am a part of together with Luciana Parisi, Oana Parvan and Ezekiel Dixon-Roman) has been working on the notion of recursive epistemologies. Recursive epistemologies are more than copying. It is not just data that is being put in, but also epistemological frameworks generated in the age of colonialism and racial capitalism. They are not simply replicated but also produce new hypotheses (the so-called hallucinations). On the other hand, do we the want powers of invention to remain a human monopoly? Do we want to maintain a strict division between Users as Masters endowed with all the attributes of humanity and AI to remain a dumb servant or slave? Or do we want to undermine and abolish all iterations of this formal model whether it works with humans, more-than-humans, not-quite-humans, and machines?

This means I also do not believe that AI is quite, as you say, colonizing humanity. I believe that there is a model of AI that is being developed that is making manifest something about the current model of language-based production. Its repetitive and highly codified nature which is the result of the implementation of procedures of quality control throughout the various forms of linguistic and affective labor. Automation is targeted where labor is more expensive – it is not targeting so much agricultural workers, care workers or miners (although of course it is also partially extending there). It is however showing us that what we do even as academics, students, writers, and “creative workers” is so codified that a machine can do it. I am thinking about grant applications for research funding. EU grant applications (all platform-based of course) are so formalized and codified that it has been one of the first uses that academics put AI to. Because software programmers are more reconciled with the procedural and technical nature of programming I suspect that they are the ones that have had less problems with it.

DZ: Your remark reminds me of a passage in The Human Condition by Hannah Arendt, that ‘The trouble with modern theories of behaviourism is not that they are wrong but that they could become true, that they actually are the best possible conceptualization of certain obvious trends in modern society’. This seems to suggest, among other things, that the relation between humans and machines is a two-way street: computers are as much modeled by and after us as we are subsequently modeled by and after them, creating a positive feedback loop – even when for Arendt it is our unique ‘humanity’ that is ultimately at stake, rather than a collective emancipation from the violent epistemological frameworks that enable and sustain the infrastructures of racial capitalism, as you seem to suggest.

Increasingly, there is also the prospect that the creative, linguistic, cognitive and affective labor on which LLMs are trained will themselves be replaced by AI – which, besides the demise of humanity by a supreme intelligence of our own making, is one of the main fears around AI. Here another Marxist dynamic seems to come into play, namely that between living and dead labor. As Marx wrote in the first volume of Capital: ‘By means of its conversion into an automaton, the instrument of labor confronts the laborer, during the labor-process, in the shape of capital, of dead labor, that dominates, and pumps dry, living labor-power’.

Apart from the highly developed cognitive labor invested by engineers and project managers, should the user-generated training data incorporated into generative AI be considered a sort of dead immaterial labor? And what are the political implications of this?

TT: I can see uses of AI, as the one by armies, where really the data you produced through geolocation for example really kills you – dead labor/fixed capital has always killed. I do think that our concepts of dead and living need rethinking though. Anthropologists such as Elizabeth Povinelli who has been working with Aboriginal Australians has really questioned this division lately. On the other hand, I also liked the way Luciana Parisi has drawn on Octavia Butler’s Afrofuturist Patternmaster series of SF novels. There are characters there who literally reproduce themselves by taking over people’s bodies – they are eternal. And there are other characters who counter them by practicing healing and by making new patterns that bring people together. Butler was drawing on African ontologies – and that is something that some scholars from that continent are already saying, that African ontologies are much more fitting contemporary technologies.  There is no simple alive/dead opposition there.

If you start with the experience of work in contemporary societies, you would possibly find that most people find it draining. It sucks your energies, there is very little replenishment or return. There is a good chance that AI will be put to work to drain even more of our energies. So how do you heal from the damage that capital does, how do you make new patterns for living together? I am thinking about Salvatore Iaconesi and Oriana Persico in Rome with the ‘neighborhood AI’ or with plant-inspired models of AI. Different imaginaries and practices are already there, but how do you make them systemic, how do you make the paradigm shift?

DZ: The crucial move by the post-workerists was to wrest away the concept of mass intellectuality or the general intellect from its association with the accumulated dead labor in industrial machinery and scientific knowledge, as Marx had originally done in Capital and the ‘Fragment on Machines’ in the Grundrisse, toward the living labor embodied by the multitude. As Virno writes: ‘Mass intellectuality is the composite group of postfordist living labour, not merely of some particularly qualified third sector: it is the depository of cognitive competences that cannot be objectified in machinery. Mass intellectuality is the prominent form in which the general intellect is manifest today’. To what extent do recent trends in AI vindicate Marx on this point, or at least reveal the existence of a historical dialectic, as the ‘digital competences’ Virno attributes to living labor are being integrated into the bleeding edge machinery of AI?

TT: Absolutely. This thesis is developed in Matteo Pasquinelli’s recent book The Eye of the Master. He calls it the labor theory of artificial intelligence. With generative AI, mass intellectuality directly feeds the neural networks. But AI is also showing how the production of wealth (for Marxists the source of labor) has changed. What does it mean, to quote something that Fred Moten said to me once, to make the model of social cooperation not the industrial division of labor in the factory, but that of a group of jazz musicians – or dancers? Data today are harvested also from activities that are done for fun, that imply a lot of mimetic behavior, lots of copying and sharing and sending on. I still return to Gabriel Tarde’s notion of psychologic economy with interest. He was really opposed to the idea that value production depended on the division of labor.

DZ: Finally, regarding the specter of job loss due to disruptive technologies like AI, it is often said that AI will augment rather than replace existing jobs (making work less intensive, more creative, etc.) and creating new kinds of jobs, thus initiating a new configuration of living and dead labor, labor and capital, including the ‘obscene’ forms of all too human labor that secretly sustain the techno-utopian ‘scene’ of AI. How do you see the future of work in general, and proposed solutions like universal basic income (UBI) in particular?

TT: It is very clear I think for everybody is that the tendency is to work more not less – I think a bit everywhere obviously with different intensities, degrees of contractual protection, and in general basic rights such as movement. We are being squeezed from all sides, some much more brutally than others, but it is happening everywhere. I know that the technoliberal class think that they can support UBI because that means that people will make do with that and the breadcrumbs of the platform gig economy. UBI is important, but there is a whole package going with that: the primacy of social reproduction: for example the so-called Commonfare, a completely renewed democratic system of institutions which provide basic needs such as health, education, insurance, environment etc. To make that the basis of a commons-based society, you cannot have that much wealth concentrated in the hands of the few. And there is the specter of a planetary apocalypse as climate change intensifies. It is a very dramatic moment that we are living through, unprecedented.