Is the Generator Class an Emerging Class Formulation?
I wrote this for my Masters degree so its put together with that in mind rather than anything else but sharing it here to a: keep a record of it and b: because the topics it covers seem particularly relevant at the moment. The media whirlwind around AI is in full hysterical flow but, from what I’ve seen, a lot of it is either divorced from reality or just completely disinterested in some of the most basic, practical questions behind it. Anyway, this isn’t a grand contribution to the debate, obviously only an amateur picking at the edges, but still – might be of interest to a tiny handful of people.
Is the Generator Class an Emerging Class Formulation?
Introduction
My focus for this essay is to explore the potential for a new class formulation amidst the rapid development of ‘AI’ and algorithmically generated art and content (broadly termed as Generative AI (GAI)). As new forms of production emerge from the process of data collection and access management we’ve already seen extensive shifts in both the public and theoretical perceptions of labour and the potential redefinition of the ‘commodity’. Within this rapid, technology driven shift represented by the Information Economy the search for new structures of analysis and understanding have emerged from both Capitalists and those seeking to challenge it as a dominant economic model. Concepts such as the Precariat, Hacker Class, Vectoralists and Creative Class all seek to consolidate potential group interests and, at times, directly re-constitute traditional Class definitions into a form more readily understood through the prism of a more information, data and generation based economy.
In exploring this topic my initial goal is to define a potential new Class emerging from the most recent notable wave of algorithmic access to new means of production, something I’ll be referring to as the ‘Generator Class’. A group I believe to be distinguished from (and sometimes within) other potential class formulations by their attempted claim to access and privilege within the new generative means of production such as prompt based, algorithmic content generation.
Following on from that I want to explore the structural validity of this potentially emergent Class. As the wide territory involved in new, generative technologies continues to rapidly expand and touch upon issues of intellectual property, access rights and the exploitation of precarious labour there are, I believe, definite questions to be asked about how enduring any attempt at evolving a new class position can be within such a landscape. However I think that the attempt to constitute such a thing is, in itself, indicative of broader potentialities and reactions to shifting economic focuses and realities.
Finally I want to explore the current and potential reactions to this process. Whether emergent economic groupings prove enduring or not the reaction they incite, both active and passive, feeds back into the development of new cultural and economic realities.
Notes on Terminology
At the outset of planning this essay one of the immediate questions was how to approach the issue of terminology. Obviously any discussion on Class is, at some point, drawn back to Marx and his definitions of Class divisions but in exploring evolving labour relations and Class constructions I chose to hold that as a semi-distant foundation and instead adopt the approach advocated by Wark (and others (Negri & Hardt’s Multitude, Richard Florida’s Creative Class etc.)) of seeking a new language of analysis and structural understanding.
“To come into an awareness of class it to speak another language. It is to refuse the terms that are given and seek other terms, other concepts.”1
Wark goes considerably further in her openness to exploration of new Class formulations than I would consider attempting. Even suggesting that Capitalism has come to an end point and what now emerges is a new system in itself2. Even though the challenge she presents is a major one to anyone writing critically about almost any political understanding there’s also plenty contained within her work which offers ample resources for useful experimentation and speculation.
“I wanted to speak
the beautiful language
of my century” – Guy Debord3
Citing Debord’s concept of détournement Wark calls for a more experimental, challenging exploration of language when approaching issues such as Class4 and it’s in that spirit that I both opted to approach the topic of this essay (the Generator Class) and also to frame it with a mixed set of references. So with that in mind I’ve decided to use the tools that seem like the best fit for the job at hand, eclectic as they are. From Wark I’ve taken the concept of the Vectoral Class as a framework for an emergent/reconfigured Ruling Class. As a group robustly defined in her work to act as a presence in this new Information Economy I believe it offers more utility within the context of this work even as it (functionally) intersects and often blends with earlier constitutions of the Capitalist hierarchy. Where I differed from Wark however was in opting not to use her idea of the Hacker Class. While her work in defining it has been a definite inspiration to me I found the boundaries of it to be too vast to be practically useful, especially when approaching ideas of creation in regard to automated generation. When talking about immediate issues of division and relations to labour there are so many gradients contained within her formulation that to apply it with definite purpose seems less useful than its value as a broad strokes conceptual playground – an inspiration for détournement.
Instead I chose to balance the Vectoral Class against the Precariat5, a concept that, in its own way, is no less all consuming but which does speak to more definite material positions. Especially in relation to the Generator Class and the Commons. The former (Precariat) I identified as a form of Worker that reflects certain aspects of recognised labour approached by the Generator Class, the latter (Commons) I frame as a more wide reaching and not (necessarily) commodified pool of culture, information and value which takes special prominence with regards to ideas around data harvesting and enclosure.
The Commons is a useful term in itself but ultimately somewhat lacking I think given that when approached by the Vectoral and Generator Classes it takes on a new data focused exchange value beyond the use and social value it’s commonly defined as holding. They make it primarily an object of information enclosure and extraction, as distinct from other, more egalitarian or historically exploitative, experiences of it. Short of creating an additional term to reflect the newer, immaterial extractive approach to it as a new term though (which would be another essay in itself) I felt sticking with it while nodding to a mediation in its meaning was the best option.
I also gave consideration to Marx’s term General Intellect6 – “where collective cooperation and knowledge become a source of value”7 but I felt that it somewhat limited the scope of what’s truly being discussed within this essay with regard to data extracted on a vast scale.
As one last nod to my conceptual and linguistic choices in framing this essay I should mention Platform Capitalism as defined by Nick Srnicek. His structural breakdown of terms such as Cloud Platforms8 approaches similar concepts to both Wark and my own efforts. I think a useful addition to this essay could certainly be had in integrating his work which – in part – is a useful compliment to Wark’s Vectoral Class and the Generator Class but ultimately I felt it would be too expansive a territory to attempt to add. Although the relation between between ideas such as Platform Capitalism, Smart Urbanism9, Platform Urbanism10, Warks Vectoral and my speculative Generator Class would (perhaps) all be interesting relations to delve into elsewhere.
The Generator Class
In framing the concept of the Generator Class there are immediate challenges present. The scope of emerging technologies around AI and algorithmic generation is, potentially, vast and to speak at any given moment is to risk missing a major shift in the next. There are also inherent issues in exploring the new means of production that this group primarily relates to, everything from art, to academia, to programming can potentially be included within their remit but to explore all of them would be a larger work than the reach of this essay can cover. So instead I will, largely, focus on one of the most active and contentious areas of technological expansion that could be said to fall under the remit of the Generator Class – that of art and cultural output. It’s with that example in mind that I’ll primarily be working.
To offer a basic grounding before exploring the wider definition – the Generator Class, as I’m seeking to define it, is constituted of a new cultural, economic and potentially political group which assumes a role somewhat analogous to the small scale manufacturer of previous eras. Using generative tools (variously referred to as GAI or algorithmic generators) such as Chat GPT, DALL-E, Midjourney and many others they represent a low level producer reliant on far larger system of production. To continue with the analogy, while traditional manufacturers may have operated a single factory or mill reliant on major mining or agricultural concerns, the Generator Class operates a largely automated production line reliant on the collection and control of information by what McKenzie Wark defined as the Vectoralist Class.
To take DALL-E, a text-to-image generator, as an example the project is a capped-profit11 enterprise (under the OpenAI organisation) which has, at various times, received funding from Elon Musk, Amazon and most recently Microsoft ($10b~). Heavily at odds with many practising, non-generative artists there are considerable questions about the source of content (primarily images) used to ‘train’ DALL-E, while its output is relatively unaffected by questions of initial labour input or potential ownership. It’s around projects like this that the Generator Class positions itself, remote from the presence of primary labour individuals use permitted access from major Vectoralists/corporations to generate new content primarily via textual prompts. Output which is in some cases already and certainly potentially commodified and monetised as part of a wider digital (and indeed physical) economy.
This relation to labour places the Generator Class at an odd and contradictory intersection of previously conceived economic groupings. Certainly they stand separate from the Vectoral Class as while they gain access to their information resources they have no more power to control it than do the original generators of such data. The commodities they extract from AI and algorithmic generation are also ultimately subject to the same measures of appropriation and enclosure as those they themselves benefit from. Their capacity to assert copyright or even defend a nebulous concept of a cultural/social Commons as applicable to the vast amount of data gathered by major tech platforms is minimal. And in that respect they find themselves positioned closer to the Precariat, if not partially absorbed by it. Finally they find themselves in proximity to the Hacker Class as defined by Wark, generators of ’new’ content within the information economy/age but it’s in this that they disavow their position as exploitable resource of this potential new economy and instead place themselves as aspirant entrepreneurs of it.
The Generator Class seeks distinction, in some cases passively and in others actively. Later in this essay I’ll provide some case studies to express this but initially I’d like to explore the key quality that enables this aspirant Class separation from invisible, precarious and even non-participatory workers. Creativity.
Creativity within the new technological landscape of generated content becomes an immediately contentious concept. At this point in time we’re seeing a constant argument between those who seek to recognise new ‘AI’ generated content as original and unique work – an act of creative labour in itself and those who dismiss it as a simple mechanical reconstitution of what already exists, a clever act of imitation mediated by impressive but ultimately unoriginal technological gimmicks. Whilst an assertion of the former is structurally essential to the Generator Class to validate its own position as producers there are clearly defined questions that perhaps undermine it.
There are undoubtedly profound questions to be asked about the potential of AI Art as a source of original and innovate forms of creative work but to focus excessively on them would be to write an essay in itself. Indeed the question of what constitutes Artificial Intelligence in the first place is a wider remit than I’m attempting to claim in this work. Instead I, like the Generator Class, am looking for the functional realities at work in our present moment and seeking the lines of labour and connections and divisions they create. As rich a ground as grander speculation may be it isn’t inherently relevant here.
In this moment then the question is one of what constitutes the new commodities of the Generator Class and what resources those commodities are drawn from and reliant on.
“Much of what passes off for AI today is really a product of coupling big data with statistical analysis. Impressive or even mysterious as some of the outcomes may look, they are the result of advanced calculations performed on large quantities of data.12”
My first example for exploring the extractive as opposed to creative nature of generative art and content is, perhaps counter-intuitively, Amazon’s Mechanical Turk. MTurk is a platform for hiring casualised labour to fulfil any given task, although it heavily promotes its capacity as a tool to establish Machine Learning systems. Defining individual acts of human labour as a Human Intelligence Task (HIT) the Amazon project provides a steady supply of workers to tag, provide and sort the data that is ultimately fed into GAI/algorithmic content generators.
“MTurk can be a great way to minimize the costs and time required for each stage of ML development. It is easy to collect and annotate the massive amounts of data required for training machine learning (ML) models with MTurk. Building an efficient machine learning model also requires continuous iterations and corrections. Another usage of MTurk for ML development is human-in-the-loop (HITL), where human feedback is used to help validate and retrain your model. An example is drawing bounding boxes to build high-quality datasets for computer vision models, where the task might be too ambiguous for a purely mechanical solution and too vast for even a large team of human experts.”13
While this service sits in something of a grey area between human and AI ‘labour’ it does reflect a wider structural system of resource extraction. While technological processes may be the end point of the Generator Classes’ commodity production there lays beneath and beyond that a vast network of invisible human labour, producing the data sets that are integral to the creation of seemingly new content. In contrast to my earlier comparison to the small scale manufacturer this process serves to create an even wider distance between the productive and labouring classes – in this case there is no factory or mill, nor is there even a designer to dictate new form – instead that middle ground between resource and commodity production is filled by algorithms whose capacity for original creation is limited to the labour of the Precariat, or Commons, upon whose actively generative work they feed. Indeed while MTurk may act as a useful, if somewhat detached, example of the means of production involved here it largely serves to reflect a broader model. Whilst Amazon has formalised the process of data extraction as a service the same principle of core, human labour is shared by most resource gathering exercises carried out by nascent AI and generative projects. Social Media platforms, large scale retailers, image and text generators – they all function on the input of a mass of people who stand distinct from the Generator Class insofar as they neither profit from nor (in most cases) even agree to participate in this extractive process.
It should be said that while I seek to define the Generator Class in relation to their extractivist nature that doesn’t dismiss their potential to act as a creative element. Nor to define them as an inherently parasitic or exploitative force. Approaches to AI and Art both offer potentialities for the positive existence of a new form of interaction with mass data holdings and their generative potential but as mentioned earlier, the focus of this essay isn’t on potential but instead on existing and emergent relationships. From that perspective, I believe, the Generator Class must be viewed primarily as extractive (and therefor notable in relation to other class groups) even where it might place a reasonable argument at intending otherwise.
“The artist no longer creates work; he creates creation” Nicolas Schöffer
A key argument in the claims to a more utopian – or at least culturally valorised – form of the Generator Class comes from those parts of the Art world which have already sought to actively engage with new technologies as a participant, rather than a resource and who in doing so have tied their own work to the Generator Class as a whole, albeit perhaps unwillingly/unwittingly.
In his project Of the Subcontract Nick Thurstow strikes up just such a position of self-conscious artistic valorisation of the Generator Class’s extractive relation to labour. In a self declared exploration of ‘expropriation14’ he paid MTurk workers to create poetry for a compilation. Again, with MTurk as the tool the labour consists of HITs carried out by flesh and blood workers (albeit ones partially designated to a job by algorithmic processes) but as anonymous contributors they offer a fair representation of the process of production from data itself. MTurk Workers, named as such by Amazon themselves, take jobs through an anonymised process and are graded for their efficiency by the system itself. While human they attempt to fill the gaps that, it seems assumed, AI will ultimately be able to occupy itself.
In the afterword to Of the Subcontract Darren Werschler suggests the project as a critique of a form of production already existent in the Art world15 in the works of such creators as Damian Hirst who regularly farm out the labour of creation to subordinate workers. An observation that hints at a proto-Generator Class operating on a considerably smaller scale and (generally) at a more immediate reach to the workers involved. Such ‘High’ Art pursuits however, while they hold the novelty of Post-Modernist observation, still retain the class distinctions that help frame the Generator Class. Thurstow, whilst seeking to explore patterns of expropriation (and exploitation) does so from the privileged position of having both the financial ability to mobilise the use of anonymised, precarious workers and the platform access to disseminate the commodified products of their labour. Certainly relative to other forms of expropriation his case represents a largely benign act but it also usefully mirrors a broader pattern that comes with considerably less artistic consideration and considerably more invisibility and exploitation.
Heading further towards that position is the ‘art’ collective Obvious whose work Portrait of Edmond Belamy (2018) was the first GAI piece to be sold by a major auction house (Christie’s). Fetching a healthy $432,500 the piece was the product of a GAI trained on a dataset of 15,000 portraits from between the 14th and 20th centuries. Given the requirements of the data involved there was no direct enclosure of the productive work of living artists but the process of exploitation was to a large degree the same. Acting as participants in the Generator Class those behind the project, Hugo Caselles-Dupré, Pierre Fautrel and Gauthier Vernier, still maintain an authorial influence over the work – which is a discussion that falls into a separate discussion on the merits of AI Art as a concept – but in their comments they adhere to my expectations of the Generator Class by ignoring the original source of the labour fed into the machine.
“If the artist is the one that creates the image, then that would be the machine,’ says Caselles-Dupré. ‘If the artist is the one that holds the vision and wants to share the message, then that would be us.” – Hugo Caselles-Dupré
A sentiment shared by others in the field of AI Art:
“There is a human in the loop, asking questions, and the machine is giving answers. That whole thing is the art, not just the picture that comes out at the end. You could say that at this point it is a collaboration between two artists — one human, one a machine. And that leads me to think about the future in which AI will become a new medium for art.”16
Returning to OpenAI we find a more ubiquitously and less conceptually framed reference for exploitation. It’s openly stated that OpenAI draws data from social media1718 while their dataset for DALL-E remains something of a mystery as they refuse to release details on what it contains. Although the existence of tools like Have I Been Trained19, which allows artists to search for instances of their own work being used to train AI datasets, suggests that data getting pulled may well not be done with the owners consent. Certainly a cursory search for a few key words will bring up a gallery full of quite clearly copyrighted content that’s been pulled. Something which other Generative AIs (GAIs) have openly admitted to, especially when the Shutterstock watermark was to be found clearly on generated images!
Ownership
The issue of ownership in relation to GAIs is, obviously, a major factor and a complex one given the contradictory and sometimes combative nature of both the Generator Class and other new formulations within the emergent tech/AI economy. Intellectual Property as a concrete progression of more traditional forms of private property rights is a necessary factor for the existence of a Vectoral Class in Wark’s terms. Information presents a novel new form of resource as it is, with some extreme limitations, beyond scarcity – without mediation it, quite easily, falls into the potential realm of the Commons. Endlessly replicable in a digital landscape there is no finite nature here. While at certain levels – that of brand, code or drug developement – ownership may trace a clearer line from innovation and investment to product with many of the tools now opening up to the Generator Class there’s a strong interest in selective ignorance of potential ownership as a clear act of emergent Class interests.
For the Vectoral Class the endless need for new datasets and content to feed AI, GAI and instigate more mainstream models of consumption through customer data harvesting makes the concept of ‘ownership’ a relative one. Microsoft will undoubtedly enforce IP and copyright over its brand and its products but in funding Open AI and its various projects they show a measure of indifference to the rights of others. Indeed when it comes to our Cultural Commons as collective dataset they show themselves to be quite willing to enclose content from the digital landscape not just in mobilising harvested data to serve AI but also in somewhat obliquely restricting rights over the final GAI generation. DALL-E for example makes a contractual claim to ownership of generated images but doesn’t assert copyright20, although it’s questionable whether they practically could anyway.
Legal questions aside however the functional truth may be closer to traditional relations within IP and copyright laws – ultimately rights come secondary to the ability to enforce them. Workers generating content, even those in the Generator Class, are unlikely to have any viable rights against major corporations – at least not individually.
Where Vectoral actors have their own interests in taking a slightly contradictory approach to IP and copyright rules those in the Generator Class follow their lead. In order to commodify their output whether through direct marketisation (sale of NFTs, licensing rights, production of physical products etc) questions of origin and primary labour pose challenges that are best not faced. A happy and intended collective ignorance which serves all sides – for now – except of course for the vast majority of workers who find their work appropriated without consent, payment or acknowledgement.
Is the Generator Class a Viable Class Constitution?
That the Generator Class exists is, I think, a fair assertion to make. The (im)material means of production that define it are there, the exploitations of those means are evident but as ever with this new and constantly evolving sphere of new technology that doesn’t necessarily lend any permanency to them as a social-cultural-economic group. Just as new technologies throw up the potential for their creation as a Class it also sows the seeds of challenges to it. Data resource and commodity ownership, as mentioned above, remain vaguely defined and often contested. The Precariat/Commons which is extracted from is in an already combative relationship with the Generator Class – not just over ownership but also over more philosophical concepts such as the nature of art and creativity. Even the Vectoral Class which by its ownership of platforms, (questionably) data and the actual infrastructure (computational and storage) stands as a key beneficiary of a new intermediate, extractive class has no necessary commitment to it. Some platforms have already banned works from GAI sources21 and it’s easily conceivable that should GAIs and the extractions of the Generator Class stray away from asset stripping a largely faceless Precariat/Commons and begin infringing on the copyrights and IPs of major corporations then the relationship may easily sour. A common joke from artists opposed to GAI is that the quickest way to kill it is to get Micky Mouse featured in the database – Disney will do the rest.
Another factor that could potentially undermine the position the Generator Class seeks to claim is within the tools themselves. AI datasets already exist that have no human factor in their production of content, either Generator Class or Precariat.
“The invisible world of images isn’t simply an alternative taxonomy of visuality. It is an active, cunning, exercise of power, one ideally suited to molecular police and market operations.“22
In his project A Study of Invisible Images Trevor Paglen focuses on ‘Machine Vision’, by which machines generate images for the reference of other machines – especially notable in facial recognition technologies, self driving cars, surveillance and policing23. How far this can and will be applied to commodity production remains to be seen but it does have the potential to pose wider threats to various economic groups and potentially even our perceived system of Capital itself – a notion Wark might perhaps say is further proof of our progression beyond it (and feel validated in suggesting that something worse might be coming). This however is still just a potential threat and one positioned towards a still emergent class.
Through GAI the Generator Class has already shown itself to be incredibly mobile in its claiming of economic space. Each failure (such as NFTs) is met with an emergent possibility for new interactions with labour (GAI art, illustrations, writing, animation, digital assets, film etc). And while the issue of ownership may provide a more daunting challenge should it ever be truly pressed by the Vectoral Class it does still leave both Public Domain and Creative Commons content as viable resources for data extraction.
In moving to occupy new spaces the Generator Class shows its rapid sense of evolution too. As mentioned earlier it already has certain practitioners within the world of High Art. One where authorial identity as a factor of authenticity is unlikely to be surpassed by either issues of primary labour, which have been debatable since Duchamp, or ownership – where concessions to a Precariat or anonymised and rightless labourer are long understood as viable (see Damian Hirst and others who have work fabricated).
The Generator Class is also coming to terms with repeated waves of new tools that open up new potentialities. As a recent example Clarkesworld, a prominent Sci-Fi publisher, has just closed submissions as a glut of GAI stories were being submitted24. Editor-in-Chief at Clarkesworld – Neil Clarke – cited Hustle culture partnered with access to generative tools (most notably ChatGPT) as a key trigger to this. A suggested connection that offers further intersection points for the Generator Class as variably Precariat-Entrepreneur25-Emergent class. It’s a progression in GAI that goes even further in blurring the speculative lines of ownership, Commons and labour. While visual artists have at least some vague tools for detecting whether their work has been pulled to train databases or may be able to recognise similarities to their own works writers have little recourse in either respect. A short story generated to emulate the style of Isaac Asimov may come out as predictably derivative, but without a mass shift towards openness by the Vectoral Class as owners of the tools generating it there’s virtually no clear way to know if it actually is trained on his work (or the endless imitations and interpretations of it already created by a multitude of other writers).
It’s a problem only primed to worsen too. Microsoft and Google have both ploughed billions into ‘chatbot’ development. In Academia for example the potential for GAI essay writing is an already recognised threat with myriad companies offering essay generation services26 and institutions struggling to find counter-measures against such submissions. Again, we have here the Generator Class inserting itself as a mediating force in the process – one which it’s, rather optimistically, suggested could ultimately take a role as moderator of its own permitted output but which practically seems more inclined towards Hustle Culture grifting than concessions to social needs.
“In this case, that would mean companies establishing a shared framework for the responsible development, deployment or release of language models to mitigate their harmful effects, especially in the hands of adversarial users.” – Rob Reich, Stanford Professor of Political Science27
There are myriad other use cases of AI being used to create new services, content and even whole commercial sectors – certainly enough to write more than a couple more essays about – but my main purpose in citing the examples I have thus far is to show the mobility of this Generator Class. Unestablished as it may broadly be it exists in a landscape offering ample opportunities to find and reinforce its own foundations. Does that make it viable in the longer term? That much I think remains debatable. Ultimately submissive to the Vectoral Class the Generators are reliant on their indifference, view of potential profitability and inability to achieve the same eye for the opportunity as a wider class formation does. Stepping back a few years you have perhaps the most prescient suggestion of how that situation may progress in the form of Content Creators. YouTube and social media in general have made massive use of individuals willing to accept access to the tools of communication/production as fair substitute for wages for years now. YouTube as a particular case made $28.8b in revenue in 2021 with the overwhelming majority of their top revenue generating channels being (relatively) independent Content Generators28. While I would differentiate these creators from my speculative Generator Class on the grounds of their use of technology I’d say there are certainly grounds for comparison and even a direct, Proto-Generator Class definition. After all, while their content is ‘original’ those acting as independent wealth generators for YouTube as Vectoral Class actor are heavily reliant on the algorithmic and user data assets of the company – relying on that analytic data service to promote themselves they may represent the more organic arm of the Generator Class.
What the example most usefully shows however is that the Vectoral Class ultimately has a relatively transient interest in actual content creation or generation. When Google or Microsoft set aside billions of dollars in order to train new AI generators with new datasets their interest isn’t in mobilising it to their own productive ends, control of the tools – ownership of the means of production – is an end in itself. The Commons and the Precariat will provide the labour for that model, the Generator Class will extract the value.
The Future of the Generator Class
In writing about the Generator Class I’ve attempted to define an emergent class entity, one bringing its own interests and its own relations to the means of production, labour and a similarly emergent (re)formulation of a/the Ruling Class. Like Wark I believe that experimentation with and exploration of class forms against an ever shifting technological, social and economic landscape has value in itself. While I’d never suggest that my own speculations are as comprehensive or insightful as hers I do think the act of at least trying to think about the structures generated by emergent technologies is an important process. Just as important though is seeking out the reactions to these potential re-constitutions and creations surrounding new means of production, exploitation and enclosure. Where technology, most notably AI in this case, creates vast new frontiers it also creates new conflicts.
As with so much of this essay though I can largely only speculate about potential paths of travel.
Emergent resistance to a commodity focused model of the Generator Class exists both within and without the class formulation itself. Art, as generally my core example, has already seen a large number of its practitioners set themselves against GAI tools. Art sharing platforms provide a fertile battleground for that as each one takes its own position on generated content. Getty29 has banned GAI ‘Art’ completely while ArtStation – a leading portfolio site – hasn’t just refused to do that but has even cracked down on attempts by artists to protest the decision30. Clarkesworld, as mentioned, have reacted by closing submissions in the face of GAI writing but, while they currently seem confident of their ability to spot such efforts, their task will only grow harder as the technology progresses. In all cases the overarching interests of the Vectoral Class remains the defining factor. Individual artists and creators can protest, smaller platforms can formulate policy – just as universities and the like can – but as long as the behemoths of the tech world continue to funnel money, resources and interest towards the creation and exploitation of new datasets then the conflict is being fought amongst those effected, more than those creating the conditions.
Those are just surface level expressions of the issue too. The vast majority of what the Generator Class extracts and what their foundational tools seek to enclose isn’t anywhere as easily outlined. I believe it’s in the realm of the Commons that the real questions of control and exploitation lay. While the vast majority of people remain so vulnerable to having their labour – or indeed their very existence – commodified to feed an information hungry system the question of how that system manifests is almost a secondary one. The Generator Class is not the Ruling one, those submitting stories to Clarkesworld, utilising the vast data stores of YouTube and Google or monetising their extractions from DALL-E aren’t irrelevant by any means but conflict against them by the majority of primary labourers is too narrow an engagement to reflect the grand scale of what’s going on.
“The barbed wire remains invisible” – Evgeny Morozov31
In writing about privacy within his article The Real Privacy Problem Evgeny Morozov talks about the balance of personal data privacy as both an individual and political issue. He speculates about individual actions of rejection in the face of data harvesting. His invisible barbed wire is the personal enclosure created by our algorithmically mediated social, economic and political lives.
Perhaps it’s necessary to take that observation and drive it further – not to see invisible barbed wire that encloses us as consumers but instead an invisible factory that exploits us as workers; as social, economic and cultural producers. In relation to democracy Morozov also suggests a balance point between the right to privacy and the transparency required for democracy – another logic worth expanding as we look to find our own balance between the Commons and the self-ownership of our own social and intimately personal output.
Still the issue of class conflict surrounding the Generator and Vectoral Class ends with two vast questions left to answer. The first being the immediate and practical – how are the exploitations of these new classes to be met? Confronted with enclosure and exploitation what immediate ground do we want to hold? Are we the artists protesting against the misuse of our creative and lived labour? Or are we Artstation, eager to pursue and exploit these new generative means of production?
The second is the broader and more abstracted question of ownership and the Commons. Lawrence Lessig, writing in his book Free Culture prior to this latest boom in AI and GAI said:
“Think about the amazing things your kid could do or make with digital technology — the film, the music, the Web page, the blog. Or think about the amazing things your community could facilitate with digital technology — a wiki, a barn raising, activism to change some- thing. Think about all those creative things, and then imagine cold molasses poured onto the machines. This is what any regime that requires permission produces.“32
At the time it was a semi-utopian argument, a defence against the enclosures of an earlier Vectoral Class which sought to limit the flow of data and lock creativity into heavily mediated IP and copyright laws. Standing here, near enough 20 years later, it seems that his Utopia has been almost entirely outflanked by new technology. The big brands, the big owners, the Big Vectoralists have grown no less militant in their pursuit of their own rights. Meanwhile the liberated Creative Commons Lessig aspired to has come to fruition in at least some warped sense – our creativity, our productivity and to a degree even our selves are no longer defined by permissions. The information economy has liberated it all, but only to the service of some.
Commons for thee, but not for me.
Bibliography
Lessig, L. Free Culture Penguin Press 2004
Lindholm, C. Culture and Authenticity, Blackwell Publishing, 2007
Loving, G. Sad by Design: On Platform Nihilism Pluto Press 2019
Srnicek, N. Platform Capitalism Polity Press 2016
Wark, M. A Hacker Manifesto Harvard University Press 2004
Wark, M. Capital Is Dead: Is This Something Worse? Verso 2021
Zylinska, J AI Art: Machine Visions and Warped Dreams Open Humanities Press 2020
Endnotes1. Wark. M Capital Is Dead. Is this Something Worse? Verso 2019 p50
2. Wark. M Capital Is Dead. Is this Something Worse? Verso 2019 p26
3. Wark. M citing Debord, G in Capital Is Dead. Is this Something Worse? Verso 2019 p21
4. Wark. M Capital Is Dead. Is this Something Worse? Verso 2019 p33
5. Standing, G The Precariat Contexts Volume 13: Issue 4 2014
6. Marx, K. Fragment on Machines (from The Grundrisse) p706
7. Srnicek, N citing Vercellone, C. Platform Capitalism Polity Press 2017 p27
8. Srnicek, N Platform Capitalism Polity Press 2017 p37
9. Sadowski, J. Cyberspace and cityscapes: on the emergence of platform urbanism Urban Geography Volume 41 2020 p449
10. Sadowski, J. Cyberspace and cityscapes: on the emergence of platform urbanism Urban Geography Volume 41 2020 p450-451
11. Coldewey, D. Open AI shifts from nonprofit to ‘capped-profit’ to attract capital TechCrunch.com 2019 https://techcrunch.com/2019/03/11/ope...
12. J. Zylinska AI Art: Machine Visions and Warped Dreams Open Humanities Press CIC 2020 p25
13. Amazon Mechanical Turk Use Case – mturk.com
14. S. Voyce Of the Subcontract: An Interview with Nick Thurston Iowa Review 2014
15. D. Wershler cited J. Zylinska AI Art: Machine Visions and Warped Dreams Open Humanities Press CIC 2020 p123
16. Is artificial intelligence set to become art’s next medium? Christies.com https://www.christies.com/features/A-...
17. D. Cooper Is DALL-E’s art borrowed or stolen? Endgadget,com 2022 https://www.engadget.com/dall-e-gener...
18. WebText – https://paperswithcode.com/dataset/we...
19. https://haveibeentrained.com/ – During research for this essay I even found instances of my own artwork being included in harvested datasets! https://haveibeentrained.com/
20. A. Guadamuz DALL-E goes commercial, but what about copyright? technollama.co.uk 2022 https://www.technollama.co.uk/dall%C2...
21. Quach, K. The Register 2022 https://www.theregister.com/2022/09/2...
22. Paglen, T. A Study of Invisible Things Brooklyn Rail 2017 https://brooklynrail.org/2017/10/arts...
23. J. Zylinska AI Art: Machine Visions and Warped Dreams Open Humanities Press CIC 2020 p88
24. Acovino, V. Sci-fi magazine ‘Clarkesworld’ stops submissions after rush of AI-Generated stories NPR 2023 https://www.npr.org/2023/02/24/115928...
25. In the most generous conception of the term.
26. One of which I experimented with to write a version of this essay, the results were – happily – terrible.
27. Reich, R. Now AI can write students’ essays for them, will everyone become a cheat? The Guardian 2022 https://www.theguardian.com/commentis...
28. Iqbal, M. YouTube Revenue and Usage Statistics 2023 BusinessofApps.com 2023 https://www.businessofapps.com/data/y...
29. Vincent, J. Getty Images bans AI-generated content over fears of legal challenges TheVerge.com 2022 https://www.theverge.com/2022/9/21/23...
30. Weatherbed, J. Artstation is hiding images protesting AI art on the platform TheVerge.com 2022 https://www.theverge.com/2022/12/23/2...
31. Morozov, E. The Real Privacy Problem MIT Technology Review 2013 https://www.technologyreview.com/2013...
32. Lessig, L. Free Culture The Penguin Press 2004 p305
