The technological surge triggered by the AI industry will revolutionize the production of ideology in the core societies of the world system.
July 7, 2024
He “lost everything” that made up his job, complained a 3D artist on Reddit around a year ago after AI software found its way into his company. The graphic artist, whose experiences were published on the Swiss GNU/Linux blog, worked at a small company with ten employees that produces mobile games.[1] When he started using the AI image synthesis service Midjourney V5, he stopped considering himself a creative artist, as he was now only concerned with reworking the models generated by the AI.
Link: https://exitinenglish.com/2024/07/07/ai-and-the-culture-industry/
The company had no choice, as models and characters for the mobile games can now be created in two to three days, whereas this work used to take several weeks. In his job, he wanted to “make, model, and create in 3D space. With my own creativity. With my own hands,” the graphic designer lamented. But now all he has to do is rework models that are “the result of internet content scavenged together” “by artists who weren’t asked.”
This is one of the foundations of the AI boom of recent years: the industry scans the entire internet, collecting gigantic amounts of data in a legal gray area in order to train its models using this data. Billions of images, texts, videos and music form the material on which the neural networks have to be painstakingly trained. The AI industry’s increasingly complex programs not only consume vast amounts of computing capacity and therefore energy (even the AI speech recognition program Whisper can only be used locally with GPU support using CUDA), they also need people to train them in their take-off phase.[2]
Back to the 18th Century With AI
Pattern recognition in the learning phase – whether language, images, music or text – is still done by “manual labor,” by cheap labor in the global South. The absurdity of the AI industry’s constitutive phase is that it destroys the few “creative” jobs created by the late capitalist culture industry, while temporarily creating an army of day laborers who first have to teach the machines to “learn.”[3] The large data sets must be coded, in mindless labor, by humans with “labels” (similar to the captchas that are often requested when logging in) in order to feed the AI systems with meaningful material.
And this manual labor for the AI industry of the 21st century is carried out under conditions that were common in the 18th century during the blood and dirt-soaked birth of the capitalist world system. The global industry of data collection and analysis, which makes the empirical material usable for AI neural networks, pays the lowest wages and is notorious for the most precarious working conditions. The Australian market leader Appen, which processes material for Amazon, Facebook, Google and Microsoft, can draw on a host of around one million day laborers in the Philippines, South America and Africa, who – when things are going well – are fobbed off with monthly wages of less than $300. The industry, whose turnover is expected to rise from $2.2 billion in 2022 to $17 billion in 2023, can relocate even faster than the textile industry, which also relies on poverty wages, as no factories or production facilities need to be built. These day laborers are often exploited in home-based work – as in the publishing system of early capitalism.
The Perfect Tool for The Culture Industry
Humans have to tell the machine which patterns carry which label so that its pattern recognition can work better and better. Building on a gigantic mountain of data that has been provided with corresponding “labels” by day laborers, the AI systems generate their images and models by matching the user’s request with the labeled material and offering its variations as output. This is the whole secret of the ridiculous “AI art” that is currently turning the concept of art into a hollow phrase. Nothing new can emerge from this; it is not a creative, aesthetic act that is based on any kind of idea that would have emerged from an examination of facets of human existence – which, in the broadest sense, is what art does.
But what the AI content systems can do better and better are variations of what already exists. The data material with the corresponding labels can be spit out in ever new combinations: new characters, new monsters, new images, new storylines that only modify what they have been fed without transforming into a different quality. And this is precisely what makes AI so valuable for the late capitalist culture industry.
In a guest article for the New York Times, left-wing linguist Noam Chomsky described the fundamental limitations of current machine learning systems such as ChatGPT, which can scan “huge amounts of data” in response to queries to generate ever better “statistically probable outputs,” creating the impression of “humanlike language and thought.”[4] However, there is a fundamental difference between the human mind and “a lumbering statistical engine for pattern matching, gorging on hundreds of terabytes of data” to spit out the “most probable answer” in a conversation or scientific query by making “brute correlations among data points.”
The current generation of AI systems is not able to draw conclusions based on “causal mechanisms or physical laws” in the way that human reasoning is able to, a “surprisingly efficient and even elegant system” that is able to “create explanations” with “small amounts of information.” ChatGPT and company, as highly sophisticated statistical pattern recognition machines, on the other hand, are not able to fundamentally distinguish “the possible from the impossible.” Even correct scientific answers and predictions come close to “pseudoscience,” as they are not based on scientific explanations but on statistical probabilities. According to Chomsky, AI systems are therefore incapable of drawing real conclusions or exercising “creative criticism” and are stuck in a “pre-human” phase of cognitive development.
However, all of the linguist’s objections have no relevance for the production of commodities in the culture industry. The basic principle of the culture industry is the thousandfold variation and reflection of the surface of reality. These are variations of the existing, which confirm the existing through their permanent repetition. Everything has to change on the surface so that basically everything can remain as it is. Whether science fiction or fantasy, AAA computer games or high-end Hollywood productions, consumers of these cultural products are in fact only living through the costumed society in which they were produced – and in which they themselves live. The culture industry is like a content machine that revolves around itself, constantly spitting out a mantra in its subtext that reliably kills all thoughts of alternatives: it is what it is. New aesthetic material is constantly needed for this dull reflection of the surface of reality in ever new variations.
Gaming and AI
Enter the AI industry. AI systems are virtually predestined to generate new forms and new material for the culture industry. Models, characters, images or scripts can be delivered at the touch of a button, in a fraction of the time previously required. The great competitive advantage of AI lies precisely in the fact that it lacks all the creative, reflective and critical abilities that are inherent in human content providers. The system varies the data accumulated in terabytes and provided with corresponding labels to spit out “new” content for films, books, comics and gaming. For the first time, AI content systems will enable the culture industry to create pure products that are free of any subtext or subversion. Capital is thus also coming to the fore in the cultural superstructure.
Until now, this social subtext has always been inevitably present. Simply because they were produced by members of society through wage labor. The monstrosities that appear in horror games, for example,[5] raise simple questions about the conditions – including the working conditions in the video game industry – that give rise to them. There is nothing left to decipher in machine-generated content – it is purely algorithmic variations. The cultural goods generated by “machine intelligence” thus represent a final ideological triumph of capital in the phase of its world-historical agony.
Especially in the video game industry, which has long since become the dominant sector of the culture industry, the possibilities for machine-generated content seem almost limitless. Valve, operator of the largest platform for PC games, announced new rules in mid-January 2024 that should allow the “vast majority” of AI games to be offered for sale on the Steam digital marketplace. The new rules for AI content also make it clear what is currently possible in the industry. Game publishers must state whether their game contains machine-generated graphics and objects, sound effects and music tracks, or even program code. In addition, it must be stated whether the games use AI systems during the game process that generate content “live,” in real time.
The pioneer of the AI game sector was the text adventure AI Dungeon, which was released in 2021,[6] which is in fact a simple dialog game with a chat system, where the shortcomings of the machine intelligence still have to be concealed by an appropriate game setting. The usual problem of the AI’s “catastrophic forgetfulness,” which repeatedly affects the game’s plot, is glossed over by the game’s objective, which is for the player to escape from a multiverse in which they are trapped.[7] Dreamino wants to go one step further,[8] to generate storylines, graphics and voice output in real time using content systems that react to the player’s actions. The text adventure is to be further developed into a graphic adventure with voice output and graphics. The game will dynamically generate text, graphics, storylines and voice output in response to player actions.
More games whose graphics, models and sound effects were generated almost entirely by AI content systems are currently in development. The graphics of the point-and-click adventure game Zarathustra[9] were largely generated by the DALL-E 3 content system.[10] Its voice output – the biggest cost driver for indie projects like this – was created using the Elevenlabs text-to-speech system.[11] Game designer Jussi Kemppainen has also already developed a prototype of a cyberpunk adventure whose backgrounds and characters were generated by AI systems.[12] In a blog post, however, the designer made it clear that the content generated by the machine still requires extensive post-processing (lighting effects, shadows).[13] Nevertheless, a qualitative upheaval is taking place here in the cultural-industrial production process, in which the roles of machine and human are reversed: The human now only corrects the content that the machine spits out. In addition, the transitions between AI content and manual work in the games industry are fluid.
It is still poor indie designers and producers of B-goods in the games industry, such as the mobile game manufacturer mentioned at the beginning, who are relying on AI content, but over time this trend will catch on due to the potential savings and new possibilities. The immensely popular segment of so-called roguelike games such as Dead Cells, Caves of Qud,[14] Teleglitch,[15] Risk of Rain 2, Jupiter Hell, Darkest Dungeon 2, Undermine and Hades is likely to act as a gateway with mass effect.
This game genre already thrives on the fact that each new game is generated anew by random generators and algorithms, so that the level structure, game items and gameplay always vary. The problem with this is that the game producers have to create a huge number of game items (weapons, armor, equipment, spells, etc.) in order to create the illusion of constantly new game sequences. The advantages of mass machine-generated content are obvious as soon as the technology is reasonably mature. Millions, not thousands, of items could be incorporated into rougelikes, even by small indie developers. It may also be possible to generate this content in real time – even with enemies that would change with each game run. Their variations have so far been very limited to a few dozen enemy types due to the amount of work involved.
Hollywood, Copyright and AI
In contrast to the games industry, which has always worked with digital content anyway, film production seemed safe from being taken over by AI systems, at least in terms of content – despite the massive use of digital technology and computer-generated graphics. Who wants to admire six-fingered actors from the incubator of clumsy pattern-recognition machines? However, the situation in Hollywood seems to be changing fundamentally, as the ever more perfected machine systems are likely to take over a large part of the production process in this sector of the cultural industry.
The protracted strike by screenwriters in 2023 was already overshadowed by the possibilities of machine-generated plots, which can easily imitate the plots of the very mass-produced films that the industry produces. The strike ended with clauses that allow the AI industry to use screenwriters’ works as data material for AI training only with their consent.[16] Nevertheless, such agreements, which are full of loopholes,[17] are reminiscent of the futile attempts of the defunct craft guilds to protect themselves from free competition in the late Middle Ages. During the strike, Netflix advertised a job posting for AI experts to help create “great content” for a fee of $900,000.[18]
Hollywood is also facing a disruptive development that will cost a lot of jobs, warned actor and producer Tyler Perry in an interview with the Hollywood Reporter.[19] Perry was about to invest $800 million in the expansion of his film studio, as part of which 12 new film stages were to be built on a 133-hectare site near Atlanta. However, this huge investment has now been put on hold after the producer attended a demonstration of OpenAI’s Sora AI system, which converts text input into video footage. Investments in film studios are simply threatened with obsolescence within a few years.
Those present were “shocked” by the content system’s performance capabilities. According to Perry, traveling to film locations, the use of stage equipment and studios will be superfluous in the future. Everything is just a text input away from realization:
“If I wanted to be in the snow in Colorado, it’s text. If I wanted to write a scene on the moon, it’s text, and this AI can generate it like nothing. If I wanted to have two people in the living room in the mountains, I don’t have to build a set in the mountains, I don’t have to put a set on my lot. I can sit in an office and do this with a computer, which is shocking to me.”
So far, AI has played a minor role in the industry, says Perry, who as an actor himself tolerated digital touch-ups that made him look older in order to “save hours on make-up.” But as he watched the AI system presentation, he said he immediately became concerned about all the wage earners who will be affected by this disruptive technology. This affects not only electricians, transporters, sound designers or editors, but also actors. The AI revolution will affect “every corner of our industry,” everything is now “up in the air” because the technology is “moving so quickly,” lamented the producer, who made a helpless appeal to the state: “There’s got to be some sort of regulations in order to protect us. If not, I just don’t see how we survive.”
The possibilities of machine content systems have thus reached production maturity. They can generate videos so well from the mountains of data available to them that even hardened, billion-dollar Hollywood producers are panicking and calling for state intervention. In view of the standardized products, the usual, hackneyed plots and the basic foundations of the culture industry outlined above, which only reproduces the surface of reality in order to affirm it, this panic on the part of content producers, who always think of themselves as “artists,” is only too justified. Precisely because AI produces nothing really new and only reproduces the given in new variations, it is superior to wage earners working in the culture industry. Humans are a potentially subversive uncertainty factor in “content production” that will be eliminated in order to save costs and streamline the production process. Precisely because of the unfolding global crisis of capital, it is an essential advantage to largely automate the production of culture industry goods.
And it is not primarily strikes in Hollywood or legislative initiatives in Washington that are getting in the way of the AI industry’s march. It is capitalism that is tripping itself up in the form of copyright. The IT companies that scanned large parts of the internet to accumulate the mountains of data they needed to train their AI systems were operating in a legal gray area. They were simply quicker than the lawmakers. In many cases, the legal battle to determine the limits of the legal use of machine content production is still ahead of the industry.[20] In addition, U.S. courts have already clearly ruled that pure AI content cannot be copyrighted.
A long series of legal disputes is looming on the horizon, in which players in the “old” culture industry based on human labor are taking action against the creations of the AI industry, as their content was formed from the “raw material” of their scanned cultural goods. So far, there are two ways in which companies are planning to deal with this legal uncertainty. Valve has given all users of the Steam gaming platform the option of immediately reporting “illegal” content that violates copyright. This delegates responsibility to the manufacturers of AI games. Microsoft, on the other hand, is turning legal uncertainty into a business: all customers who get into legal disputes through the use of its own AI tools will receive legal protection from the Group. This gives Microsoft an important competitive advantage in the market for AI systems, as it also acts as a deterrent. Who wants to go to court against one of the largest corporations in the world?
Nevertheless, these legal and political battles – in which lobbies will also fight for the concrete form of the legal framework – are likely to delay the success of machine-generated “content” in the sphere of the culture industry at best. With the full implementation of AI in film, video games, music and writing – the B-good of journalism, the photojournalist, is already being replaced by AI in everyday tasks[21] – capital will finally come into its own in the cultural superstructure. The empty value abstraction will produce pure forms without any depth, which cannot even be what they pretend to be on the outside.
Where is the whole thing heading? Ultimately, producers like Tyler Perry or game designers like Todd Howard will also become largely superfluous as AI systems interlock and synthesize with the established network services that have long since locked Internet users in a gilded cage of algorithms. It is likely that highly personalized and newly generated everyday commodities of the AI culture industry are for the few privileged wage earners who will still be able to afford ideology in the disaster capitalism of the 21st century. The personalized video game, the personalized film, which is generated after work on the basis of the data trail that people already leave behind on the internet every day, should be feasible in the medium term. The various AI systems are then likely to compete primarily to provide customers with media content that they don’t even know they want.
[1] https://gnulinux.ch/ich-habe-alles-verloren
[2] https://github.com/mkiol/dsnote
[3] https://www.wired.co.uk/article/low-paid-workers-are-training-ai-models-for-tech-giants
[4] https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html?searchResultPosition=1
[5] https://doomwiki.org/wiki/Models
[6] https://store.steampowered.com/app/1519310/AI_Dungeon/
[7] https://hessian.ai/de/warum-neuronale-netze-katastrophal-vergesslich-sind/
[8] https://store.steampowered.com/app/2795060/DREAMIO_AIPowered_Adventures/
[9] https://www.gamingonlinux.com/2023/11/point-and-click-adventure-zarathustra-uses-ai-art-and-ai-voices/
[10] https://openai.com/dall-e-3
[11] https://elevenlabs.io/
[12] https://80.lv/articles/this-adventure-game-prototype-has-ai-generated-graphics/
[13] https://www.traffickinggame.com/ai-assisted-graphics/
[14] https://www.youtube.com/watch?v=o_PBfLbd3zw
[15] https://www.youtube.com/watch?v=nJTdwbutW9k
[16] https://www.wired.com/story/hollywood-actors-strike-ai-future-distruption/
[17] https://www.wired.com/story/writers-strike-hollywood-ai-protections/
[18] https://www.spiegel.de/netzwelt/web/netflix-bietet-ki-experten-900-000-dollar-streikende-schauspieler-empoert-a-7bac7f4a-782a-42d3-bef3-1c3f14cc8392
[19] https://www.hollywoodreporter.com/business/business-news/tyler-perry-ai-alarm-1235833276/
[20] https://hbr.org/2023/04/generative-ai-has-an-intellectual-property-problem
[21] https://www.forschung-und-wissen.de/nachrichten/technik/bild-zeitung-ersetzt-redakteure-durch-kuenstliche-intelligenz-13377679
Originally published on konicz.info on 03/05/2024
This text is part of the e-book Crisis Ideology: The Delusion and Reality of Late Capitalist Crisis Management, which was published at the beginning of March.