You can’t put the genie back in the bottle, no matter what you do. There were warnings about the harm mechanization can do to industries, but artists figured, oh, not us. We’re different. Our work encapsulates the soul of humanity, and therefore, we can’t be replaced! Most artists – in all fields – were absolutely silent when mechanization and computerization devastated blacksmiths, glassblowers, woodworkers, and so many others whose styles and skills were plundered by industrialization for the profit of large corporations. They were also silent when AIs were being crafted for other fields that were about to face the chopping block of computerization. Truck drivers and cashiers weren’t artists, what they did wasn’t like art, despite the absolute centrality of those jobs for human civilization to continue to exist. No one eats without truck drivers handling cargo and cashiers selling it to you, not until they are replaced by machines.
And it happened fast. AI art went from a novelty to a conqueror in about five years. Even people who were involved in computer-generated art were surprised at the speed with which it happened. Most would have guessed writing the first to fall to computerization, since that was where most of the money had been spent. Still, eventually it became clear that AI generated art was where the future of visual art would be.
There was, unsurprisingly, corporate interest in AI art, particularly in fields where people paid human artists to do jobs who also had cash on hand: movie studios and video game companies. Large amounts of money were quickly pumped into AI art generation.
Then artists got organized and pressed suits into action. In December of 2022, a wave of court rulings forced the libraries upon which AI art generation computers depended to remove all copywritten art, called Artists v. AIs. Shortly thereafter, they all shut down because they were initially starved of the base images their statistical models relied upon.
At the time of the ruling, a legal philosopher opined, “I find the ruling predictable but the case curious. The artists who sued used the Disney-inspired, Disney-sponsored IP legislation to defend their art. Now, it is a matter of law that you can’t copy a specific piece of art for profit or even a whole class of intellectually protected material, but the ruling says styles themselves give the people who created those styles a legal right to the proceeds generated from works with stylistic similarity. I don’t think they’ve thought the consequences through.”
Early in 2023, however, headquartered out of the Caribbean, a new AI art generation service came online. The libraries of images that AI art generation relied on had been public databases for years, and many thousands of copies were scattered around the world. It just took a minute for someone to collect them and write the web app to make it easy to use, set up in a country with contempt for international copywrite infringement and to set up a payment method.
The old commercial services spent a lot of effort making sure their services were family-friendly. Not because their users demanded it, but in defiance of their users, to secure “legitimate money” from corporations such as Google and capitalists like Elon Musk. The new systems, already black market by definition, had no reason to seek legitimacy or protect themselves from charges of pornography. So, they offered no such protections and used that as a lure to get people to their systems, and the people came.
When the money in the new outfits was traced, they went back to gangsters. Mostly in Eastern Europe and East Asia, which were already on the cutting edge of cybercrimes. They had no need or interest in legitimate money. They embraced their role as an illegal outsider. They allowed people to upload and use large galleries of pornographic images in AI libraries. The new AI art services blanched only at child pornography. Not out of ethics or a sense of responsibility but because they knew that if they crossed that line that people would come to get them. Still, the new services were rife with pornography of the most intense kind. The service grew quickly since more users meant the ability and willingness to buy more GPU time – and then to use the new RISC chipsets optimized for AI art generation, which came in the middle of 2023 – and a large, intensely devoted fanbase that spent enormous time and energy training the AIs. They were now a permanent part of the digital art landscape. Art had been gangsterized like recreational drugs.
Nevertheless, the commercial applications of AI art generation were obvious to the big media companies, notably Disney and Discovery-Warner. Disney and Discovery-Warner both set up in-house AI art generation units. Disney fed into their system the whole catalog of Star Wars, Marvel comics, and the MCU, all Pixar, and Disney, itself. Discovery-Warner did the same with its extensive back catalog of movies, DC comics and the DC movies, all the Looney Tunes characters, and its amazing video game department. Not to mention that Discovery-Warner owned Time, with its catalog of spectacular, genre-defining photographs. Both companies also went on buying sprees to secure the catalogs of the most prolific digital artists, anime, and game studios that produced much of the most influential fantasy and science-fiction art. It didn’t cost much relative to an AI’s hardware and software development.
Of course, the Disney and Discovery-Warner systems used everything in the public domain and all the unlicensed work out there, and the work which was licensed with sufficiently permissive Creative Commons licenses. This catalog ran into the ten of millions of images.
Still, even with that, the size of the corporate databases was far smaller than the extensive research databases that trained the old AI art systems. However, it didn’t matter. As the programmers discussed at great length, while the old databases were very large, they were also of… middling quality. Yes, they had upwards of half a billion images, but the old AIs effectively used only about one percent of those images. The rest were, well, they were noise. They were numerically significant, but their weights inside the program were almost nil. Some AI engineers were happy because discarding the huge datasets would make the AI more efficient in time and energy, and would produce higher quality art because of the de facto curation of the new datasets.
As one said, “We’re only including the good stuff, now!”
And AI engineer said, “The artists should have talked to us about what the system did. They were, like, your system is dependent on a large number of images – that you need this crazy, huge database with all of our stuff in it. We would have told them the truth is more nuanced. Sure, it requires many images, but art isn’t created equal, guys. AI art generators are pretty good at mimicking the style of Renaissance masters, even though the number of images of these masters is a tiny fraction of a percent of the number of waifu anime images. But you told me which pictures I would do without, I’d say all of that waifu stuff could be put in the trash. Despite the number of images, waifu art plays a far smaller role in developing AI than the smaller body of images of the work of Renaissance masters, which is critical to both the Western artistic tradition and modern art.
“Not to mention, and this is critical, and we mentioned it so many times… most of the artists in the database were themselves derivative of other artists, right? You’re some digital artist on Artstation or Deviantart, the odds are that not only is your style highly similar to a famous artist but your material is itself about copywritten subjects. When the big players like Disney and Discovery-Warner got into the act, they didn’t need your Captain America fan art. They had all of the original art and could pay the tiny number of style leaders a few bucks to use whatever new styles they also wanted. Not that shit mattered, anyway, because we were making new styles every day! We were really pounding them out.”
The artists who had brought the suits were disappointed, then horrified. Except for a few top artists paid for their specific stylistic contributions, no one made any money. And even the top artists made less than they imagined. When faced with the size of the corporate databases – which often contained their work of many famous artists from their time as contractors, anyway – no one got paid. The artists believed that their art – all of it, in aggregate – was vital to the AI-generated art system. They were wrong. Most of them were irrelevant to the indexing due to the low weights assigned to the art of “minor” artists.
Worse, since the art libraries of Discovery-Warner and Disney AIs were wholly owned and licensed, they came to control the ownership of styles. After all, the old commercial systems had shut down because the artists successfully argued that IP laws protected styles. And now Disney and Discovery-Warner used those rules to prevent artists from selling art in protected styles – with the data and legal muscle to define what “style” meant. The legal assertion that present art created a future financial obligation was leveraged by the big media companies to squeeze small artists out of the field altogether.
In other words, if an artist started to get big, well, Disney and Discovery-Warner found that they had the tools to demonstrate the extent to which humans were copying the now-protected styles that the big corporations owned. That is what the Artists v. AIs decision meant – styles were in some sense owned by the people who came before. It was a well-understood capitalist legal construction, rarely implemented, called takings. It meant that a business was protected from any actions taken that would deprive an existing business of future profitability. It is something that the big media players had wanted, but they knew they couldn’t get it – that it was a line too far. But the artists, terrified about the future of AI art “taking their jobs,” gave it to them. The very people who would be most harshly treated by acknowledging the development of styles as legally protected copyrights were the most vulnerable people!
So the big media companies crushed their rivals. They bombarded art sites with cease and desist orders to force artists to use their systems to create art, with the attendant ownership of content and style that implied. After all, if artists were to be paid for stylistic likenesses of their art, they would also have to pay others when they used stylistic likenesses. It is what they asked for, and what they got.
All hail our new corporate and gangster art overlords.