ARTISTS VS. ARTIFICIAL INTELLIGENCE – a Manifesto

For approximately a year now I have been fighting the indiscriminate and accelerated rise of Generative Artificial Intelligence platforms as best as I can. The name itself is misleading: they are not intelligent, much less artificial. An immeasurable number of human beings work in real time behind the scenes to guarantee the “quality” of what is spewed by these applications. A notable example are Kenyan moderators, responsible for filtering content and preventing harmful, biased or stereotypical results from being generated. They receive less than 2 dollars an hour to do the dirty work that ChatGPT is incapable of learning on its own and end up with irreparable damage to their mental health as a result of continued exposure to the most toxic things on the internet.

They are not intelligent because they are incapable of abstracting, learning, conceiving or creating. When we talk about generative image applications, such as Dall-E and Midjourney, as much as they are capable of deceiving some people with shiny images, the illusion breaks down when we look at the figures’ hands: some with six, eight, three fingers on each limb. Thanks to the manual action (who would’ve thought!) of the human beings who operate and moderate the results, some of the platforms have become more adept at turning out hands that are less scary and a little more convincing. Even so, far from what is intended to be perfect.

This is because the systems are obscure even to their creators. It is not known exactly how they work or what connections they make. Open AI, for example, is extremely obscure about how it develops its platforms, without revealing any documentation regarding how the models are trained or how they actually work. It is known, however, that this is a statistical game. The databases that feed the servers of such aberrations do so in a volume so high that the machine is capable of generating very convincing results – even if it doesn’t know what it spit out (search the “Chinese Room argument” and you’ll understand what I mean). The application collects fragments and, according to what it “learned”, recombines the pieces on an obscene scale and assembles what is later presented as the result of the prompt. Therefore, applications often hallucinate and spit out false, erroneous or harmful results. Keep this in mind before asking ChatGPT for medical or marital advice.

It is all just euphemism for high-tech plagiarism: large-scale copying of other people’s texts and images excavated from the internet. Stability AI and Midjourney are declared plagiarists, to the point of having suffered legal charges as a result. And even if OpenAI and Dall-E haven’t (yet) been sued, they are probably guilty of the same crimes. This is the first and most important argument which there is no way around: there is no ethical use of A.I. applications, since they a) exploit workers in subhuman conditions and b) literally steal other people’s intellectual property to feed their databases. Anyone who has used or regularly uses AI platforms, even if it is just to satisfy their curiosity, needs to know that they are actively contributing to an ecosystem that makes many people’s lives much worse. There is no longer any way to plead innocence on this matter.

Furthermore, synthetic image generating platforms directly affect artists. Firstly because, as I said, they use stolen images to feed their databases. AI-generated images are derivative by definition. Without my consent, my artworks have been used by applications that make millions in profit for the companies that control them. And if you have published or posted any of your work online on the internet or social media, it is likely that they have used your art too.

Secondly, because they steal work opportunities from artists. At a cost close to zero, they can artificially generate illustrations in different styles, or else emulate the lines of a renowned artist who would (fairly!) charge a high price to produce the same image that the plagiarizing machine spits out for free to those who write the prompt. In fact, A.I. reduces costs for companies that would have to resort to human artists. And these artists, in many cases, are left without work.

“To accepting the new technology” is not an option in this case. The artist who believes that his future lies in mastering these platforms (which, let’s be honest, is not difficult at all. One can learn to write these commands in less than an afternoon) is doomed to one of two possible fates: first, you will be the only survivor in a department or company, condemned to do the work of 10 or more professionals using A.I. platforms. Second, you will be replaced as soon as the application starts writting its own command prompts (which already happens in some paid versions of ChatGPT). Therefore, do not be fooled: “prompt master” or “prompt writer” will not be a profession or future for any artist or worker.

In short, generative A.I. platforms (I will never call such monstrosities “tools”) were not developed to assist you, but to replace you. And if you use them frequently, you are helping to train them and hasten your own defeat. In the words of Steven Zapata, a formidable fellow artist, “I am perplexed by my friends and colleagues who are looking into the lion’s mouth and saying that, after all, it might be hospitable in there, because they glimpsed a few pieces of food stuck in its teeth.”

THEIR ARGUMENTS AND OUR RESPONSES

1 – I.A.s ONLY COLLECT REFERENCES, IN THE SAME WAY ARTISTS DO.

Or, also, “there is nothing original in the world anymore, everything is a copy of something else”. It is a glaring false symmetry. This fallacious argument usually comes from non-artists, but I’ve seen it among our ranks as well. The data that feeds the bases behind Stable Diffusion, Imagen and Midjourney exceeds billions of images accompanied by descriptive texts. Art by me, by you, by established artists, beginner artists, living and dead artists make up these data bases, but not only. Private images, from closed photo albums, medical records and many more that I couldn’t even name. Images that wouldn’t even be available to me or to you if we wanted to use them, it would never be allowed. And under the pretext of carrying out this search for academic research and non-profit purposes, these companies train their applications on a free basis and then resell them to the end user, who only has access to the “premium” version of Midjourney or a more advanced version of ChatGPT by paying for it. If it sounds hypocritical, that’s because it is.

Furthermore, even though human artists collect, combine and transform references, they do so on an infinitely smaller scale and without the mechanical precision that a machine possesses. Even the most skilled artist is unable to reproduce exactly what he sees. Again quoting Zapata, “you could not trace a Michelangelo well enough to create confusion about which one is the original, and if you could, you are almost certainly already a talented draftsman in your own right, by the sweat of your brow. ”.

2 – IF THE ARTS ARE ON THE INTERNET, EVERYONE CAN USE THEM.

This is simply a lack of knowledge about how copyright works. In most countries, legislation protects creator’s authorship and the right to use and exploit any artistic production by its creator from the moment of conception. Brazilian copyright legislation is such a case.

3 – I CAN USE A.I. TO DO THE OPERATIONAL PARTS OF MY WORK, OR TO GENERATE THE INITIAL IMAGE ON WHICH I WORK NEXT.

You can, but it is unwise to do so. The most recent legislation in the US determines that images and texts originating from generative A.I. applications are in the public domain (which will likely be echoed in the EU and Canada). This means they cannot be licensed or commercially exploited unless they are later edited and modified by a human artist. But then, get this: you only have copyright ownership for the part of the image that you edited, not the entire image. The part that came from the A.I. remains in the public domain. You can even say that it will be very difficult to monitor this, but large game and animation studios have already come forward and started demanding that their artists, whether in-house or freelancers, do not use A.I. in any stage of their creative process so as not to run any legal risks later.

4 – IT’S JUST ANOTHER TECHNOLOGY, A TOOL. WITH PHOTOGRAPHY WAS THE SAME THING! ADAPT AND ACCEPT IT!

It is, in fact, a new technology that drastically changes our relationship with art and consumption. However, as I said a moment ago, it is far from being a tool for artists: it is a substitute, something created to eliminate the need for human artists. Photography did not replace painters: it freed them to explore other forms of art that were not restricted to representing “reality”. You would be surprised to know how many painters use photography and how many photographers use collages, drawings and other forms of interference. Not to mention that, even though it seems like something “easier” than drawing or painting, the photographer is not just pressing a button. He still needs to be there, on location, to capture his object. And he decides every step between his vision and the click: which lighting will it be? What objects should be shown in the background? What should be the focus, and what shouldn’t? What should the apperture be? the depth of field? the angle? It only looks easy for those who don’t know what they are talking about. And, because it is in fact a tool, it has never been a threat to the work of other artists, but rather a valuable addition to our arsenal of resources.

5 – WITH A.I. ANYONE CAN BE AN ARTIST, NOT JUST YOU, YOU ELITIST! STOP DISMISSING AI ARTISTS!

There’s no nice way to say this, but this is the truth: there is no such thing as an A.I. artist. This is because the artist is the one who produces, not the one who delegates. And, after writing the prompt, the person has zero power to interfere with the result. It can rewrite and edit the prompt, but not the individual elements that make up the final image. In other words, the “prompt master” is, in this regard, much more like the client than the artist. He may know what he wants, but all he does is ordering it. The level of sophistication of the prompt itself is irrelevant and the result it spews is generated by the machine, not by whoever wrote the command. So don’t ask others (or machines) to do it for you and then sign your name on the final work. If you want to be an artist, do like the rest of us: sit your butt in the chair and make art.

WHAT WE FIGHT FOR

Regulation: consent, fair compensation and credit

We fight for regulation that is capable of protecting not just artists, but everyone. Generative Artificial Intelligence systems can be immensely harmful due to their potential for plagiarism, slander and fraud (such as deep fakes), in addition to providing, in an extremely convincing way, false or misleading information that people take seriously. For starters, corporations should be held responsible for the content generated by these machines: do they give erroneous medical information? generate legal damages for someone (defamation or other type)? If a young person decides to take their own life after being advised by ChatGPT, who should be held responsible?

We also fight for regulation that determines that databases are fed ethically, using only images provided by artists who volunteer to do so. Furthermore, since such platforms are often used for profit, it is essential that artists receive financial compensation, in addition to credit when their images make up some synthetic result generated by A.I. And no, Adobe’s Firefly is not a saint. There are already documented cases of living artists being used as a prompt on Firefly without authorization, such as “make an illustration in the style of artist Dan Aroeira”. How would Firefly have learned this artist’s style without using his images in its stock? Although it was sold as a fairer system, which only uses images from free or loaned databases, it seems very clear that this is not the case at all. In short: we want our work to be used only with our consent; that when we give it, we are properly compensated and credited for its use. “Ah, but it’s impossible, there would be too many artists, blah blah”. That, dear tech companies, is not my problem. Go figure it out.

Finally, we fight for a “label” system. All of us internet users of the world have the right to know whether we are looking at an image produced by a human being or by a synthetic A.I. system. The same goes for text-based systems such as ChatGPT. Some type of identification is necessary to prevent us from being taken by surprise or deceived by images and content that may be more (or less, much less) than they appear: fraud, plagiarism or worse.

It may even be that victory is completely denied to us. But if we do not fight, we have already accepted defeat. Do we offer our heads and hearts to these platforms? That I refuse to do.

Believe it or not, fighting is infinitely more important than winning. Join me in the trenches, join the humans. To the artists.

WHAT ABOUT THE FUTURE?

A.I. is here to stay, there is no way around it. It is likely that a future in which the majority of available images are the result of A.I. systemsis nearby. As contradictory as it may seem, the best thing to do in this scenario is to continue drawing and making art, evolving our skills and our sensitivity, without even touching platforms such as these (just thinking about using one of these aberrations makes me shiver).

If you are a beginner, you may be seduced by the pretense facilities that these applications offer. However, you will end up failing to train your eyes, your skills and your sensitivity. It would be like a body with healthy legs using a wheelchair: you end up atrophying your muscles. Likewise, if you use ChatGPT to come up with ideas or Midjourney to develop concepts or to render your drawings, you will be stunting your creativity and your skills.

If you’re an experienced artist, on the other hand, you don’t need to use any of this crap. Nothing these freakish platforms spew out will compare to what you do. I speak from experience: my comics have never been more read, my illustrations more valued and my clients more present. In a year of popularization of A.I.s, shouldn’t it be the opposite? It turns out that those who take art seriously don’t accept less than human artists. And, in a future when most of the images available are likely from artificial intelligence, human sensitivity will have weight of gold. Mark my words.

Therefore, preserve and develop your artistic sensitivity as you have always done.

And remember: these inferior platforms do not produce a single drawing. They just steal, chop up and recombine pre-existing images. Therefore, comrade artist, every day that you pick up your pencil, digital stylus or brush to draw, is a day that you exercise your supremacy over these cold machines. As long as you make art, victory is already yours. It’s already ours.

– Dan Aroeira

REFERENCES

Steven Zapata – The end of art: an argument against image AIs: https://youtu.be/tjSxFAGP9Ss?si

A website to investigate whether your art has been used in the databases: https://haveibeentrained.com/

Stable Diffusion release without any mention of “artists”: https://stability.ai/news/stable-diffusion-announcement

Release from Imagen explaining how the dataset it is trained on is uncurated and contains “a wide variety of inappropriate content”: https://imagen.research.google/

Interview with David Holz, founder of Midjourney: https://bit.ly/davidholtz

A great article that summarizes data laundering techniques from AI companies: https://bit.ly/corruptdata

Pedro Loos – The Chinese Room Argument: https://bit.ly/quartochines

California Congresswoman Anna G. Eshoo writes to OSTP calling for investigation into unsafe AI release models: https://bit.ly/eshoonsa

Young Belgian dies by suicide after dialogue with ChatGPT: https://bit.ly/killergpt

US Copyright Office on AI-generated art: https://bit.ly/copywriteoffice

“On the Dangers of Stochastic Parrots” by Bender, Gebru, et al: https://bit.ly/3G7O6iC

Timnit Gebru is fired from Google: https://nyti.ms/3M71vv0

OpenAI paid Kenyan workers $2 a day to label disturbing content: https://bit.ly/40xX94s

Stable Diffusion is sued by artists: https://cbsn.ws/40vkdka