In Defense of AI Art
This essay was originally concieved a few months ago (beginning of October?) after reading The Work of Art in the Age of Mechanical Reproduction by Walter Benjamin. Admittedly, much of his essay went over my head, especially on first reading, but at the least it put me into a more thoughtful mindset (reading it more closely while writing this, I think I understand it more). This was around the same time AI art was starting to become a more mainstream topic on Twitter, and it made me more resistant to the reactionary response to AI art's emergence. For example, in the months since, a lot of people have adopted the rhetoric that AI art isn't art at all, based on dubious arguments spread through Twitter artist circles. With this essay, I'm hoping to open people up to consider more closely their thoughts on AI art.
To begin, consider AI art as an analog to photography. Many anti-AI arguments have a close analog as anti-photography arguments. "A machine is doing all the work!" Sure, but the artist still has to set the machine up to match their artistic vision. Just as a photographer has to set the ISO and shutter speed to get the correct exposure, so too does an AI artist have to add "trending on artstation" to their prompt. I'm joking of course, but dedicated AI artists can spend hours trying to get the program to give them what they want (As an example, the guest on the latest Chapo Trap House episode told an anecdote about someone in the Midjourney Discord spending 2 hours getting the AI to create an anime girl with a perfectly sized ass). Or another argument, "AI art can't make anything new", because it is an amalgam of other works. While this is, in a way, true of all art, it is most clearly seen in photography. A photo can only contain something that already exists in the physical world—a photographer cannot create anything that does not already exist.
Moving on, however, and implied by the previous argument, we come upon the hairier question of "does AI art steal from artists?" Here we could also apply the camera metaphor: is taking a photo of something the same as stealing it? Obviously not, as Benjamin argues himself.
Even the most perfect reproduction of a work of art is lacking in one element: its presence in time and space, its unique existence at the place where it happens to be. This unique existence of the work of art determined the history to which it was subject throughout the time of its existence.
Benjamin terms this unique existence of an artwork its "Aura". A photograph cannot reproduce the aura of a captured subject, only those facets visible to the camera's lens. But here we run into the limits of this metaphor: a physical artwork can have a verifiable original, confirmed by the paints used or other dating methods. On the other hand, a reproducable artwork has no original:
To an ever greater degree the work of art reproduced becomes the work of art designed for reproducibility. From a photographic negative, for example, one can make any number of prints; to ask for the “authentic” print makes no sense.
Similarly to photographs, digital art can have no proper "authentic" source. Sure, there is the original file on the artist's hard drive, but a copy of this file is indistinguishable from the original. The "original" file of a digital artwork is nontransferable—this is one reason NFTs are nonsensical.
That which withers in the age of mechanical reproduction is the aura of the work of art.
Digital art of course exists within the age of mechanical reproduction. Therefore digital art, like a photograph, is created without any aura to be lost by reproductions. And so we can see that no, AI art is not stealing from artists. Even if someone constructed an "identity AI" that spits out an exact copy of whatever input is given, the AI would not, and could not, be stealing anything. There would simply be one more copy of the artwork in the world. (Incidentally, this is also why piracy isn't theft :])
...
But that's not the whole story, is it?
What these artists have a problem with is not that their art is being copied. What they have an issue with is not being credited for their work. An AI's neural network is insanely complicated and its inner workings are functionally unknowable, meaning tracing what training data is actually used in the construction of a given output is impossible. So, the only feasible solution would be to credit every single artist whose work is included in an AI's training dataset. Whether this is an acceptable solution to affected artists is ultimately up to them, so having an opt-in system for training data inclusion seems like a good idea. Unfortunately this would automatically exclude any works produced by artists that are no longer active, which is not ideal. Perhaps this could eventually be remedied if a positive consensus were reached on this crediting solution?
Another interesting question raised by the advent of AI art is one of copyright. Who owns the copyright for an AI art piece? The person who input the prompt? All the artists that contributed to the training database? The real question I think we should be asking is whether anyone should have copyright over these images. More blatantly than any other art form, AI art is a collective creation, built on the work of innumerable people. Not only artists, but the programmers and mathematicians whose work was used in actually building the models. I think that AI art can represent a new dawn against the oppressive copyright system, and create space for a new, open, and collective ownership of culture. (I'll save my diatribe about why we need open-source culture for another time)
In any case, thanks for reading. I'm not really a great essayist I think, but hopefully this made sense. If I haven't changed your mind on the issue then maybe you've at least found this interesting. I guess if you've gotten this far you probably did. Anyway, thanks again for reading.
Addendum
I'm writing this addendum both in response to this piece by Mr. Connor Crow, and to clarify my position in general. Let's begin briefly with the clarifications.
When I first wrote this essay, I was mostly concerned about the philosophical arguments being made against AI art. Frequently people were simply discarding it as "not art" rather than making any proper case against it. My goal was to offer resistance to this purely subjective argumentation, not make a full analysis of AI art—which leads us into my response to the wonderful Mr. Crow.
The 1st paragraph of Connor's 1st section makes an argument that the value of art is determined by how much the work challenged or advanced the skills of the artist. I find this argument somewhat strange. The aesthetic enjoyment of an artwork has no inherent connection to the artist. I can appreciate a painting without even knowing the name of the painter, let alone what challenges the work presented them.
His 2nd paragraph discusses the loss of humanity represented by AI art. I would argue that the "humanness" of the art is preserved in which pieces the artist chooses to publish. One might generate hundreds of pieces before settling on a satisfactory result, the properties of which are ultimately decided by a human. The AI may be regurgitating concepts, but a human still filters those results through their own experiences.
In response to his 3rd paragraph, I must restate my allegory with photography. A painter might criticise a photographer by saying the photographer hasn't had to learn the skills to create their desired image, but simply click a button and instantly get a result. This is clearly silly. A photographer may not need to know how to mix paints, but there are still principles they must learn.
Now onto the 2nd section. This section I largely agree with, albeit with some minor differences. He argues that artists will be put out of work by AI—which, although it seems unlikely to happen in the near future, almost certainly will eventually. I agree with his concern on this point, but take a broader view of the phenomenon. As with any automation technology in a capitalist economy, jobs will be lost in pursuit of higher and higher profit margins. This is not a flaw of the technology itself, but rather a consequence of the incentives created by capitalism. Capital will optimise toward one thing and one thing only—profit. Paying workers eats into those profits, and so if a new technology allows a captialist to pay fewer workers, it must be implemented in order to stay competitive. Connor's solution (expressed only implicitly in his essay) is banning of the use of AI by corporations. I agree with this as a sort of harm-reduction strategy, although of course in an ideal world we'd simply ban corporations period.
In closing, I think most people don't actually have a problem with AI art as such, but moreso with who is using it and how they are using it (read: annoying crypto bros making shitty copies of Ghibli art or whatever).