Many artists are opposed to AI art, and there seem to be three central arguments. First, that these AI models were trained artists’ images without their permission. Second, that the use of AI technology endangers the livelihoods of artists. Third, that AI art is bad art. The first is a rights-based argument while the second is a consequentialist argument, and the third is an aesthetic argument.

1. Rights

Starting with the first argument, I think it’s an open question whether artists truly ought to have the right to be excluded from AI training. If an artist-in-training looked at images on Deviant Art to learn how to draw, we would not say that the artists of Deviant Art had the right to prevent them. If you look at art with an artist’s eye, you’re going to naturally learn from it, and it seems unreasonable to allow people to look at your art just as long as they don’t look too deeply, lest they learn something from it.

However, an AI model is substantially distinct from a human artist-in-training, in that it trains on vast number of images–and also it isn’t a person. You could argue that there’s no rule against humans because it’s simply impractical to delineate how they consume art, while it’s far more practical to restrain AI. AI also poses the risk of accidental plagiarism–which is not unlike human artists! But the risk of plagiarism might be significantly be higher in AI, which could affect our moral judgment.

It’s worth noting that AI trained on images on the internet was already widespread prior to AI art generators. Google’s image search and reverse image search are examples of AI. Social platforms like Pinterest are most certainly training their AI recommender systems on images. The main difference is that these older algorithms were used to generate recommendations and search results, not to generate new images.

If it’s purely a rights issue, why is a recommender algorithm okay, but an image generator not okay? Maybe neither is okay? We could say, recommender algorithms are different, because they draw our attention to artists, while AI art generators displace artists.

But at this point, the argument is no longer based on artist’s rights, and more about the consequences of using artwork in one way vs another. So let’s move on to the second argument, the consequentialist argument.

2. Consequences

I suspect that some artists’ livelihoods are in fact in danger. And there’s a lot of historical precedent for this, with all sorts of technologies either eliminating or greatly reducing certain types of work. In the field of art, people commonly cite the printing press, which destroyed the art of illuminated manuscripts, and photography, which destroyed portraiture.

I would also point to stuff like digital audio workstations (DAWs), vector graphics software, or just widespread accessibility of self-publishing, as things that have made certain art-related tasks easier to perform. In principle, technological tools that make an artist’s job more efficient may reduce the number of available jobs, because now the same work can be performed by fewer artists

That’s true if artists are being hired to perform the same work, but technological shifts also mean that artists are no longer doing the same work anymore. Technology may give entirely new work to artists. Photography may have destroyed portraiture, but it also introduced the new art form of photography. DAWs may have made it a lot easier to record, edit, and master music, but they also shifted listeners’ expectations, created new genres and modes of musical expression, and require a lot of specialized skill to use professionally.

So is technology reducing jobs for artists, or is it creating more of them? That’s really hard to say. I do not know how many portrait artists would be supported by society in an alternate timeline where photography doesn’t exist. There are a tremendous number of artists employed to produce movies and video games, but I can only guess how that compares to… whatever artists would be doing if not for that technology. There are also some unanswerable questions about the overall quality of jobs, and which jobs count as “art”.

These unanswerable questions may not even be relevant. If a new technology eliminates some jobs, and introduces new ones, why should you care about the new jobs if your job is the one being eliminated?

If you’re a certain kind of visual artist who draws things for commission, it looks bad. You’ve invested in a particular skill—drawing—and suddenly a lot of people can imitate it, fairly cheaply, using AI. Sure, there might be some new jobs for artists down the line, but that’s cold comfort to the artist who did not spend their time learning to use AI art, and instead spent their time learning to draw.

However, as these examples show, the loss of artists’ jobs isn’t necessarily connected to the alleged violation of artist’s consent. It’s perfectly possible for new technology to endanger jobs without doing anything to violate consent! While yes, photography may have materially harmed artists who spent a long time learning portraiture, I don’t think that means photography was wrong or unethical. By analogy, if AI art endangers artists’ jobs, that doesn’t mean AI art is unethical. AI art may be unethical for other reasons, such as violation of artists’ consent, but not because it eliminates jobs.

3. Aesthetics

So far I’ve been speculating on the consequences of AI art, but the truth is we don’t know. It may be the case that AI will cause rapid and disruptive transformation, but another possibility is that it will fizzle after the novelty wears off. Even if AI art is eventually transformative, that doesn’t necessarily make it disruptive. It may be that AI art still has a long way to go, and that artists will have plenty of time to adapt.

So this brings me to the third argument about AI art, that it’s bad art. Either it’s ugly, or it lacks creativity, or perhaps the lack of human input makes it just inherently unappealing. The implication is that AI art is not transformative; or if it is transformative, then it should not be.

This is an interesting argument, because it doesn’t imply any stance on whether AI art is ethical or not. Anyone on either side can hold a negative opinion on the artistic merits of AI art, and I’m not so high on it myself. But in the context of arguing about AI art, I think it’s more than just an aesthetic opinion. People loudly talk about how ugly AI art is because they want to persuade others of something. The goal is to persuade people to place less value on AI art, and more value on human artists.

To some extent, this comes across as self-serving. It’s artists trash talking another category of art in hopes that it will increase the economic value of their own work. If I went around promoting my origami by talking about how ugly other kinds of art is, it wouldn’t make me look good. (Although… trash talking other genres is a lot more common in music, so I guess it’s not entirely a unique thing.)

But regardless of whether it’s self-serving, there remains the question: is AI art bad?

One way to argue this question is to point at a bunch of examples and offer our personal opinions like. And it may disappoint you that I’m not going to do that, and I discourage doing it in the comments. I’m not into 2D visual art in the first place, I simply don’t have an opinion. Also it seems rude to artists to bring up their art solely for the purpose of swiping left or right in an internet argument.

But I have a few observations. First, our aesthetic judgments are often informed by what the art represents to us. If one person looks at AI art, and thinks about how it’s eliminating artists’ jobs and violating consent, that’s what the art represents to them, and that makes it ugly. Likewise, if someone else looks at AI art and thinks of the new possibilities and increased accessibility, they will have a higher opinion. And if those two people tried to persuade each other of their aesthetic viewpoints, they’re going to come to an impasse. This is all to say, maybe the aesthetic argument simply isn’t all that powerful, because our aesthetic opinions are determined by our stance on the other two arguments.

My second observation is that to some extent, the aesthetic value of art depends on the (perceived) effort that went into it. It’s only natural to look at intricately detailed drawings or technically challenging musical performances, and be in awe of the effort or skill on display. This also ties into the value of scarcity—technical skill is impressive, because it is perceived to be scarce. If you see an AI artwork, you may be unimpressed because it is easy to make (lack of effort) and that anyone could do it (lack of scarcity).

However, effortfulness is not a universal value. I’d like to develop this thought more in a later article, but as an origamist, I have a deep appreciation of art that invites the audience to think: you could do this too. There are also plenty of artforms that would commonly be considered too effortful. Blockbuster films represent an artform that is so effortful, you practically need a megacorp to do it, and some would say that diminishes their artistic merit. Although for the time being, I do not think AI will help with blockbuster films, and I do not think people would like it if it did.

With regard to scarcity… I’ve seen many people liken AI art to the recent NFT craze, but they’re quite opposite to each other in their views of scarcity. NFTs were premised on the idea that the value of art arises from its scarcity, while AI art imagines a world without scarcity. I don’t know, I think scarcity is bad, even if it has monetary value.

Something that complicates the aesthetic argument, is that AI art may improve in the future. This is often taken for granted, but I really don’t think we should take anything for granted. AI art might suffer from fundamental limitations. I also think that AI art could become more effortful in the future. People are writing single sentence prompts now, but it doesn’t give very fine control. I think artists will start to want more advanced tools that may require much more skill to use.

Conclusion

In my analysis, there are three main arguments in opposition to AI art: the rights argument, the consequentialist argument, and the aesthetics argument. But I find that none of arguments alone are decisive, and they tend to lean on each other to do the work. The most “load-bearing” argument is the argument from consequences. I think this has some merit, as AI art is likely to cause some people to lose their income source; however, this is true of many technologies and does not typically translate into saying the technology is unethical.

One weakness of this analysis is that it is purely analytical. That is to say, I did not try to answer any factual questions, such as the frequency of accidental plagiarism in AI art. I do not know the degree to which AI art is causing loss of income for artists, nor do I know whether it is introducing any new jobs. I do not know what is the future direction of AI art. These factual questions are significant enough that they could change our opinion on AI art, so I think this may be worth reassessing in the future.