GEMINILOGGBOOKOBERDADAISTICUS
Art & Images: AI
First, a fact that is not obvious to everyone: Images are not necessarily art. I'm not primarily thinking of those images that clearly serve other purposes than being art, including photos in news reporting, scientific diagrams, or the cover of a fashion magazine. No, I refer to images that may have an artistic appeal, that are debatable candidates for being art, those images that depend on the art world's embrace for being recognised as art.
For the sake of the argument, I'll take the perspective of the institutional theory of art, according to which something is art only insofar as the art world recognises it as such. Usually, that involves getting your work of art exhibited in a gallery or a museum. There are problems of circularity with the institutional theory, that's for sure, but it can be useful to remind ourselves of the fact that there is an art world with a certain set of quality criteria, or ways of distinguishing those works that are deemed worthy of circulating in the system from those that are not.
There is inevitably some confusion (or, in polite terms, differing views) as to what constitutes a work of art. Still today, almost a century after Walter Benjamin wrote his essay on art in the age of mechanical reproduction, we may insist on the difference between original work of art and reproduction. Those cheap posters of Modernist paintings that may be bought and put on your wall are obviously reproductions, as are many of those fancy giclée prints in limited editions that you may buy for a much stiffer price. Adding to the confusion, some giclée prints are considered "original digital prints" and fine art. Images of art found on the internet are usually reproductions, unless they are in fact original digital art. But always when the words "digital" and "original" cross paths there is trouble around the corner. This is seen perhaps most clearly with NFTs.
AI and art has been discussed recently here in the gemini sphere, either from the perspective of "it creeps me out, I don't like it," or defended as an interesting creative and expressive possibility for the "digital native" artist. Generative AI art also raises the question of who's the author. After a full century of readymades, we should be prepared to accept the fact that what an artist declares to be art can in fact be art, as long as the rest of the art world agrees (still reasoning according the the institutional theory).
The invention of the camera changed painting forever. It happened to coincide, roughly, with the invention of the paint tube, which permitted plein air painting and led to impressionism. The camera didn't kill art, photography even became recognised as its own art form. Something similar, I think, will happen with AI as an artistic tool. As long as the artificially generated pictures are made for a human audience, there needs to be humans in the loop evaluating or curating the output and picking out items of interest. Generative adversarial networks may aim to automate the process, but ultimately human viewers will have their opinion and that's what matters.
The internet in general isn't the ideal exhibition space for art. There are exceptions, which I have written about earlier, but most images found on the internet do not participate in the art world unless they are also exhibited in galleries or museums or, at least, made by established artists. In particular, that is the case with AI generated images.
Conflating the categories of image and art also leads to an underestimation of the potential of art spaces and all the fascinating things that can be done with pictures as material objects. It's not necessarily a question of scale – blowing up images to monumental format is a safe trick to evoke awe in a way some scant pixels on a screen won't live up to. Pictures hung on a wall or, why not, projected onto the ceiling, stand in a relation to the room they're viewed in. Even the frame, if there is one, may contribute something of its own. The same is true of images viewed on your laptop or cell phone, with the difference that the context is anonymous and homogenous: same for all and not chosen by the artist. On the other hand, it's fair to argue that digital images should be viewed in their natural habitat.
For artists, I believe self-expression and stylistic distinction still matter. For that reason alone, I would assume that artists who choose to work with AI as a tool would want to come up with something that carries their recognisable stylistic fingerprint. It could be done by picking the diet to feed the monster (technical jargon for choosing corpora for training the network), or it could be done by selecting, combining, and post-processing output.
There is the question of the derivative nature of art that relies on big data and the exploitation of numerous previous artists. Standing on the shoulders of giants and all that. Ideally, artists should have the right to refuse anyone to train a neural network on their production, although it's not very realistic to expect the prohibition to be enforced. The energy expenditure for training the network (and additional hardware demands) should always be considered. These silly images do not come without a certain cost, and it's always appropriate to ask ourselves, is it still worth doing?
Finally, another important point. For AI generated art to successfully enter the contemporary art world, the artist would need to have a convincing project or a narrative that takes it beyond a mere demonstration of a novel technique. Maybe you can get away with it this year, but sooner or later the novelty of AI imagery as such will become tiresome old news.
For the rest of the discussion, here's some threads:
gemini://sol.cities.yesterweb.org/blog/20220603.gmi
gemini://rawtext.club/~deerbard/art/2022-06-02-lament-for-the-future.gmi
gemini://idiomdrottning.org/ai-art
also on the web:
https://idiomdrottning.org/ai-art
There's more to say on the topic, but that's for another day.