Part 3 of 4 in a series about AI and artistic creation.
Finally, we get to the good stuff. This question bugs me every time I see it online. Or, more accurately, the assertion that AI art isn’t art bugs me every time I see it online. Not because I disagree, mind you, but because everything surrounding the discussion feels undefined and flimsy. What do we mean art? If anyone considers something art, is it even possible to refute that? What’s even going on when someone generates something with an LLM?
We’ll start with the last question, because I think I’ve come up with an interesting framing for this whole debacle. Let’s say that you, dear reader, want to see a really cool picture of a dragon. Specifically, you want to see the sketch of a dragon with realistic anatomy like you’d find in a biology book from the 19th century. Well, if you looked up “Realistic Dragon Biology Sketch,” on Pinterest you’d find plenty of sketches that roughly fit your description. However, since Pinterest is a user platform with a search engine, you might struggle to find exactly what you’re looking for (though given the sheer amount of stuff on Pinterest, there’s probably some keyword combination that would do the trick). In some senses, you have now “generated” this image of a dragon.
Now, imagine doing the same thing but with the DALL-E image generator. It would, similarly to Pinterest, spit out something close to what you’re looking for, but not exactly. And, like Pinterest, DALL-E’s database is large, but what you can “search” in it is limited by what’s already there. And in both cases, there is a clear distinction between “searching” on Pinterest/DALL-E and drawing the image yourself. Searching,” on Pinterest or DALL-E is, well, searching. You’re trying to find existing content, either by having an algorithm give you things in its database according to keywords, or by having an algorithm tear apart existing things and smash them together according to your keywords. In both cases, you’re trying to “search for” or “generate” something to consume.
In my opinion, the best argument that Generative AI Art is not actually art is that all art involves both composition and consumption. GenAI art skips the composition and goes straight to the consumption. Someone using GenAI may be responsible for something “new,” but they are no more composing than someone looking up an image on Pinterest.
This becomes more clear – and more easily applicable to other art forms like writing – when looking at the learning process. In any compositional art there are recognizable mistakes beginners make. For example, new writers often overuse the passive voice, or tell their readers something is true without showing them. Similarly, amateur videographers often forget to leave headspace for their actors, and those new to drawing often suffer from “same-face syndrome.” This type of error cannot exist, by definition, in art made solely using GenAI.
Therefore, if you consider art to require composition, then AI art cannot be art. It can be argued, however, that art doesn’t require composition. In this worldview, the most important aspect of art is simply that it can be consumed. To rebuke this argument, we must finally address the aspect of art which I, personally find most important. It’s something I’ve been talking around for the last few paragraphs.
Continuity.