Is AI generated Art "Art"? (and Other Concerns)
[Written on 9/02/2022]←← Take me back to the writings directory!
The exigence to write this post came from a Motherboard article I saw on Tumblr that read "An AI-Generated Artwork Won First Place at a State Fair Fine Arts Competition, and Artists Are Pissed". Likewise, the artists in the notes were pissed. I was wrapped up in this feeling too for a bit until I sat my brain down to think about things. A common theme for people responding in the notes of that post is that they kept trying to define what was "real" art. Often when you try to define a thing that's loose in nature (such as the concept of being a woman, which is a whole other can of worms), you end up excluding things that are very much a part of the thing you're trying to define. To the right you'll see a picute of the generated work in question. If you're an artist, you may have some reservations, but to me I think this piece has clear artistic value. The colors used work great with the composition, and while it still has that 'AI-generated fuzz' about it, some areas of the piece this kind of texture serves it well. This being said, I do not think generated art has the same kind of merit more established artforms do.
This kind of discourse surrounding what is art is not exclusive to generated art. While doing research for this topic, another medium that used to be in a similar place as generated art was photography. People didn't believe it to be real art since all a photographer had to do was point their camera and push a button. I'm not a photographer myself, but I know it takes more than that. Anyone can push a button on a camera, but it takes extended effort to get the right focus on your camera, the patience waiting for a good composition to form and good lighting conditions to create what we tend to know as art. Generated artwork can be seen in a similar way. When I made my initial argument on Tumblr, I did some investigating and found the algorithm used to create the artwork mentioned in the article. It was Midjourney, a Discord-based bot to which you fed prompts, and you could give it input based on how well it interpretted your prompts, as well as how to improve the images it has generated for you. Algorithms as I know it are just loose piles of information, so it's up to a human to feed it inputs to shape the algorithm's outputs. I imagine there's a good amount of technical knowledge you'd need to keep a hold of in order to best use keywords to shape a loose ball into something that can create great artistic works, but it would be unfair to say that the skills needed to produce generated works are of equal merit as photography, or to say that the skills needed to take a good photo are the same as the skills needed to create a good painting. A prompter doesn't have to know how to work a camera, find the right lens, understand how light reflects onto objects, or know how a line of action leads the eye to generate 'effective' art. Certain consumers may not care about the skill needed to create a piece, but regardless of taste, I think this will remain an immutable fact.
Going back to the artwork that inspired this writing, I want to say that the prompter (Jason Allen) didn't enter a generated piece into a digital painting competition. If he did, I would be jumping more quickly to say that the submission of the piece was unfair to the other artists who had to make theirs by hand — but that's not what happened. He submitted his piece under the "Digital Arts / Digitally-Manipulated Photography". For a piece to qualify as digital art/digitally-manipulated photography in this competition, all it had to be was "artistic practice that uses digital technology as part of the creative or presentation process", and this criteria it meets. It was up in the air whether the judges knew about the generated nature of the piece, but fortunately they did know, clearing up some moral ambiguity for me. This being said, the potential for some moral shenanigigans to pop up is worrying for other prompters who aren't as forward with labelling their art is worrying — not in the sense that I think it'd ruin other artforms, but because of the ensuing drama that would ensue if people found out the true nature of these works.
On the topic of generated art "ruining" or otherwise threatening the jobs of artists, as I saw often in the notes of the original Tumblr post, at a first glance the rise of this art definitely seems scary. While researching this topic, I came across the subreddit for Midjourney and this post show-casing fake Marvel villains. The quality of these generations almost meet those of photos or 3D renders if it weren't for some of the imperfections that give these away as generated art. Even with this in mind though, there is a trend with every good looking piece that I've noticed. Everything that looks good is of what I call a 'safe subject', these being prompts that the algorithm has tons of reference for, such as human faces, things related to nature (animals, landscapes, etc), and generally stuff for which there are tons of pictures of. The concept of a safe subject also applies to things that benefit from an abtract look (which is serviced by the look of 'AI generated fuzz'). Where Midjourney might struggle, a human artist could take and make more refined (the gist being no, in it's current state Midjourney won't replace human artists).
I know DALLE-mini is not the same as Midjourney, but they're similar artificial intelligences that could be used to understand one another.
A kind of example for what I mean is the concept of making a Pokemon realistic with Midjourney, which this person has already tried, by the way. You can probably recognize a few of the pieces as the Pokemon Pikachu, Eevee, Squirtle, and perhaps Charizard. These Pokemon are safe subjects because they're rather popular Pokemon and thus are more likely to have preexisting realistic renders done of them by human artists. But what if I asked this algorithm to generate me a realistic Gorebyss? Unfortunately I don't have the means or the will to try this myself, but based on how other algorithms like it work, I can speculate. What an algorithm like this might try to do is look for preexisting stuff that fits into the description you provided, which is already kind of a bad start since from what I've found there is very little realistic art of Gorebyss, especially in comparison to more popular Pokemon. From here you could describe other animals for the algorithm to reference in order to make the Gorebyss. With a human artist, the process of combining known features is more straight forward, but for Midjourney I imagine the process would involve some very heavy tuning in order to get the desired results onto the right places of the Pokemon's body. I wouldn't know how hard this would be to accomplish, nor how long, but I personally think past a certain point, it'd be better to leave this stuff up for a human artist.
From the small glimpses I've seen, it also seems like Midjourney seems to struggle with stylization, particularly with anime. If you describe to it a style that's quite popular (like the style of Dragon Ball Z), it can do it, but not without imperfections. Someone did the style I described in Midjourney, and from a glance it looks quite like DBZ! Upon closer inspection though, there are some flaws that detract from this, like the undefined hands and the eldritch horror that lives in the mouth of the right image. In other instances of anime-esque art I've seen, it looks like the algorithm struggles with connecting/making sense of thin, clean lines. A part of my theory is that with some anime artworks they use solid lines while others use inconsistent lines with gaps. Artists sometimes use inconsistent lineart to convey a lightness about their art, but an algorithm can't read intent, so it ends up unceremoniously putting the two together without knowing what it's doing. It could also just be that thicker lines are easier for the algorithm to connect, as I did see that clean, cartoony lines are possible.
Whenever I write these long articles I forget that I need to have some sort of conclusion, so I guess my bottom line is that algorithmically generated art does and should have its own place within the artistic community. Aside from allowing those who cannot naturally draw to create, I can see something like Midjourney be great for disabled folks who have lower ability, or are otherwise unable to reproduce the mechnical movements needed to draw. Less understanding is needed to make art with Midjourney, but it also feels callous to completely disregard it as an artform just because it might take over artists' careers.