Updated: Dec 13, 2022
Yesterday, I wrote a post on social media and here that outlined as many of the arguments against AI-generated imagery as I could think of. Those posts have seen a bit of traffic, and so far nobody has suggested any additional criticisms, so let's confirm for now that these are the arguments that need exploration. In summary:
1. AI-generated imagery steals work opportunities from artists.
2. Using AI art engines to mimic an artist's style is a kind of copyright theft.
3. AI doesn't have a soul, or at least genuine artistic inspiration, and so anything they produce is without heart or real feeling.
4. AI-generated images are made instantly by typing in some words and so don't have the same artistic merit as art that is laboured over with time and effort.
5. AI-generated images can be produced by anyone, regardless of skill or training, and so the results possess less depth, subtlety, and value than work produced by an artist who has trained and practiced for years to achieve a high level of skill.
6. AI art engines tend to produce images that conform to stereotypes of attractiveness.
7. Some AI art engines sexualize subjects in the absence of an instruction to do so.
8. The ability of AI to convincingly mimic art by human artists creates confusion for the audience as to what is "real."
9. Inevitably, disturbed individuals are going to use AI art engines to create horrible shit.
If you can think of a problem that I have overlooked, please let me know. Regardless, there's a lot there to deal with. I should note that, so far, all of my own personal experience with AI art has been via Midjourney, and that all the images in this post were made by me using that engine.
An Important First Distinction
Of course there are, broadly speaking, two types of use that AI-generated art can be put to, namely professional or amateur. Is there money involved in some way? If a person is making AI art for fun, as a hobby, a diversion, or a method of personal creative exploration, then the issues of stolen work prospects and copyright violation are fairly moot; nothing material is being taken away from anyone. The other problems are still relevant, though, perhaps even more so because images created "for fun" likely see a wider spread in the spaces where art lives. There is one possibility, though, that makes all of these arguments irrelevant, and that involves using AI art engines privately for reasons of personal growth or therapy. Obviously, if the resulting images are never shared, there are few implications for the wider world. There are some possible exceptions here that will be discussed later.
As I write this thing, it's becoming clear that this discussion will likely be very long and will definitely be broken up into different parts. In light of that, I'm going to quickly summarize my own positions on these criticisms, then spend the rest of this post on just one aspect of them. Here we go.
1. AI generated art does and will deprive working artists of work and income. It will also provide work and income to those artists who use it as a tool in their creative process. This is a society-wide problem that affects many different occupations, not just artists, and it behooves us to come up with a way to deal with it.
2. For professional applications, using specific artists' names as prompts and producing work in their style without their permission is unethical and shitty, but probably not illegal. However, I'm not a lawyer.
3. It's irrelevant that AI doesn't have a soul, or a heart, or genuine artistic inspiration. Those things can be found in the person using the AI who is telling it what to do.
4. AI art engines do make images very quickly, but there is the opportunity for the user to provide a lot of detailed instruction to the AI before each image is made. There is also the ability to refine images after they are made (at least, there is with Midjourney). To get results out of AI that meet specific criteria and look the way you want them to takes quite a bit of work.
5. It's true that anyone can make a pretty picture using AI art engines, but that's also true of the camera (to be discussed at length below).
6. AI art will tend towards stereotypes of attractiveness, but usually only if the user is vague in their instruction to the AI or uses loaded terms like "beautiful" in their prompts. It's very easy to make ugly things, if you want to.
7. I would recommend avoiding any company that has trained its AI to unnecessarily sexualize its output. That says something about the people who work there, and reeks of toxicity. Gross.
8. The issue of the quality of AI art and its ability to fool people is too big to encapsulate here. Plus, I need to think about it some more. Rain check.
9. People absolutely will make horrible shit with AI, but people make horrible shit already, and we have laws about what is and isn't legal. Those laws should probably be updated, though, and the whole issue examined thoroughly in light of new technologies.
How AI-Generated Art is Like a Camera
There is an analogy that I keep coming back to because the correspondences are numerous and fitting, and that is the relationship between photography and AI imagery. When photography was invented, many artists at the time made arguments against it that sound very similar to some of the criticisms mentioned here.
"Photography was invented in the 1820s and though it remained a fledgling technology in the few decades thereafter, many artists and art critics still saw it as a threat... When critics weren’t wringing their hands about photography, they were deriding it. They saw photography merely as a thoughtless mechanism for replication, one that lacked, “that refined feeling and sentiment which animate the productions of a man of genius,” as one expressed in an 1855 issue of The Crayon. As long as “invention and feeling constitute essential qualities in a work of Art,” the writer argued, “Photography can never assume a higher rank than engraving.” (JSTOR Daily)
Like AI imagery, photographs can be generated very quickly with little effort, they can be devoid of meaning or artistic quality, and anyone of any skill level can easily make them. The internet is wallpapered with our selfies and holiday pics and fluffy kitty pics. Few people, though, would say that the casual photos captured by our phones are art. Photographs are elevated to art when they they are made or manipulated with artistic intent. I do not have a formal education in Fine Art or the Philosophy of Art (my unused degree is in Forensic Psychology), but I have been working as a scenic artist for over two decades. My experience leads me to believe that art happens when a creator has an artistic goal. Good art happens when that creator also has a well-developed skill set to help manifest that goal.
In the same way that a manual camera has settings that can be adjusted to control the appearance of a photo, AI art engines require a set of instructions, a "prompt," to tell the system what kind of image to produce. In Midjourney, the maximum length of a prompt is 6000 characters, with prompts consisting of plain language (e.g., "full-body-shot of a film-noir squid-detective-hybrid, shining a flashlight, city street, night") and commands that set values for things like the version of the engine, style, aspect ratio, resolution, and much else (e.g, --style 4b --stylize 1000 --chaos 100 --v 4). A photographer can flip the camera to automatic and simply press the button, or go full manual and specify from film speed to focal length. Similarly, an AI art engine user can simply enter a prompt of "a cyberpunk woman" or they can strongly shape what the AI generates at a very fine-grained level by understanding how prompts work and using that to full advantage. For both cameras and AI art engines, there is a a wide range in level of control over the process in order to make them accessible to all skill levels. In both cases, the vast majority of users will be casual, but there will also be a dedicated minority who are fascinated with the technology and its use as a creative tool.
This brings us to the moment of image creation, either the click of a shutter of the hitting of a Return key. The next installment will continue this analogy by looking at post-processing images.