There’s a lot of talk right now around AI art, and I’ve mostly just patiently listened to it. However, more and more I feel the need to yammer at length about it, because several aspects of how this gets discussed are frustrating to me.
“AI” stands for Artificial Intelligence, and is used with all of the exacting precision we’ve come to expect from how the terms “artificial” and “intelligence” are themselves deployed. AI gets used to mean a lot of things – from a carefully programmed set of instructions to control the behavior of entities in games, to science-fiction computer super-intelligences, to algorithms for mashing a bunch of randomized inputs together until they match an expected output. The last example here are referred to as Machine Learning algorithms (the acronym “ML” gets pretty confusing if you spend a lot of time around left-leaning programmers). There’s three parts to this process: First, getting the list of stuff to mash together, second, doing all the mashing, and third, developing a way to assess how close this mishmash is to what you want. Any one of these three steps can introduce unforeseen biases – and, due to the black-box nature of these meta-algorithms, none of these biases are easily identified once they’re introduced.
Regardless, this technique can be, in certain circumstances, kind of cool. The process of masticating a bunch of input and regurgitating it is not necessarily a creative one, but it can be a generative one: I think most artists have probably played with processes for gleaning inspiration from random occurrences, from looking at passing clouds or from Rorscach tests or from pulling words from a hat, and these processes can be satisfying and elucidating in their own right. ML is a particularly powerful version of such processes, since it has the vast iterative power offered by modern processing, but is still fundamentally pretty similar to a set of dice with words printed on them. Artists also have a number of processes where the specific result doesn’t matter so much as that it creates the correct impression; creating the texture of tree bark or of flowing water, it’s seldom necessary to get any particular line or highlight correct, but vital to create the right texture, convey the correct impression, and uncontrolled processes like throwing gobs or specks of paint or scraping the canvas might work to generate this substance – here, too, such ML algorithms could be useful in digital art, generating the right kind of noisy detail fills.
What ML is not useful for, what it will never be useful for so long as it uses this methodology, is generating art. The reasons for this get down to the core of what art is: Art is a method of communication, an intentionally engineered solution to the problem of how to convey a complex and abstract idea. ML can never make art because it is a conversation with no one: What it generates could be interesting, but it’s just random noise filtered in a particular way. Whoever interprets meaning from that could find something interesting, could be inspired or moved – but the same is true of sitting in a forest and listening to a river. The process may be beautiful, but without intent it fails the one and only test of art, building a connection between two or more human beings. In other words, art is art because it has an artist.
Perhaps the thing I find most insulting and frustrating about this is I actually love the idea of artificial intelligence art. Imagine a whole other class of sapient being who lives with us, who was born of our effort! What would such a being have to say? What would their perspective be? What concepts would they try to convey to us through art, and how would we receive them? But of course, that isn’t what this is. Nor is this a tool created by an artist to express an idea, some abstracted set of rules for generating meaning, some meta-medium that can generate many artistic experiences by building off a set of authored rules (though, tangentially, it could be argued that this is what a video game is). It is, in practice, simply a system for chewing up and spitting out existing work. I’m not one to be prescriptivist when it comes to art. I think anything a person cares to make and to call art can be art – but they do have to make it, curate it, place it, assign meaning to it. I know the objection that’s springing to every defender’s virtual lips now: “I wrote the prompt, I chose the result, therefore I’m the artist. I provided the intent, the machine just filled in the details – how is this different from an artist using a textured brush or other tool?”
There are two reasons why I still don’t buy this argument. First: The direction these tools are being developed in, towards ever finer detail and rendering, and the way the people supposedly advocating towards the future of them present the results, build towards a shallow aesthetic beauty, something which looks nice on first glance, rather than towards results which are surprising and expressive – that is, the way ML art is currently rhetorically deployed is the furthest possible thing from the aesthetic goals that define an artist’s approach. The ML tools making the rounds a few years ago, the tools that spat out grotesque nonsense, where you could see the connections the machine was trying to make and failing to articulate, were often fascinating prompts for the imagination, a weird dream slurry of related ideas, but as these tools “improve” they only drift further from the realm of the interesting and expressive. Second: While ML tools may spit up solutions to problems, they are in fact regurgitating existing solutions to problems – an aspect which, as well, is exacerbated by their current rhetorical deployment in “real art” arguments made by people with a rotted view of what real art is supposed to look like. The output that manifests this way is doomed towards genericism (in the case of harvesting too many inputs) or plagiarism (in the case of harvesting too few) as an innate property of the methodology of the tool.
Of course, these mistakes aren’t limited to machines. Countless human artists, many who should know better, make the mistake of imitating the form of something they liked rather than attempting to understand its construction and apply the lessons that can be inferred from it. So much of art, especially popular art, consists of these endless repetitions of contextless form, because we have come to identify certain shallow traits as the correct way to do things. Waves of movies with the same visual and production style, echoes of the same style of dialogue writing, signaling at fun or importance without bothering to construct anything fun or important; eras of game design with pointless leveling or crafting systems or unnecessary turret sections; novels which mistake maudlin sentiment for profoundness, television shows which mistake meanness for genius, and so forth.
Perhaps what really frustrates me so acutely in this advocacy of a future of machine art is what it reveals under the surface: So much of what people see in art is the immediate presentation, the basic aesthetic, and so little the craftsmanship, the meaning, the message. Bad enough were it limited to art, but the pattern repeats elsewhere: Sham experts citing sham science to convict real people; sham coordinators making sham foam pits for people to shatter their real spines in; sham politicians stoking sham outrages to get their real enemies murdered. It may have always been this way, but it’s moving faster and the stakes are higher, and I will never stop being angry about it. The awareness of impostor syndrome has been inverted into a belief that everyone is an impostor, that doing things well doesn’t matter, that simply mimicking the appearance of competence is identical to competence, and it’s getting people killed and it’s going to get more people killed. All of this is only possible in a world that holds deep contempt for both the concept of education, that which makes one expert, and labor, that which gives one experience – we hold capital to be the sole arbiters of correctness, to decide what’s real and what’s fake, what’s expertise and what’s bias, what’s art and what’s trash.