Heat, Compute, and Hard Reboots: In Defense of AI Art
Why "Blood, Sweat, and Tears" is not the only currency when it comes to art + a story of how I bumped my head against an echo chamber.
Although I’d hoped to focus this blog on technically challenging, tricky aspects of AI, something interesting happened to me today which I can’t help but share. This post will be about the trickiest thing of all: ART, about whether or not AI-generated images can ever be called “art”, and about my first ever direct brush with cancel culture.
Backstory
My Relationship with Arts
I’ve spent many hours learning and engaging in traditional arts, both visual (mostly drawing), and performative (piano & singing). Here are a few example sketches/drawings/exercises:
As you can see above, I am not particularly good, but also not entirely clueless, though I don’t find my works especially artful or worthy of sharing. The point of showing you these is to earn a bit of your trust when I say that I do understand how much effort goes into creating visual artwork, and most forms of art in general.
Naturally, given this hobby of mine, I follow a number of Facebook groups dedicated to various forms of art. Earlier today, Facebook recommended me the “Artists Against Generative AI” community, which I joined, curious to learn what the discussions look like.
One post in particular caught my attention, raising the topic of whether or not AI creations can be ever called “art”. And that’s how it all started.
The Case Against Generative AI Art
Here is the post almost in its entirety (though the attached image got cropped a little, its gist should be clear too):
Here is its main argument as text (highlighting is mine):
[I’m] interested in the philosophy of art (specifically the question of which things can be called art and which cannot) and I'm compiling a number of arguments about whether or not ai images are art or whether ai app customers can call themselves artists. Though I'm trying to be as objective as possible, it really seems that all the good arguments are firmly against the possibility of ai images being art, while all the arguments in favour are easily countered (probably because the spotty oiks who make them are too lazy to have ever read a book on the subject).
For instance, some say that ai is just a tool; art materials like paint or clay.
When Mr and Mrs Andrews commissioned Thomas Gainsborough to paint their portrait, they issued him with a prompt and a text description of the image which they wanted him to produce. Would anyone say that the Andrews were the artists and that Gainsborough was a tool which they used? There is no ambiguity about who the artist was and who the patrons were. But when ai bros give their computers instructions, they perform the same act that the patron does, not that which the artist does.
Although slightly combative (referring to AI proponents as “spotty oiks”), the post seemed to genuinely invite a discussion, especially the “I'm trying to be as objective as possible” part.
Owing entirely to naivete and poor judgment, I volunteered to play the devil’s advocate (or the “spotty oik”, if you will), since a) I’d previously given this topic plenty of thought and arrived to a different conclusion b) I’m in an interesting position of being an AI professional who has a reasonably good familiarity with arts.
Thus, after spending nearly two hours to craft a reply, I clicked “submit” and… got immediately banned from the group (the reply was never even posted, as all comments by new members are manually moderated). This is despite the group’s rule #1 (screenshot below) which seemed to welcome polite discussions:
In all my years as a semi-professional contrarian, this’d never happened to me before. I’ve politely expressed dissenting opinions in various circumstances, and have never met such a caricature case of “if I ignore it, it might go away” response from anyone.
I will present my argument here since I would like to actually find out if it has any flaws. You are very much invited to disagree, as I would love to read your thoughts in the comments and engage in a productive discussion.
The Argument In Defense of AI Art
Let's first agree that we don't have a consensus on what constitutes art in general, whether or not AI is involved. Many scoff at contemporary art, for example, like when you display a glass cast your own genitalia in a museum and call it an "art piece" (this is not an exaggeration, by the way; I’ve recently stumbled upon one such “exhibit”). Intuitively, I tend to agree, in my view, contemporary art veered too much into "showmanship/expression" and too far from actual art.
To start somewhere, I offer this as a working definition (it’s not some new word in theory of arts, but rather how I tend to think about it):
For me, true art is defined by a genuine search for beauty/truth, the desire to express it in some medium, and a strife for excellence in that endeavor.
I hope we can more-or-less agree on this definition. If through such artistic search one arrives at casting his privates in glass — so be it, I would be fine calling it art too. As an aside, my reservation with much of contemporary art is not that it “looks weird,” but that I’m not convinced that the search was actually for artistic value rather than hype/controversy/etc.
With this definition in mind, below I give three arguments why Generative AI can be involved in creating true art.
1) Generative AI as a Tool
While the Gainsborough example in the post seems to defy the "AI is just a tool" argument, there are numerous counterexamples, e.g.
We often recognize film directors as one of the sources (if not the source) of the movie's artistic value, even though they might have never touched the camera while orchestrating the production in accordance with their vision.
A conductor might direct an orchestra performance and create art (an artistic interpretation) even though he is not "directly" involved in producing sounds, and haven't written the score.
Next, upon a closer look, the Gainsborough example itself is not very fair. Arguably, the "art" in that piece comes from Gainsborough not-so-subtly defying the given instructions (there is a great video on that, if you are interested).
Moreover, when instructions are closely followed, we often credit instruction-givers for the artistic product. For example, Michelangelo did not actually paint all of the Sistine Chapel, but can you name a single assistant of his? Historically, numerous famous masters delegated (to various degrees) the execution of their paintings to their apprentices, while still getting credited for the result. They might provide a sketch at the beginning or step in at the end to add final touches, but, fundamentally, is a rough sketch that far off from a very detailed AI prompt? What if the prompt is given in a sketch form? What if the AI "artist" does some editing before publication to fix things like the number of fingers or some other details?
Overall, I think it's fair to conclude that art is often created through a set of intermediaries, and the person who is orchestrating the process in accordance with his or her artistic vision is often the one credited (at least in part) with the creation of that art product. I don’t see why we can’t treat Generative AIs as very skilled but initiative-less apprentices, eager to follow instructions; such apprentices are not the “great artist mastermind” behind the project, but their involvement also doesn’t automatically strip the project of its artistic value.
2) Why AI Art Does Not “Feel” Like Art?
There might be a bit of confusion when it comes to AI work because people jump from "some AI work might be art" to "all pretty AI images are art". If a person types "'\imagine beautiful landscape, Hudson River School Style" into Midjourney — sure, he won't get art. But it's not because one can not, by definition, create art though AI, but rather because this person is a crappy "artist": there was no idea worth speaking about, no search for beauty in any meaningful sense, no "blood, sweat & tears" shed to create this “work” (shown below).
What we get is a bland pretty picture which rings false, like a run-off-the-mill pop song or the Hollywood music cues that remind semi-conscious viewers how they are supposed to feel about any given scene.
Now, don’t get me wrong, the picture is, in some ways, visually pleasing, and even has a bit of the Hudson River School feel it it, but it’s not art. I trust your artistic taste, dear reader, that it rings “false” for you too, that you can feel that something is lacking.
The best analogy I can think of is the the dominant seventh chord in Theodor Adorno’s “Philosophy Of New Music.” Adorno’s idea was that when the dominant seventh chord was first discovered, it was genuine art, an expression of tension, feeling, beauty. But today, a dominant seventh is just a cookie cutter "symbol" that telegraphs "tension is here," and has little to do with artistic expression. Similarly, we have a picture that telegraphs “I am a pretty landscape in the style of the Hudson River School,” but it does not represent anyone’s search for anything.
In other words, this picture is like a bad review paper turned into a painting: it regurgitates what has been done, but does not offer any novel perspectives or thoughts.
On the other hand, imagine an aspiring movie director who writes a scenario, and, not having the funds for hiring an artist, creates a short movie where all visuals are made by an AI.
Our director spends hundreds of hours to make it look consistent and accord with his artistic vision, painstakingly refining/reprompting every image until it is "just right." Let's also say that in the end, it's a beautifully executed, touching, heart-breaking short film. Would you say we can't call it art because Generative AI was involved? I think it should be art by any reasonable definition.
Basically, in a typical non-AI "director-artist" interaction, both the director and the artist contribute an element of "effort" and "search" to the artistic product. With today's Generative AI, it’s different: all the "true art" aspect has to come from the director while the AI serves in a workmanlike, subordinate, initiative-less role. In that sense, the AI becomes a very complicated tool, which is very different from a typical director-artist scenario, but does not disqualify the product from being art and the director from being an artist.
3) Beyond AI as a Tool:
"AI is just a tool" argument, while defending AI-using human artists, does so by throwing AI itself under the bus. It assumes right away that AIs are not capable of true artistic endeavor, that its products are derivative by definition as they simply “rehash” existing works.
At present, I agree that AI’s on their own can’t create art, but not because of the “rehashing” per se. No artist developed in a vacuum. Ask any famous writer/musician/artist (or even a chess player or a boxer) — they will list a number figures whose style and vision affected their works the most. We also see that early in their careers, many are somewhat trapped in imitation, where they can’t quite shake off the influence of their predecessors. But sooner or later, they develop their own “voice” which adds individuality to the “rehashed” output.
So it’s not “rehashing” that is problematic in current Generative AI art — if anything, AIs have an edge, being able to integrate and master more styles than any human. It’s the absence of the extra “search”, the extra toil that every artist must go through to discover themselves.
However, it's not clear that it will always stay this way. It might be possible to formalize what that "artistic search" is and come up with algorithmic analogs of it. One thing that will still be lacking is the element of "human effort", and even "human suffering" which sometimes seems to add value to art. E.g. if I know that somebody truly suffered “in the name of art,” in their attempts to achieve excellence/create their artwork, I tend to give more respect to their creation compared to the case when they just picked up the craft & came up with something over the weekend.
But then again, following this line of logic would give a very self-serving definition. If we say "art requires human effort," we don't give AIs a fair chance. I don't yet know what AI's “blood, sweat, and tears" might look like, but I don't see why, one day, they can't shed them alongside with human artists, even if their tears are rather "Heat, Compute, and Hard reboots" instead of bodily fluids.
I hope that by defending the frontier of human artistic achievement against AI intrusions, we don’t limit the amount of true art discovered and created by entities surpassing us in many ways. As one Russian Chess commentator put it, looking at Alpha Zero’s games, “it’s as if Aliens or Gods descended upon earth to show us how chess should be played.” I want to see what such Aliens or Gods might show us in visual arts, music, writing, and cinema. And it seems to me that anyone who truly loves art should be looking forward to it as well, even if one day we, humans, have to pay the price of being deeply humbled when Da Vinci’s works begin to look like cave art in comparison.
P.S.
I resonate with the many artists who see their jobs at risk, and I understand that apart from the abstract "can AI make art?" question, there are many very real human lives affected by this. I understand that it feels very unfair to see a skill that one had spent thousands of hours on potentially substituted by a machine. To be intellectually honest, however, we have to try to think of it in a more abstract sense, to give the "machine" a fair consideration, apart from how we personally might be affected by it.
On my optimistic days, I hope that the overabundance of AI art increases the value of physical arts, and the value of live artistic performances, e.g. live drawing sessions. Humans do still value excellence, even when human ability is surpassed by technology. We still watch the Olympics even though no runner could outrun a car and no weightlifter could beat a forklift. We admire not only the product, but also the person who pushes the boundaries of what is humanly possible, while also understanding that the “machine-possible” or “AI-possible” boundary might be far ahead.
On my most optimistic days, I even dare to hope that for some people, the abundance of crappy AI "art" might actually help to grasp the difference between art and pretty pictures, mass-produced pop music, and the like.
P.P.S.
Let me know what you think on the topic by replying to these polls & please share your thoughts in the comments.