DALL·E 2, Artificial Intelligence, and Objective Art

How do we define art and artist when it comes to image-generating programs?

“A.I.”-made “art” has never inspired anything more than a shrugging response in me. This creates a bit of difficulty with understanding the explosive appeal of the recently publicly released DALL-E 2 and Midjourney technologies. To describe their functions and productions as novelty-based would, I think, be accurate; but to leave it at that — as an explaining-away dismissal — doesn’t do much intellectual lifting either, since novelty, a techno-corporate conveyor belt of things which continually overrides the previous item or “experience”, is predominantly what our cultural productions run on now. So, let me just state: I am distinctly unmoved and unexcited by these programs’ productions (even if they represent a technological leap (they do)) and the novelty of a thing very rarely moves me to action outside of an attempt at critical exploration.

So, I’m going to do a bit of that critical exploration, with a focus on DALL-E.

On one level, it is mildly sociologically interesting how DALL-E 2′s capabilities tie into the wider social media phenomenon wherein the first impulse is to make a sort of joke. One can see this phenomenon play out countless times a day on Reddit by looking at the top rated comment(s) of a given thread, or by searching for the most popular posts on Twitter concerning a recent event or hot (sub-)cultural item. The aim of these graphic combinations is distinctly not artistic-poetic sophistication but, rather, a brute-childish “lmao i squished the two things together” act of play.

This would appear to tie into how the idea of a meme (a term which, in its fullest sense, comprises a crucial part of languages’ operative power) is now implicitly comedic. The things that spread the fastest online are charged by humor or anger, or a mixture of the two, and anything that can facilitate that process in a sort of automatic way is likely to be adopted on a large-scale by users. DALL-E 2 has acquired a functional near-equivalence to something like Meme Generator, although its element of chance — Is it able to create the visual combination I’ve specified? — endows user interaction with a game-like aspect, giving it added entertainment value.

What often seems to be most impressive to both audiences and the prompt-feeders about these examples is not the images themselves, as standalone “artistic” works, but the program’s ability to approximate the desired combination with either vague recognizability or convincing fidelity. It is in that way that I understand the phenomenon as novelty-based. There’s a sort of admiration of raw technical success here, a kind of humored “gee, wouldja look at that” sentimentality over feeding the machine one’s absurd idea and it succeeding, on some level, representationally.

I am tempted to form a connection with the multiverse/universe-colliding trend in superhero media and other comparable media. With those projects, I get the sense that some wall of (corporate) imagination has been reached, and the only purported way to simulate/stimulate creative development is to push everything beforehand together for a sense of cohesion and narrative elongation. Individuation is lost as the conglomeration of All of the Content takes precedence. Similarly, with DALL-E 2′s images, what is driving a lot of the content’s production is a “What-if” desire to form mounds of history-/chronology-jumbling ersatz that may be amusing in the moment, because of that frisson of combinatory recognizability, but which also seems historically dislocated, bound to the inattention-forming habits of the Internet.

When someone in a server I attend (half-jokingly) wrote, “think of all the porn this can generate”, their comment seemed unintentionally spot on to me, comparatively speaking. Pornography — or the (masculine) pornographying impulse within sexual play — and technically advanced image-generation such as that afforded by DALL-E 2 offer a similar, frisson-like pleasure in putting things together which “shouldn’t” go to together and suddenly seeing their absurd, sometimes grotesque, overlap. Prompts such as “Bugs Bunny as a Greek statue in Athens” or “still of Shrek in Spirited Away (2001)″ are really not that far off from “asian bbw inflating a balloon.”

Another person elsewhere wrote, “[I don’t] like the way AI art interacts with a physical multiverse which contains all that can be imagined. There’s something distasteful about the way this reaches into the realm of… platonic realism, idk.” This reminded me of the trend of grotesque, algorithm-generated “kid-oriented content” on YouTube, described by author James Bridle in a 2017 essay. The machine cycles through an Internet-derived bank of content, producing countless combinations of material with a simultaneous effortless resourcefulness and somehow-wrong ugliness.

All of this, however, fails to broach the question: “How are we defining art here?”

When we say or write the term “A.I.-art”, we are propagating a misnomer, not unrelated to the term “artificial intelligence”, on the basis of some surface level resemblances. Claiming that an image generated by a computer program is art — because it is a picture which has been made — or that a program which responds to text is intelligent, is like pointing to a submarine and fish and saying that, because they both “swim”, the submarine is a fish. Resemblances do not make equivalents.

Sociologist Elena Esposito has remarked that to engage an “A.I.” like ChatGPT is to stimulate and encounter a form of communication which we do not otherwise have outside of human society; and, lacking other referents for literate written communication, we often seem inclined to regard similitude as exactitude. But it’s not clear at all that, within these interactions, anything more than mere communication is occurring. Despite the protestations of persons who accusingly shout “Anthropocentrism!”, communication is no more an indication of intelligence on the non-human side than the chirps and warbles of birds are an indication of language.

Let’s reconsider the question above in tandem with: “Who is the artist here?”

One crucially important distinction of these image-making programs is that the machine cannot (yet?) refuse a request. It can only acquiesce and then approximate. The productive function of DALL-E2 orbits a programmatic requirement of curatorial appropriateness, the variables of which are content relevance and graphical fidelity. Without the mediation of the (inter)subjective psyche, then, what we seem to have here are the stirrings of an “objective ‘art’” couched among a larger industrial context. The nearest equivalent of this “art” might be something akin to poems making poems about poems.

Note that, even within the designation of “objective ‘art’”, I’ve continued to use quotation marks around “art”, for I’m not convinced that art-making can exist without the (inter)subjective psyche. Subjectivity — the porous, embodied state of discriminating being within personhood — is central to the creative process because it is the poetic response of a body-mind to other subjectivities and to qualia. Programs categorically do not experience bodies, disparity, or qualia. In fact, it’s not evident that programs have experiences at all. What they “have” are information states which proceed according to a predeterminable order of operations.

It is critical that we do not confuse qualia for information (do you experience the world as information?), or the mind for a will-overriding computer, as a number of figureheads within the tech spheres seem intent on having us do. Specialists, after all, are obsessed with seeing the world according to the strictures of their disciplinary domain. Propagating this narrative is fairly easy, though, when culture is technophilic.

As a person with whom I was in conversation about this topic wrote:

…these particular AIs don't do any sort of creative reinterpretation. We know this because we have specifically trained them to merely produce images that they "think" would be best described by the prompt. They just so happen to also be probabilistic and can produce a whole class of different outputs for the same prompts; and I think it's plain to see that is not what an artist does at all.

So, if the program which produces the images is not the artist, then what about the person who feeds it the prompts? There is an obvious issue with this, because the person isn’t making anything. They are simply inputting the conditions for the making to occur. If this were how we defined the artist, then the concept artist who is hired for contracted work is not the artist; it is whoever hired them (I am sure some companies would like for this to be our definition; too bad for them).

One might counter this — saying that, in some sense, the person is making something, because they are the one with the idea. This would appear to bring us around to some rubric of intentionality. But if our definition of the artist is ultimately contingent upon not the executive function, but the conceptual function, then no one would ever need to make anything to be an artist. They’d just need to have ideas.

Well: we do, actually, find a slight parallel with this criterion in the cases of so-called art world celebrities such as Jeff Koons, Takashi Murakami, and Damien Hirst who long ago abandoned the practice of art-making as autopoetic tékhnē and have since converted to the Warholian model of works made by usually uncredited (and likely often unpaid or underpaid) assistants. It should be unsurprising to know that there have been attendant controversies about art and authorship with respect to these celebrities.

Perhaps this area should be a starting point for future debates about “A.I. art” and where creative loci apparently reside. It does push me in the direction that the closest analog we have for the artist here is the prompt-person (even though I think, with collaboration, we need humans on the side opposite of the decision-maker for it to be understood as an artistic collaboration). Yet here we run up against another problem — namely, that members of what may be very loosely designated as the “A.I. art community” do not identify themselves as the artists. Surely some people have attempted to pass off program-generated material as their own, but the standard for attribution within these domains is synonymous with a citation of the program used.

As such, I am left with the sense that “A.I. art” presents a bizarre situation where there is neither art nor artist and, instead, the barest facsimiles of the pair.

I will end (noting that, obviously, a lot of avenues have been left unexplored, or only partially tread) with a quote from a friend during a discussion about this topic.

I think this sort of thing could ultimately be good for the arts. It could cause a reckoning that leads us to understand that the value of made things-as-products falls away at a certain point, and we create in order to practice our humanity, our humanness, and to find whatever we find, as individuals, in artistic creation.

I came to this conclusion when I read a lot of praise for a poetry generator that turned out some genuinely interesting material, and then paired that with the tech-minded praise & enthusiasm over having “automated poetry.” […] There’s this notion that Heidegger explores in his essay on technology called Bestand, which can be translated as “standing in reserve.” He used frozen chicken in warehouses as an example. We store and store resources with the intent to use them, but they do not always actually get used, so our priority isn’t to use later but to store. And the units become meaningful in themselves, not because we have a relationship to them, but because they are stored.

And I believe this is very close to our attitudes on information, now, with the internet. I think it was a governing (if unintentional) concept behind the mass surveillance push after 9–11, which certainly wasn’t optimal surveillance. And this extends to computer generated poetry & art. What does it exist for? It is not a response. It is a thing. And the things are entertaining. But valuing art as things rather than as responses, and that is exclusively a property of human creation, turns them into Bestand. Things as things to exist.