This is Jessica. In a new essay reflecting on how we get tempted to aestheticize generative AI, Ari Holtzman, Andrew, and I write:
Generative AIs produce creative outputs in the style of human expression. We argue that encounters with the outputs of modern generative AI models are mediated by the same kinds of aesthetic judgments that organize our interactions with artwork. The interpretation procedure we use on art we find in museums is not an innate human faculty, but one developed over history by disciplines such as art history and art criticism to fulfill certain social functions. This gives us pause when considering our reactions to generative AI, how we should approach this new medium, and why generative AI seems to incite so much fear about the future. We naturally inherit a conundrum of causal inference from the history of art: a work can be read as a symptom of the cultural conditions that influenced its creation while simultaneously being framed as a timeless, seemingly acausal distillation of an eternal human condition. In this essay, we focus on an unresolved tension when we bring this dilemma to bear in the context of generative AI: are we looking for proof that generated media reflects something about the conditions that created it or some eternal human essence? Are current modes of interpretation sufficient for this task? Historically, new forms of art have changed how art is interpreted, with such influence used as evidence that a work of art has touched some essential human truth. As generative AI influences contemporary aesthetic judgment we outline some of the pitfalls and traps in attempting to scrutinize what AI generated media “means.”
I’ve worked on a lot of articles in the past year or so, but this one is probably the most out-of-character. We are not exactly humanities scholars. And yet, I think there is some truth to the analogies we are making. Everywhere we seem to be witnessing the same sort of beauty contest, where some interaction with ChatGPT or another generative model is held up for scrutiny, and the conclusion drawn that it lacks a certain emergent “je ne sais quoi” that human creative expressions like great works of art achieve. We approach our interactions as though they have the same kind of heightened status as going to a museum, where it’s up to us to peer into the work to cultivate the right perspective on the significance of what we are seeing, and try to anticipate the future trajectory of the universal principle behind it.
At the same time, we postulate all sorts of causal relationships where conditions under which the model is created are thought to leave traces in the outputs – from technical details about the training process to the values of the organizations that give us the latest models – just like we analyze the hell out of what a work of art says about the culture that created it. And so we end up in a position where we can only recognize what we’re looking for when we see it, but what we are looking for can only be identified by what is lacking. Meanwhile, the artifacts that we judge can be read as a signal of anything and everything at once.
If this sounds counterproductive (because it is), it’s worth considering why these kinds of contradictory modes of reading objects have arisen in the past over the history of art: to keep fears at bay. By making our judgments as spectators seem essential to understanding the current moment, we gain a feeling of control.
And so, despite these contradictions, we see our appraisals of model outputs in the the current moment as correct and arising from some innate ability we have to recognize human intelligence. But aesthetic judgments have never been fixed – they have always evolved along with innovations in our ability to represent the world, whether through painting or photography or contemporary art. And so we should expect that with judgments of generative AI as well. We conclude by considering how the idea of taste and aesthetic judgment might continue to shape our interactions with generative model outputs, from “wireheading” to generative AI as a kind of art historical tool we can turn toward taste itself.
Yup.
As we discuss in our article, many computer-generated works do contain that “je ne sais quoi” that is familiar to us from human-created artworks, which is no surprise, given that (a) the computer generates the work by estimating parameters from, and thus abstracting from, corpuses of human-created art, and (b) when we, humans, create art, we often do it by a shuffling process, altering recombining existing ideas in a way not so much different from how these computer programs work.
Looking at it from another direction, as we also discuss in our article, much of human-created art and entertainment aspires to algorithmic production, in the sense that human authors are often looking for procedures that will get to the solution with minimal effort. Not always, but this is often the case.
>many computer-generated works do contain that “je ne sais quoi” that is familiar to us from human-created artworks, which is no surprise, given that (a) the computer generates the work by estimating parameters from, and thus abstracting from, corpuses of human-created art, and (b) when we, humans, create art, we often do it by a shuffling process, altering recombining existing ideas in a way not so much different from how these computer programs work.
It occurred to me we don’t say this quite so directly in the text, but we should. So, I updated the article.
Maybe next time we can write the whole paper via blog interactions!
Thank you for the interpretation, Andrew. A few weeks ago, I visited the Yayoi Kusama museum in Tokyo, where I saw her earlier works with more colors and patterns than in her more recent or famous works. It is as if she has been abstracting from her own earlier works, simplifying processes or reducing efforts, over the years…
Really appreciate this post!
“…are we looking for proof that generated media reflects something about the conditions that created it or some eternal human essence?”
I’m not, but I have a relative who publishes in research psychology who is fascinated by that NYTimes article about the chatbot interaction where the chatbot tried to get the guy to divorce his wife. He now takes for granted that there is something indistinguishable from a humanlike homunculus in there, and he has the skills and experience to torture it until it confesses.
I argued that he has spent his career studying human personality but has no experience with an entity that exists entirely to convince the user that it is something that it is not. Under the circumstances, his evaluation of the results will be based entirely upon aesthetics – the stuff where chatbots often do well – without any roots in scientific enquiry, the part he is supposed to provide. That imaginary homunculus is not quite like a greased pig, it’s a greased thing that morphs to satisfy your expectations of what it should be while remaining just beyond your grasp.
I find this a fascinating topic. I think there is some question about what is defined as ‘art’ – there is the sort of high art that ends up in galleries, but there is also huge amounts of ‘everyday’ art that doesn’t. For an AI there is no difference, the difference only depends on the viewer. Nick Cave said about AI songwriting: “Songs arise out of suffering, by which I mean they are predicated upon the complex, internal human struggle of creation and, well, as far as I know, algorithms don’t feel. Data doesn’t suffer.” That is fair enough and that is what music means to him, but there are huge proportions of society who simply see music as something to dance to on a Friday night and would not really care if the material came from a suffering artist or was generated automatically.