Empire of the senseless
Criticism has always been plagued by a certain skepticism about whether it is even necessary. Shouldn’t works of art be able to speak for themselves, to each in their own idiom? When a critic tries to describes a work, isn’t this just redundant at best, if not intrinsically distortive? Why should anyone defer to a critic’s tastes or interpretations when this could come at the expense of our own? Why let critics, for their own aggrandizement, place themselves between audiences and artists, mucking up the parasocial communion? As TikTok paintfluencer Devon Rodriguez wrote to art critic Ben Davis after siccing his legion of fans on him, “love will always outshine being a hater, I hope I taught you that today.”
Davis may have made a categorical error in trying to subject social media influencers to the kinds of contextualization traditionally applied to fine art. In insisting that we assess Rodriguez’s promotional videos as part of his art practice, Davis seems determined to treat Rodriguez as a potential figure in the art world, even though Rodriguez doesn’t operate in that market and has no apparent designs of intervening in that discourse. He seems more like Rupi Kaur, who is operating in a world of “pop poetry” apart from establishment poetry. She writes poems for people who will most likely never be invested in the distinctions and concerns of the institutionalized poetry world; what she and her readers call “poems” has little to do with what those institutions mean by the word.
The same could be said about Rodriguez and his massive audience. Davis points out that Rodriguez “has managed to rise to this status of apex visibility without any kind of critical writing about him at all,” as though this indicates a systemic flaw — as if popularity and critical attention should be aligned, or as if critics should seek to internalize pop culture figures to reinvigorate their marginal discourse. But critics aren’t needed to rationalize a performer’s popularity after the fact, not by their fans anyway. Davis’s claim implies a distinction between critical writing — a visibility among recognized authorities that legitimates artists within the tradition in which they purportedly operate — and “apex visibility” of the kind that can be whipped up algorithmically on platforms designed to aggregate it. But if you reject the relevance of tradition and the specialized “serious” discourse communities that administer it, there is no distinction. Visibility speaks for itself in one voice. It is its own form of revealed criticism: If you are highly visible, the platforms and their audiences and algorithms endorse you. Look at the scoreboard; it measures the “love.”
Davis argues that “the majority of Rodriguez’s fans are most engaged by his appealing social-media persona, not his actual artworks,” but that doesn’t mean his persona is amenable to being assessed as performance art. What that persona achieves — parasocial profitability — is completely separate from what art criticism needs “serious” art to do, and in bullying Davis, the legions of Rodriguez’s fans weren’t rejecting critics’ ideas about art’s higher calling; they were just strengthening their sense of the parasocial bond and extracting more value from it. (Whether or not swatting could be considered an art practice is also a separate question.) Being a fan has nothing to do with “art appreciation” or aesthetics. Critic and fan are antithetical standpoints, though it may not even make sense to say they are looking at the same object. The ability to accumulate fans, while of deep sociological significance, doesn’t seem like an artistic skill (is MrBeast art?), though I guess Andy Warhol thought differently. But the purpose of criticism isn’t to make fans or administrate fandoms.
***
What, then, do “serious” critics expect “serious” art to do, given that it is not to make money or to provide emotional comfort or culinary enjoyment? One answer to that (and I’m deriving this from the Adorno-driven art criticism in J.M. Bernstein’s book Against Voluptuous Bodies) might be that art brackets off a space in which our ways of thinking and experiencing and representing the world can be tested for their continued coherence and validity. Art allows for epistemological problems to be articulated, if not solved.
Another related answer is that art holds open a space between experience and how it is conceptualized, seeming to manifest the otherwise indescribable, ineffable aspects of experience — the stuff that resists discursivity — and assures us that such a realm (the realm of freedom, if you believe Kant) really exists. If something can be completely described, then it is subject to full, mechanized determination; it can’t be free. Proper artworks can’t be fully described or “put to use” — they can’t be exhausted by critical discourse or ordinary consumption — so they reveal freedom to us. A critic’s work, from that perspective, succeeds by failing — when its strenuously efforts to describe a piece serve to reveal its inexhaustibility, its ability to renew its meanings from some impenetrable, possibly noumenal source.
An artwork itself embodies the same paradox: It may most succeed when it “eludes and fails visual and perceptual claiming,” as Bernstein puts it in describing a piece by Jeanette Christensen. A work’s “own power of proliferating discourse” is what it both “wants and refuses” because its significance ultimately depends on manifesting and holding open the gap between what there is and what can be described (or mediated, or simulated, or reproduced, or predictively generated), the gap between words and things, between the meanings we project onto things and “things in themselves.” That is, art can make palpable what Bernstein calls an “aporia of the sensible,” which makes it a reflection our experience of the crisis of modernity: the rationalizing disenchantment of the world, the scientistic instrumentalist mode of grasping reality, the commodification of experience under the pressures of capitalism, the “all that is solid melts into air” condition.
Here is one of Bernstein’s rehearsals of this story:
If, the argument runs, things only have meaning through what we project onto them, then in themselves things are meaningless and thus ought to be understood in the visionless medium of pure mathematics. The same movement of demythologization that fashioned the death of God is carried forward by a rationalism that limits meaning and value to the satisfaction of human desires and interests as processed through a practical reasoning that is instrumental, means-ends rational, through and through. All else is mythology and illusion.
Here is another:
What the instrumental rationality exemplified by natural science begets, and what is socially borne into everyday life by industrialization and technology (for Dewey), and by the ever-expanding domination of exchange value over use value, the ever-expanding commodity form (for Adorno), is the disenchantment of the world, the creation of an unnatural wound, a diremption, between human nature and nature. This wound is unnatural, or contrary to nature, because the human animal is part of the natural world. In raising ourselves above it — in (cognitively) making the world an object of representational knowing, and (practically) making exchange value the measure of all worth — all subjective response to the world, and thus the world as it gains its constitutive sense in its appearing to human subjects, is qualified, curtailed, elided to the point of disappearance, to the point where worldly things become mere fungible props for an allegorical system whose truth is number and quantity. The cultural crisis generated by science, technology, and capital is a crisis of subjectivity and meaning.
Art is recruited to stand against this process of reification (or perhaps to serve a respite that makes it tolerable).
Modernity pushes art to become modernist, with artists compelled to find ever new ways of claiming grounding while evading reification, mainly by ostensibly purifying the mediums they worked in. Only in this way could their work “compel conviction,” as Michael Fried would put it. Audiences would both see the work as art and see the world as not fully predetermined, predigested, always already formulated into stock ideas. “Modernist works are sites of transcendental claiming,” Bernstein writes, paraphrasing Stanley Cavell. “They are material formations of mastered subjectivity that claim human activity can be meaningful, that we can intend our lives, that even in the midst of indifferent nature and determined society meaning and action are possible.” An artwork stands as significance, itself, in the abstract, as opposed to some particular significant object with a particular purpose.
Hence art from this perspective basically works the antithesis of capitalist consumer culture — it is what resists paraphrase, resists being packaged, resists circulation, resists consumption. Pop culture, by contrast, exemplifies “unmastered subjectivity” and elicits “meaningless” reflex behavior from audiences content to be driven and determined by their appetites when they are not altogether passive in the face of the heteronomy of “determined society.” This is the sort of argumentative line that makes Adorno unpopular: Consuming pop culture is a “regression” into infantility, if not a “liquidation of the individual,” and so forth.
But the point of my detour through modernist art criticism here isn’t to castigate pop culture or to convince you to go spend time in a room in the sanctifying presence of an Anthony Caro sculpture. I’m trying instead to establish stakes for the disjuncture between works and their descriptions, and for the ongoing efforts to force some computational identity between them. When Bernstein describes modernity’s processes of abstraction as ““the indefinite recruitment of ever more domains into the grasp of an indifferent system of commensuration,” something that “reaches down into everyday life and tendentially robs it of subjective qualification,” he is basically describing what we experience primarily as digitization and datafication.
It’s not hard to see AI models as one of the most elaborate material expressions yet of the “indifferent system of commensuration” that underlies capitalism. AI models are touted as though the “aporia of the sensible” (and the “crisis of modernity” it betokens) will eventually be computed away as long as continue to feed them data, as though the idea of “data” itself and what it can’t capture weren’t the essence of the problem. (“The solution to reification is more reification!”)
In a recent piece for Pitchfork, Jayson Greene writes about feeding music criticism into Google’s text-to-sound model to see how well what it produced matched the material the words originally meant to describe. After some experiments, he decides that the model comes “close enough in its aura to give pause” but also recognizes that the whole process is paradoxical, if not logically impossible. In moving from words to sounds, he writes,
there is a sensory transposition occurring — from eyes to ears, or vice versa — with the opportunity for all kinds of vital data to drop off or get lost in the process. Words are already leaky containers, spilling out context whenever they’re jostled. How could this technology even happen?
The simplest answer is that it can’t. It is based on a false and falsifying equivalence between incommensurate things, and it deploys technology to force quality into appearing as quantity, and resolves any subjective perspectives into a pseudo-objectivity. (AI models are machines for eradicating subjectivity.) Asking words to be “containers” of sensible meaning reduces language, the sensible world, and our subjective experience of them both. AI models manufacture a spurious identity among essential differences; they impose a compulsory unity on the world’s irreducible heterogeneity as if all the world can be made to speak in the same voice, in mathematics, in data, and it is all to be made thereby subservient to our demands, to our unchecked instrumentalism.
One could rephrase Greene’s question as “Who is trying to make this technology happen and why?” Who wants us to believe that we should delegate to statistical models the capability to dictate what sounds are “really” convey by certain descriptions? Who wants a post-critical world where works can be fully described algorithmically and functionally deployed to accomplish various ends?
This can be asked of every “AI” application: chatbots, text-to-image models, emotion-detection technology, brain-wave translators, and on and on. On what terms can these projects even be understood to “work” — in what sort of world are emotions, aesthetic experiences, thoughts, etc., directly translatable into specific data — and who wants us to believe in them? Who is demanding that we live in that world? For whom is that world appealing?
When a model makes something that we experience as “close enough in its aura,” it hasn’t approximated that forced harmonization of word and sound; it’s not in any sense “accurate” or “true.” There is no ground truth to refer to: “arpeggiated synths and light-up dancefloor grooves” doesn’t have a single essential referent in the ideal realm of forms. Instead that “close enough” measures how effectively the technology (and more precisely the hype around it, the insistence of its deployment and the leverage of its deployers) has managed to coerce us into hearing those harmonies between word and sound, into obediently recouping the deficit between what technology promises and what it achieves, so that the tech’s actual achievement is transforming us all into its apologists.
But isn’t that a reassuring role to assume? Don’t we want “close enough” rather than “not even close”? Wouldn’t it be nice to believe that a technology could fix the world’s meaning in specific representational formulas so that it made universal sense for us? Don’t we want the “crisis of modernity” resolved? Wouldn’t it be easier to get along if we could all compute correct taste? Wouldn’t it be nice to believe that there are ultimate answers? (This helps explain the tendency among some to worship technology and to exhibit a religious faith in it.) Wouldn’t it be great to be liberated from the burden and isolation of subjectivity, from the alienation and confusion of having a limited standpoint on the world and oneself that places them forever beyond one’s complete comprehension?
Andrea Agostinelli, one of the computer scientists involved with the music generation project, tells Greene that after many years of searching, “the first actual great results are starting now.” But when Greene sensibly asks him and his colleague Chris Donahue how they know what “great” is, they seem perplexed about the nature of their own project:
They spoke, confusingly, of using “proxy measures of audio quality” that were “quantitative and easy to compute.” When I press them on what, exactly, they mean by this, Donahue admits, “You just had to listen.”
Whether or not there can be defensible “proxy measures of audio quality” is the crux of what their research is about. If there are proxy measures, then the specific qualities of a specific piece of music are ultimately translatable into other terms and are thus basically irrelevant to assessing true quality.
Agostinelli admits that music is (in Greene’s paraphrase) “dense in information,” but not that any aspect of it exceeds information; it just has taken the researchers longer to analyze and quantify it all. If whatever they mean by quality can be deduced from something other than the work in its full sensuous, social, and conceptual totality, then the work itself is just an arbitrary assemblage of those underlying determinations. You can use the same raw material to produce something equally worthy in an infinite number of quality configurations — which is, of course, the ideology of generative models. “You just have to listen” is basically a total negation of their work, because what their work is trying to show is that you don’t, in fact, have to listen; you just need to make the calculations. Audio quality is just some number.
***
Even though AI models are often implicitly pitched as solving the problems of language and representation — with data and statistics we can generate a picture of the socially average understanding of any proposition — it may be that their contrived identities prove unpalatable, incapable of “compelling conviction” at any epistemological level that matters. Their proclivity for producing kitsch, which Max Read discusses here, is one index of that. Clement Greenberg’s description of kitsch — “synthetic art” — just is a description of the logic by which generative AI operates, as many by now must have pointed out:
Kitsch, using for raw material the debased and academicized simulacra of genuine culture, welcomes and cultivates this insensibility [to the values of genuine culture]. It is the source of its profits. Kitsch is mechanical and operates by formulas. Kitsch is vicarious experience and faked sensations. Kitsch changes according to style, but remains always the same. Kitsch is the epitome of all that is spurious in the life of our times. Kitsch pretends to demand nothing of its customers except their money — not even their time.
Generative tools (like social media tools) will allow lots of kitsch makers to flourish and achieve mass audiences with their gimmicks. But at the same time, they also popularize the limitations intrinsic to AI’s procedures for representing the world. All the shortcomings of AI culture — the “hallucinations” and the stereotypes and the mediocrity — and the hustle-bro and hack-consultant milieu it attracts, makes it plain that it is not resolving but expressing the crisis of modernity at a higher level of intensity. But at the same time, it can’t complete the mission of fully reifying the world, or convincing people once and for all that everything can be fully rationalized without remainder.
Greenberg polarizes avant-garde and kitsch, but they may also be related dialectically, such that the development of kitsch also pushes the development of an avant-garde, sometimes through an appropriation of kitsch’s means and methods. Generative models, like criticism, can be turned to the project of revealing their own shortcomings in order to mark out the aporia of the sensible. The models can succeed by failing too — not by insisting on an identity between words and things that people are compelled to passively accept, but by strenuously revealing the irreducible nonidentity between them, and disenchanting the disenchantment, marking out negatively what persistently escapes datafication, all that still requires a specific subject in a specific situation to be experienced, realized.
Rather than view transformer models as generating a convincing harmony between text and sound, or text and image, or whatever — one that compels acknowledgement, one to which we surrender, whether out of exhaustion or expediency or social inertia or coercion or some other reason — we might see them as revealing the resiliency of the gap between words and things, the stubborn resistance of incommensurateness itself. After all, the models are not absolute; they can always be trained and retrained on more and more data. We can always be dissatisfied with the identities they propose and demand more, pulling the handle of the slot machine over and over again. In their being “close enough in their aura,” we can retain the feeling of their shortcomings, and the shortcomings of conceptuality in general to capture the nonconceptual residue of all experience. In evoking an aura, or a vibe, AI models may suggest to us that there is nothing but vibes and no conclusive matches.
Just as the crisis of modernity compelled modernism as a response, current sociocultural conditions can be seen as compelling desperate and futile (fatal?) strategies like the development of generative AI — a belief that systematizing and datafying all of past creation is somehow making it available instead of neutralizing it. The mounting chaos of the world that has ensued from the human instrumentalization of it can seem to demand that the project be carried all the way through. Datafication, even as it fails, might seem to reassure is that the past was made for and by humans; that the world is not actually and insurmountably indifferent to us. Perhaps the pursuit of AI will help clarify its insanity; perhaps in the fear of an evil AI that hates humanity, we can come to recognize ourselves and our species’ long and mounting track record of self-destructiveness.
On this view, AI models are the latest and most desperate attempt to heal the division between ourselves and the world and our capacity to dictate or even believe in some sort of definitive truth — a brute force attempt at calculating the name of god. Their continual failure lets us believe there might even be one.