As you may have noticed, I typically use a generated image to “illustrate” these posts. Usually I generate them with a random phrase that would seem to have no visual overtones, cut-and-pasted from any of the tabs or documents I have open at the time. (The image above was generated with the prompt “a random phrase that would seem to have no visual overtones.”)
I don’t do this to make some subtle point about generative AI or the conundrums involved in producing images through textual associations; I do it because having an image is part of the boilerplate for a newsletter post. Not having an image makes the post seem incomplete or less promotable. Since I can’t afford professionally made images and don’t have the expertise to choose among them anyway, I generate a placeholder that more or less adequately fills the content hole that the format imposes on me. Having a pertinent image would be a bonus, but it seems far more important to have anything at all there rather than nothing.
It may be that some people see writing in general this way, as nothing but an assembly of placeholders, so many TKs that need to be methodically filled. When everything is finally compiled, it magically becomes more than the sum of its rote parts. (I should have made this a dissertation-writing mantra.) When I was in middle school, I had an English teacher who taught essay writing by passing out a mimeographed Mad Lib for students to fill in with details from whatever short story they had been assigned to analyze. “In AUTHOR TK’s story TITLE TK, the theme of THEME TK is explored. In this essay I will show …” It seemed like a pointless exercise to me at the time, but it was probably meant to help students past the paralyzing fear that writing requires special inspiration or intuitive leaps or imaginative juxtapositions. It doesn’t have to be original to be functional. Whereas I stupidly thought that the teacher actually wanted my profound insights into Ernest Hemingway’s “Big Two-Hearted River.”
As a reaction to that kind of pedagogy, I am susceptible to believing that “good” writing must in some way be dysfunctional to seem truly inspired, and that straightforward presentation of information impedes the reader’s experience of “inspiration” or “artistry.” In a post called “Why AI Will Never Rival Human Creativity,” William Deresiewicz articulates a version of this position, arguing that LLMs can never be “original” and thus can never make art.
AI operates by making high-probability choices: the most likely next word, in the case of written texts. Artists—painters and sculptors, novelists and poets, filmmakers, composers, choreographers—do the opposite. They make low-probability choices. They make choices that are unexpected, strange, that look like mistakes. Sometimes they are mistakes, recognized, in retrospect, as happy accidents. That is what originality is, by definition: a low-probability choice, a choice that has never been made.
I don’t find this convincing as a description of “AI” anymore; you can adjust the probability of a generative model’s output, which often has the feeling of being riven with “happy accidents,” especially when you take into account multiple generations and iterations derived from the same procedures. Chatbots aren’t restricted to generating “high-probability” outputs, but they are restricted to producing outputs probablistically. If you assess work in terms of how original or probable it is, you are generating a kind of AI content in your mind.
Deresiewicz is also attacking a straw man in positing someone who argues in good faith that generative models make art or simulate human creativity. Anyone who claims that is giving themselves away as a scammer or a salesperson. The far more frequent claim is that AI “tools” will “unlock” new echelons of human potential; it will help us fill in the TKs so we can reserve our energy for the imaginative leaps and spontaneous syntheses. We can use prompts and refinements to go on a “voyage of discovery,” as Deresiewicz characterizes true art-making, and let generative models choreograph the “dialogic dance between the artist and the work.”
Stephen Marche, who recently published a novel generated with the aid of LLMs, describes his “dive headfirst into a practical extended application of the new technology” in this Atlantic piece.
My process mostly involved the use of ChatGPT … and Sudowrite, a GPT-based, stochastic writing instrument. I would give ChatGPT instructions such as “Write an article in the style of the Toronto Star containing the following information: Peggy Firmin was a Canadian writer who was murdered on a bridge on the Leslie Street Spit on August 14 with no witnesses.” Then I’d paste the output into Sudowrite, which gives you a series of AI-assisted options to customize text: You can expand, shorten, rephrase, and “customize” a selection. For example, you can tell Sudowrite to “make it more active” or “make it more conversational”
This is very much in the spirit of my middle-school English teacher, with a novel being conceived as a set of genre-dictated TKs, a quota of plot points, descriptive passages, aphoristic sentences, and stylistic homages to other writers. Where Deresiewicz derides the “everything is a remix” approach to writing as a “degraded conception” and a “banality that grew up to become a stupidity,” Marche proclaims that “AI may be an escape from the formulaic exactly because it is derivative art.” He likens writing with AI to hip-hop producers’ manipulating samples, rescuing artistic practice from commercially imposed sameness.
But Marche and Deresiewicz are in agreement about one thing: their own aesthetic acuity. “Unlike the techies and pundits, in their glorious ignorant smugness, I have some sense of what art is and how it is created,” Deresiewicz announces. Likewise, Marche declares, “I know what good writing looks like,” by way of explaining why he could use AI tools to make something “compulsively readable” with AI when an army of tasteless poetasters on Amazon could not. Their confidence on this front suggests that their disagreements about originality and so forth are irrelevant relative to their shared view that AI-generated materials will allow special humans to once again assert their superiority over the smugly ignorant.
In a sense, both are using AI in the way John Herrman describes here:
Whoever or whatever this odd internet stranger didn’t like, AI was coming for it. It’s AI as a reckoning, a punisher, a revealer of frauds. It’s AI as a future vindicator of their hunches about how the world works, and as an extension of their politics. It’s AI as a cleansing force that humbles your enemies and proves you right — AI as economic rapture. It’s AI as your army-in-waiting just over the horizon, your punishing angel, or maybe just as the thing that’s going to embarrass the people who annoy you online. A lot of sunnier AI speculation is clearly wish fulfillment, and so is this. AI is my big, strong friend, and he’s going to beat you up.
In a review of Mrs. Davis, a new TV series about a future world where a chatbot in an earpiece serves as everyone’s superego, Phillip Maciak shows how this bullying plays out when the smug tech pundits go after artists:
One of the most vexing aspects of tech prophecy in the present moment is its contempt for art, whether it’s writing or the kind of image-making that’s supposed to be replaced by tools like DALL-E. Art, in the eyes of lots of these kinds of futurists, is not a mystery, but a problem to be solved. That the creative act is difficult to perform, difficult to explain, difficult even to understand makes it an uncomfortably irreducible remainder to someone who’s trying, Mrs. Davis–style, to optimize the world. And so, of course, AI’s makers come for it first, seeking at long last to reveal the trick at the center of art, the proof that the idea of romantic genius and inspiration is a glib illusion, a grift meant to shame and embarrass everyone who can’t draw or dance or write a good sentence.
But people certainly don’t have to be AI enthusiasts to use the idea of it as a “vindicator of their hunches about how the world works.” Skeptics can use AI as a cudgel as well. If your hunch is that “romantic genius” is real and manifests in gifted individuals, than AI art will reveal that also. If your hunch is that you have an exceptional ability to appreciate real art, the “AI debate” gives you an opportunity to try to demonstrate that. If you think that artistic ability is more of a habitus, a mark of social preference and entitlement, you can use the existence of and resistance to AI to try to make that case too.
When I read essays of writers’ patting themselves on the back for their creativity or artistic insight, I’m tempted to celebrate AI for how it could undermine the mystifications of “creative genius” that are often used to justify various social exclusions and hierarchies. But I also think that the bulk of AI hype has made it clear that, as Maciak argues, the people most invested in claiming art is a grift are grifters with a different kind of scam. They don’t want to do away with art’s elitism; they just want to ground social hierarchies in a different form of domination that can be implemented automatically through a totalizing technology. And perhaps most obviously, they want to devalue artists’ work to undermine their negotiating leverage when it comes to paying them. One way to do that is to redefine writing work as a new kind of job, as a custodial process of sub-editing the machine-generated material produced to fill in the TKs demanded by management.
Art as an “irreducible remainder” — a highly Adorno-esque formulation — sometimes strikes me as self-evidently correct; sometimes it seems dogmatic and unverifiable, a principle taken on faith to sustain belief that something persists beyond the fully administered society, the full reification and exchangeability of everything. There is something incommensurate and “nonidentical,” something that can’t be rationally and mechanically produced on a commodity basis, something autonomous from capitalist culture that we must insist on but can’t definitively identify without destroying it, something which we must and can’t ever call art. When I drop an image into a template, I don’t feel like I am responding to a formal, institutionalized demand to make and distribute more art; I feel like I am being called on to destroy art.
If art is simply defined as irreducible and unprogrammable (which would make “AI-generated art” just a contradiction in terms rather than any kind of threat), then it ceases to serve as any kind of beacon for a particular kind of social practice, whether liberatory or reactionary. What specific kind of social experience resists the attempts to reduce society to automaticity, or does that resistance just flow automatically from sociality? If AI tools allow everyone who can access them to make art-like things, is everyone now an artist or no one?
Sprachkunstwerk
“They don’t want to do away with art’s elitism; they just want to ground social hierarchies in a different form of domination that can be implemented automatically through a totalizing technology.” - This made me reflect on capitalism’s incessant, natural movement towards removing the middle. To be radically simplistic about it - Artistic elites are fine, an army of people pumping out banner ads and producing writing prompts in contained systems are fine. That mushy middle class that includes designers and art directors and writers that expect livable incomes for that “ambiguous” performance of “making art” has, in my experience, always been positioned as a kind of “theft” from the “real” engines of capital - investments, financial “risk” and the production of material goods and extraction of labour value therein. AI is exciting for the owners of capital because of its promise to automate away much of that ambiguity into a more totalizing (if never complete) domination over social relations of which “artists” have always seemed to move through in uncomfortably uncontrollable ways.
---As a reaction to that kind of pedagogy, I am susceptible to believing that “good” writing must in some way be dysfunctional to seem truly inspired, and that straightforward presentation of information impedes the reader’s experience of “inspiration” or “artistry.”---
That's what I appreciate most about your writing. I actually have some offhand remarks (in addition to entire pieces) saved in word doc format that I open up and reread occasionally for pleasure.
We are not aligned about a lot. I truly just appreciate your writing. It's the only Substack I pay for and pay real attention to.