writing is not the point of writing

on bad arguments to good ends

One of the more humiliating things about writing a newsletter where you get to write about whatever you want is that sometimes you run around saying, to yourself and others, “I don’t want to write about this thing, I hate writing about this thing” and yet there you are, writing about it all the time. I hate writing about “AI” but here I am, writing about “AI.” Clearly I don’t hate writing about it, and yet my hatred of writing about it also feels genuine, so.…

In truth though today I’m not trying to talk about “AI.” I would guess most people reading this newsletter are uninterested in reading things “written by AI” and disapprove of “writing with AI” even if their definitions of “writing with AI” and their levels of disapproval vary. While I’m not above playing the hits, I’m just taking it as a given that at least ninety percent of you are on the same page as me here. Instead, I want to talk about a type of “anti-AI” argument I run into often and which I find very frustrating. This argument is generally made up of the following statements:

• The process of writing is part of thinking.

• All of the work that goes into a piece of writing is writing.

• You write to find out what you think.

• Ergo: The process is the point.

Now I would say the first three statements here are completely true about writing, by the way. I agree with them totally. Nothing in this post should be taken as disagreement with those first three statements. However, what this line actually argues is that the practice of writing is personally rewarding and clarifying, not that the writing produced from such a process is worth taking your time to read. This argument implies that for readers, the writing process is also the point. However, to the extent that this argument can be made for readers, the “process” is reading, not your personal writing journey. For readers, reading is the point.

In short, the problem to me with this line of argument is that:

• The process of writing is not the point of writing, because

• The point of writing is to produce a work of writing, and

• The point of a work of writing is to be read.

Not “read by a lot of people,” maybe to be read by just one person, but nonetheless read by somebody who is not yourself. I don’t know why it feels almost taboo to me to say “the point of a work of writing is to be read,” that’s a matter for my non-existent therapist to ponder. Note-taking, diary entries, and student papers, however, are just about the only kinds of writing I can think of that are not meant to be read.¹

And the question for people who want to prioritize process, to me, is why pieces need to be published if the result is not the point of the process. Why not simply write journal entries?

---

However, there’s another reason this line irks me, which is that I’ve found one hallmark of people who become interested in creative uses for AI is an obsession with process. For instance, in the latest issue of the Yale Review, Christopher Sorrentino has an essay that fits into my own biases here almost too perfectly.²

In his piece, he says that LLMs have freed him from, essentially, worrying about the audience. Instead he can engage in a process in which he prompts the LLM over and over, iterating and re-iterating prompts, producing in one case eighty pages of process that he then refined into twenty pages of story.³ Sorrentino places this use of LLMs into a tradition of experimental artwork, invoking the formal restrictions of OULIPO and the different ways in which visual artists have played with their personae as technicians:

Was I abdicating responsibility whenever I allowed GPT to generate the bulk of a draft in response to specific prompts? It’s a fair question, but blurring the boundary between artists’ purely creative work and their more technical role as operators, or even as curators, has a long (and controversial) history, from Marcel Duchamp’s ready-mades to Donald Judd’s fabricated boxes. I’m reminded of Sol LeWitt’s remark about his wall drawings, which were executed by assistants and installers based on written instructions: “The idea becomes the machine that makes the art.”

Sorrentino is correct to say that his use of LLMs fits neatly into certain kinds of experimental art. He’s more convincing when he’s talking about visual art, though, not writing. The thing about OULIPO projects is that they are interesting books in the sense that they exist, but they are rarely (if ever) discussed as books people read and evaluate as opposed to achievements people admire. I’m not presenting this observation as my own judgment of the books in question, because I have not read them, merely as my understanding of how people seem to relate to them. Unlike a sonnet, where the form is used to express something, in many of these formally restricted writing stunts the form is the only thing being expressed, or at any rate the only thing people can register.⁴ There are more analogies available to Sorrentino in the visual arts, in which something like pure form is more accessible.

Restrictive formal challenges are exercises that can be of use to writers, but writers are not the only people who determine what makes a piece of writing, even their own, good or worth reading. I find Sorrentino’s claims that playing around with GPT-4 has freed him from a creative prison neither offensive nor hard to believe, but the piece itself gives every indication that he is going to get stuck there.⁵ Here, for instance, he asks his chatbot to generate a story from the prompt “eight people entered a room where a holiday meal was to be served and they were…shown selecting their seats on the basis of various criteria—this could be almost infinite”:

GPT-4 instantly generated a draft of about four hundred words, establishing the setting (an overheated dining room), the cast (members of a family reuniting for a holiday), and various internecine conflicts and complications. It sounded a bit like a workshop student doing Franzen. Amused, I had it create variations in style and register, then instructed it to progressively pare them down. Once I had about twenty different versions, I selected eight, which I revised line by line, teasing out and amplifying motifs and recurring phrases to give the individual pieces resonance beyond their shared premise. The finished story moves from one version to the next, progressing from most elaborate to simplest: a critical text copiously annotated with mock academic citations, a baroquely figurative and simile-laden vignette, “contemporary literary fiction,” a Hemingwayesque pastiche, unattributed dialogue, a theater director’s notes blocking the scene, anthropological field notes, Beckettian reduction.

I am sure that this was amusing to do. But as I said, this piece fits my own biases a little too well. Generating endless seating arrangements in pastiche voices is the kind of artistic use I’d come up with for LLMs if I were feeling especially mean and unfair. My reaction to reading somebody else present it as an example of creative energy unleashed is something like “doctor, I’m afraid the infection has reached the brain.” Mostly, though, anybody who uses LLMs this way is not going to disagree with the claim that the process is the point, or even that writing is itself thinking; instead they are going to disagree that their of LLMs conflicts with those statements.⁶

Again, you know, the difficult thing about what I’m saying here is that my point of disagreement is not “writing involves a lot of work that is not literally writing” or “writing helps us to refine our thoughts and develop arguments.” I believe there are goods gained through the practice of writing that are not easy to explain. I agree completely with Meghan O’Gieblyn’s comments along those lines in this recent interview, for instance. My difficulty is that the contexts in which these statements can make up a complete argument for eschewing LLMs are limited. They’re true of the classroom, where writing exists in a specific way. They also make good intra-writer pep talks, where we all remind each other of why we do the stuff we do. Everywhere else, they are an explanation of why writing matters to people who do it, and that’s precisely as compelling as why any pasttime matters to people who do it.

There is something missing here; a place people don’t want to go, or maybe don’t know they have to, or maybe feels too obvious to state. Or maybe it’s just that writers have forgotten the people they address do not ideally consist of each other. I’m not really sure. Furthermore, I don’t have a particularly great argument myself “against” the kind of writing Sorrentino wants to do—I only have the sense that it will not produce anything I want to read or that I will not experience as an elaborate prank.

I’m not super online right now, but I’ll check in when I can.

1. Student papers a little dicier but You Know What I Mean.

2. I don’t know if it’s a necessary disclosure or just annoying to say I am “working on something” for the Yale Review but in case it’s the former I mention it.

3. You could say that he’s gotten rid of the audience by constructing a situation in which he and the LLM co-create for an audience of one, himself.

4. This is not true of the way in which people talk about Finnegan’s Wake, for instance.

5. Also stuck, judging from her contribution to this issue: Sheila Heti.

6. Maybe related:

ai and the individual talent