Hand-wringing about AI, Part I: “I Want to Read it All”
Many many moons ago, in a dark bar on a wintry Rochester night, I sketched out a series of eight posts/topics that would roughly correspond with my plan of reading all of In Search of Lost Time (in the semi-recent Penguin set with each of the seven volumes translated by a different translator), and would investigate hand-wringing topics that plague discussions of publishing and reading. Things like, “the kids are all on TikTok, no one reads!,” or “Amazon is ruining book culture,” or “AI is going to be the death of literature.” Statements that definitely have some grain of truth to them, but ones in which—in my opinion—we can’t really discuss in nuanced, informed, philosophical ways simply because there are far fewer cultural spaces where those sort of discussions can take place than there were in the past (and the ones that exist seem to have shrinking audience engagement) and because the people who have the most experience and insight into this issues are probably severely overworked. (Another statement that feels true, but which I can’t prove: “we’re all so much busier at work since the pandemic.”) All of which, in one way or another, fall under the rubric of “the impact of decision fatigue on the business of culture.”
I thought it would be fun to try and unpack some of these ideas against Proust’s very slow, very over-stuffed masterpiece, since most of these are very contemporary concerns that revolve around the explosion of information and ways of accessing it, whereas Marcel spends so much time idling, chasing chicks, attending parties, explaining the intricacies of ballroom and salon life by namedropping hundreds of ducs, duchesses, marquises, and the like, with the Dreyfus Affair being one of the few topical topics presented in the book of which most (some? a few?) of today’s readers will have previous knowledge.
[There’s a reason that, for volume four alone, there are 402 footnotes spread across 30 pages. That’s one of the seven volumes. One.]
I wrote an intro post alluding to this scheme, then one when I finished Swann’s Way, then . . . disappeared for most of 2024.
That was bound to happen—I’m not good at carrying through on my schemes—but there’s also a larger reason, which is where I want to start this, the second official entry in my scheme (with parts three and four coming later this week). Namely, I want to be a polyamorous reader.
I mean, it always sounds good at the start of year to commit to some big, overarching goal. (Hence, New Year’s resolutions.) It’s exciting to think that every day I’d have 10, um, pages of Proust penetrating my mind until I had consumed him whole. But after 60, 70, 200 days of mostly the same things—party, girls, jealousy, striations of high society, it all gets a little . . . vanilla. I need some variety. I’m not a one-author sort of man!
Given that 99% of my reading time is dedicated to editing books for Open Letter (Melvill, Of Beasts and Fowls, The Fake Muse, River) or Dalkey Archive (Midnight Is Not in Everyone’s Reach, A Girl is Lost in Her Century, Looking For Her Father) and cranking out OCR conversions so that as many Dalkey titles as possible are available to interested readers (for a bit more detail about this, backlists, POD, etc., see forthcoming Mining the Dalkey Archive post), I have very little time for “fun” reading. And there are new books by Kate Atkinson, Rachel Kushner, Michel Houellebecq, David Peace, Virginie Despentes, Mark Haber, Richard Powers, László Krasnahorkai, Sarah Moss.[1] That doesn’t even include other books, like Megan McDowell’s retranslation of The Obscene Bird of Night (dying to get to), the Jean-Patrick Manchette books I’ve been carrying around for years, or the Slow Horses books that I’m addicted to thanks to the all-too brilliant Apple+ TV App Streaming TV show. And there are thousands more backlist titles I’d like to discover or reread—including the hundreds of Dalkey titles that I want to either a) make available available again for readers like you, or b) that I need to reread for my longer book project. (I’m not giving up on this project. Trust me?)
In short: Proust is great, but I need to have some other connections. I’m sorry, Marcel, but we’re gonna need to open this shit up. It’s not you . . . well, it’s sort of you . . . but really, it’s just that there are so so many attractive options out there, all available to me, all offering up varying levels of joy and satisfaction. And what would life be like if you didn’t explore your larger interests? I’m not leaving you, Marcel, I’m not. I just don’t need your 10 pages a day . . . We can extend this relationship . . . cool things off a bit . . . deescalate, as the kids say . . . I mean, I only have 1,301 pages left. We’ll get there. It’ll be OK.
*
Before we get to the personal, let’s start at the practical. At the highest level of publishing as business. Publishing as activity to produce product that produces money. A level that makes me uncomfortable—where not everybody on staff could possibly read all their own books—but one that AI, our hand-wringing meta-topic for the next three posts could play a significant role. So that’s where we’ll start: how AI (ChatGPT being the tool most people are familiar with, and is what I used while writing this) could impact larger publishing issues of strategy and production.
Totally oversimplifying here, but: All budgets consist of fixed costs (salaries, operational expenses like rent and supplies, things that are constant regardless of how many books are produced), and variable costs (advances, individual book marketing expenses, printing costs, etc., things that are contingent on the details of the individual products and how many of them are produced), which are offset by revenues generated by sales (frontlist and backlist, subrights, merchandise [haha]) and, in the case of nonprofits, donations. (If you’re a press like Biblioasis or Deep Vellum with a indie bookstore associated with the press, that could also be a revenue stream. And you should all get to Windsor, Ontario. Biblioasis’s store is a dream.) Which, from a capitalist viewpoint, flows into the idea that the more profitable products you can produce while keeping fixed costs low will lead to long-term financial success. Profits. Bonuses. Private jets.
For example, let’s lowball it all and say a press has five employees and fixed costs of $400,000/year ($320,000 in salaries and benefits, $50,000 in rent and utilities, $30,000 in other business expenses). And let’s just use $20,000 as a starting point for the cost of doing an individual book—in a traditional fashion with advances and fees paid to the author and translator (when applicable)—costs $20,000 to produce ($10,000 in artist expenses, $8,000 in printing, $2,000 in marketing).
If you do one single book a year, your overall expenses are $420,000 and you would have to sell approximately 60,000 units of a book with a list price of $20 to simply break even—which would necessitate and additional $140,000 in printing bills (on top of the initial $8,000), and at the end of which, you would owe around $90,000 in royalties . . .
[Just to clarify, of that $20 list price, $10 goes as a discount to retailers, and ~30% of net receipts to your distributor, so you’d earn about $7/unit. And that doesn’t include extraneous fees, which tend to be legion if you’re with any of the main distributors—aka Ingram Publishing Services—available to indie presses. If you’re able to do your own distribution, this whole model would change, adding on astronomically high warehouse and technology costs, many new employees, etc., but you wouldn’t have to pay yourself for doing the distribution, so your net income per book should be higher.]
From here, things get scaled: Publish 10 books and your overall budget needs to be $620,000 and you need to sell 88,571 total units or 8,857 per title just to break even. Move it to 50 titles a year and, well, your employees will hate you (remember, we set the number of staff at five for this exercise) and you’ll face issues of differentiation and sales velocity (topics for later), but the overall budget is $1,420,000, requiring 202,857 total net sales, or 4,057 per title. Again, this is only to offset expenses—not even turn a profit! We’re starting to get into a space where you can squint and see a pathway to the black . . . (Although, again, the printing cost of $8,000 does not accommodate this volume of sales. I’m working on an Excel model of all this mostly to see if I can. And because I’m a dork.)
I’m leaving out several things, intentionally. Sales of foreign rights and audiobooks would contribute to the bottom line, sure, but not in a way significant enough to really alter this model. Donations would, but, again, that’s a different post. Backlist sales would though, significantly. As John O’Brien would say, “If we had 500 backlist titles selling an average of 75 copies a year, that would be 37,500 units and a couple hundred thousand in sales and would act like a functioning endowment of sorts.” This is especially fruitful if you already have the stock sitting in a warehouse, or if you just put all your titles up in LightningSource (a print-on-demand service at Ingram) and save on all upfront printing costs for books that will likely sell a consistent 40-100 copies a year.
Plugging this backlist treasure-trove into my model above, if you did 50 books a year, had that five employee set-up, cranked out product, and had a giant backlist (say 1,000 titles) to rely on, you would only need to sell 2,557 units on average of each frontlist title (the new books coming out in a given year, versus backlist, which is everything from the past) over the course of a year.
And this sort of scheming plays a large role in the Mining the Dalkey Archive book I’m slowly working on (forthcoming sometime, maybe, fingers crossed). Publishing runs like a Ponzi scheme, printing Peter to pay Mary to steal from Paul, and John added his own twists to this.
But here and now, we’re about to take two very bleak turns, play a game, and set up the next two pieces of this series (which are shorter! Most of the heavy lifting is here).
*
I promise you that by the end, I’ll reassert the integral importance of editorial vision as it relates to readership, being a global literary citizen, and value, but for today, we’re going Pure Amoral Capitalist.
As anyone in the industry knows, predicting overall book sales is a bit of a fool’s errand. There’s not enough market research to provide useful comps, things breakout for random reasons (remember the year of adult coloring books?), the best titles fail because of a pandemic. But if you do this for long enough—like twenty-five years—you can get pretty good at assessing the floor and ceiling for a given book. “That’ll probably sell 700, but could do 2,000 if things broke the right way, or 300 if they don’t” is a typical statement with insanely broad error bars. Give me enough historical data and time, and I can probably get this within 300 or so units for 90% of books, but the point is: Assuming you can sell, on average, 3,500 copies of a book in one year is a wild swing, and hard to pull off for most presses smaller than New Directions, Graywolf, or New York Review Books.
This used to be possible, but, tipping my pitches here, there are just too many books nowadays. With so many options, so much noise, the average sales per title have been depressed over the past couple decades. A certain percentage of titles take off and sell far more than 3,000, but the rest don’t even sniff that number.
As such, and employing Dark Capitalist Mojo, the most likely route to broad financial success is throwing more and more books into the marketplace while keeping a chokehold on fixed expenses. Something is sure to stick! Turn publishing away from an editorial enterprise into a numbers game. And what if you can do that by keeping down both fixed costs and variable ones? Like, what if you didn’t have to pay a printing bill unless it was tied to actual sales (the print to order strategy mentioned above, but for everything)? And/or cut down on artist expenses?
Print-to-order is a reality to be addressed elsewhere, but limiting payments to artists is where AI comes in.
*
For years now, in my World Literature and Translation class, the first assignment is an evaluation of multiple translations of the same piece of literature. I’ve used multiple versions of The Master and Margarita for years—always fun and, in lieu of this article, it’s interesting that the Pevear and Volokhonsky never fared well, and to be honest, that was also my least favorite of the translations—but this year I got bored and decided to switch things up. And trick my students in the process.
What I did was provide them with six unmarked samples (you can download all six versions if you want to play along at home 1, 2, 3, 4, 5, 6) of the opening pages of Proust’s Swann’s Way. (Now that I’m blowing this game, next time I teach this, I should probably use part of In the Shadow of Young Girls in Flower. Following with my prescribed rubric, the spirit of volume two—especially the Balbec section about all the young girls blossoming around Marcel and the peripatetic nature of his attraction to each—hovers over the rest of this post.) I told them nothing about where these came from—just that they had to pick the one they would publish and write a paper articulating why that version, both in terms of what sets it apart from the others and how those others fall short.
Here’s the opening from each of the six versions:
For a long time, I went to bed early. Sometimes, my candle scarcely out, my eyes would close so quickly that I did not have time to say to myself: “I’m falling asleep.” And, half an hour later, the thought that it was time to try to sleep would wake me; I wanted to put down the book I thought I still had in my hands and blow out my light; I had not ceased while sleeping to form reflections on what I had just read, but these reflections had taken a rather peculiar turn; it seemed to me that I myself was what the book was talking about: a church, a quartet, the rivalry between François I and Charles V. (Version 1)
For a long time, I went to bed early. Sometimes, barely had I blown out my candle, my eyes would close so quickly that I did not have time to say to myself, “I’m falling asleep.” And, half an hour later, the thought that it was time to seek sleep would awaken me; I wanted to put down the volume which I believed was still in my hands and blow out my light; I had not ceased to make reflections in my sleep on what I had just read, but these reflections had taken on a somewhat particular turn; it seemed to me that I myself was what the work spoke of: a church, a quartet, the rivalry of François I and Charles V. (Version 2)
For a long time I went to bed early. Sometimes, when I had put out my candle, my eyes would close so quickly that I had not even time to say “I’m going to sleep.” And half an hour later the thought that it was time to go to sleep would awaken me; I would try to put away the book which, I imagined, was still in my hands, and to blow out the light; I had been thinking all the time, while I was asleep, of what I had just been reading, but my thoughts had run into a channel of their own, until I myself seemed actually to have become the subject of my book: a church, a quartet, the rivalry between François I and Charles V. (Version 3)
For a long time, I went to bed early. Sometimes, as soon as my candle had gone out, my eyes closed so quickly that I didn’t have time to say to myself: “I’m falling asleep.” And, half an hour later, the thought that it was time to go to sleep woke me up; I wanted to put down the volume that I believed I still had in my hands and blow out my light; I had not stopped thinking while sleeping about what I had just read, but these reflections had taken a somewhat particular turn; it seemed to me that I myself was what the work was talking about: a church, a quartet, the rivalry of Francis I and Charles V. (Version 4)
Time was when I always went to bed early. Sometimes, as soon as I snuffed my candle, my eyes would close before I even had time to think, “I’m falling asleep.” And half an hour later, wakened by the idea that it must be time to go to sleep, I would feel the desire to put away my book, which I thought I was still holding, and blow out the light. While I had been sleeping, my mind had gone on thinking over what I had just been reading, although these thoughts had taken an odd turn—I had the impression that I myself had turned into the subject of the book, whether it was a church, a string quartet or the rivalry between François I and Charles V. (Version 5)
For a long time I would go to bed early. Sometimes, the candle barely out, my eyes closed so quickly that I did not have time to tell myself: “I’m falling asleep.” And half an hour later the thought that it was time to look for sleep would awaken me; I would make as if to put away the book which I imagined was still in my hands, and to blow out the light; I had gone on thinking, while I was asleep, about what I had just been reading, but these thoughts had taken a rather peculiar tum; it seemed to me that I myself was the immediate subject of my book: a church, a quartet, the rivalry between Fran9ois I and Charles V. (Version 6)
I’ll give you the results of my mini-survey later, but before anything else is revealed, you should take a minute and pick out a) which one(s) you would read/publish, and b) which one(s) you absolutely would not. Hint: Four are by human translators, two aren’t.
*
This class took place at the height of AI Anxiety within our industry. Petitions were signed. Think pieces written. Lawsuits filed. Nobody wanted to lose their job to a fucking AI—especially AI “trained” on the backs of human labor.
You can probably see where this is going . . . Operating under the core-capitalist principle of making as much money as possible, consequences be damned, a publisher could be pulled to employ AI to create its works. At first blush, I suspect we’d all hesitate at the idea that AI could invent a full book . . . but, really? Have you read John Locke? (No. Not the one you’re thinking of.) There’s so much shit out there in the world that is competently written, but not particularly good. Using various prompts and scripts, reference points, and craft books, I’m willing to bet that we’re not far away from version X.X of ChatGPT being produce something that’s passable. (Not quite there yet.) And the same goes for translation. (We wouldn’t be so anxious if we didn’t think it was at least possible.) The real question is: What makes a piece of writing “passable”? And if it’s “passable,” is that good enough? Would you really know if that totally mid book you stumbled up and decided to read for whatever reason was written by a “real,” living writer.
My argument—which is more of a hot take, a wish-fulfillment fantasy—is that there is something inherently human in human-created art, something we can’t identify but that we’re attracted to. As if our subconscious can feel the uncanny valley and then subtly rejects it. What does it mean to be inhuman though? And what does that look like when there are just words on a page? But you have to wait for Part III to get the full picture of my bold—and way too optimistic—statement about the future of culture.
For now though . . . AI shit passes when it comes to translation. It’s not perfect, it may not even be great, but in a void with no referents, it probably won’t stand out. Poor writing is. Poorly written books are published. AI is “good enough.”
By now, I’m sure you’ve picked out one of the AI versions above—but did you identify both?
*
Remember the 3% Problem? I’ve been distancing myself from it for a bit to focus less on production and more on the books that are published, but for anyone unaware, this “problem”—and the name of this blog—comes from the fact that 3% of the total number of books published in the U.S. are in translation. I can give you a mathematical argument that in 2024 that percentage is actually much lower; I can show you that the average number of translations actually published has been on the rise, with peaks and valleys, over the past couple decades.
The point is: There are so many great books in the world, so many essential voices, and no possible way to publish even a fraction of them given the publishing constraints detailed above. The sales aren’t there to make the money work. Full stop.Beyond that, how many books could all the professional English-language translators in the world produce in a given year? Ignore the low wages and fact these translators would need to sustain themselves, somehow, I’m just thinking of the time. It’s a fascinating thought experiment to try and establish the upper limit of translation possibility. Is it . . . 10,000 books? 25,000? 100,000? Whatever the ceiling might be, it’s for sure only a fraction of a fraction of a fraction of the total works published in a given year—and a microscopic percentage of all that already exists out there.
But, you know, if AI was “good enough” . . . and you invested heavily in production . . . there’s no reason you couldn’t churn out 1,000 translations a year, paying no one except for the original rights holder, a typesetter (soon to be replaced by AI that can take InDesign templates and go) and a proofreader (hahahaha who the fuck cares) and someone to upload the files into LightningSource. These books likely wouldn’t sell very many copies, but if they each sold a minimum of 200? And, say, 100 of them sell 1,000 units, 1o sell 5,000, and 1 sells 25,000? It’s a numbers game, baby, and the more titles you make available, the more likely you are to hit the lotto with one of them.
*
Thinking about publishing 1,000 books a year gives me hives. Hell, more that 24 is a bit of a nightmare for a small press or anyone hoping to read and think about each of the books they publish. I can’t even decide which book to read next for fun, in my spare time. I finished Spook Street by Mick Herron last night as my “book for fun” (currently editing António Lobo Antunes’s Midnight Is Not in Everyone’s Reach, proofing Stanley Elkin’s The Dick Gibson Show, doing a final pass on Javier Serena’s Attila, correcting the OCR of Eshkol Nevo’s Homesick, and reading Layla Martinez’s Woodworm for a different post—all “work” books) and have no fewer than thirty titles in front of me (plus a dozen or more on my Kindle) that I really want to read. But how does one make a choice? You’ll never read everything, so just pick something . . .
But what if I want to watch something instead? I can’t even guesstimate how many movies I’ve missed out on in the past decade, and I have subscriptions to Netflix, Hulu, Max, Disney+, Apple+ TV, Paramount+ (enough with the plusses already!), and probably something I forgot about, like Tubi or Criterion.
Although honestly, I just want to find time to listen to music. I have 1,929 songs in my “2024 Music” playlist to explore. And that’s only releases from this year. (And doesn’t include a single Chappell Roan song or Taylor Swift remix.)
All of that is real, and even if it’s presented in a cheesy way (or corny? what food metaphor is most dismissive these days?), the point is valid: there’s no shortage of content. If I were 30 years younger (or 40??), you could add to my content consumption habits hours of YouTube videos, short-form video content on Instagram and TikTok, and other social media platforms that traffic in information sharing as entertainment and time-wasting—it’s overwhelming. There is simply too much information.
Making a choice in the face of infinite choices requires heuristics to help trim things down to a reasonable number. Although I don’t think you can get to a reasonable number while engaging with the “new.” You can stay in the 1990s music wise and forget all things that came out since. Revisit the list of movies you imprinted on. Or, what I tend to do with books (remember that list of authors above?), is read all the new titles by authors I already know I like. That’s a good bet, and gives my reading life a sort of continuity. Even if it’s not their best work, it’s likely to be comfortable, familiar. (Although, to be honest, I choose the books I read by setting up strange schemes for myself related to ideas I want to write about, such as the NBA and NTA longlists for Translated Literature. Or fucking Proust.)
Translated literature functions as a sort of sub-genre that helps in terms of selection. You could, if you so chose, read a significant percentage of translations published in a given year. But that’s because the new number of translated works of fiction and poetry coming out in a given year is around 600. Now, imagine if there were an additional 10,000 literary translations available every year. How would you parse that? What are you missing out on? How do you find a community when, if everyone you know followed their own impulse, it’s possible that none of you ever read the same book?
Decision-making is a notoriously energy consuming activity. It’s why certain successful businessmen own endless variations of the same suit—they don’t have to waste time deciding what to wear. But seriously, it’s exhausting trying to wade through all the entertainment possibilities available to us. And the available information multiplies all the time. As a result, no matter how much media one consumes, it’s a smaller and smaller percentage of the whole. Which, for some, can be anxiety making. It’s daunting! And, for the average person, leads back to consuming the things you already know, or the things that are incredibly popular (see: nostalgia).
And that’s where we’ll leave things for today. Up next: AI and marketing.
[1] These names will come up again later in an important context.
Leave a Reply