LLMs and the Business of Fiction: Execution (part 3)
A look at the way that writing in different media will be impacted by Large Language Models like ChatGPT.
Today I’m going to take a bit of a deeper look into the way that writing in different media is going to be affected by LLMs like ChatGPT. In this installment I want to look more at execution: how the technology is likely to be used in the the creation of commercial writing.
Some of this is happening already. A lot of this is speculative, but I think if you look at the trends in the publishing business over the last couple of decades (discussed in some detail in part 2 of this series) and put that together with an understanding of some of the capabilities of the technology (covered in part 1), a lot of this looks inevitable.
Film & TV
Let’s start with film and TV, since the Hollywood has currently ground to a halt on the back of the WGA strike, which has been joined by the Screen Actor’s Guild. AI is a key bone of contention.
The studios appear to want to use AI to write scripts and to digitize actors. They have been doing the latter since the 1990s, in low budget TV shows like the Mighty Morphin Power Rangers to ultrabudget fare like Terminator: Salvation with its Fake Arnold Schwarzenegger and the posthumous appearances of Carrie Fisher and Peter Cushing in recent Star Wars. Writing AI scripts is a newer idea for them, but I think has the same motivation. It’s presented as either cost-cutting or an homage to older talent, but I think, if you look at the amount of profit the studios are already making, that is only a small part of the play.
What we have seen, with the recent box office domination by IP franchises, is that scale is everything. The bigger you are, the more difficult it is to compete. It’s becoming more and more difficult to even get your movies screened, if you’re not Disney. The studios want to put out more material, more cheaply.
Corporations want the IP to be the big draw, first and foremost, followed by the acting talent. The film production process is liable to deviate from the script, mediated through the skills of the director, the DOP, the actors, the editors, and I think the studios just don’t see the value in the screenplay, no matter who writes it.
This is a huge frustration to me, as a writer and a viewer. The script is the cheapest part of a movie to develop, and it’s much easier to fix problems there than it is to rewrite, re-edit or re-shoot once production has started. I know that scheduling talent is hard, but it drives me nuts to see a big budget film with a half-baked script—but Hollywood in particular seems fine with it. AI scripts are going to yield a further drop in quality, but with AI screen talent, I imagine the studios think that reshoots will be cheaper and easier.
Can they get by with scripts written by LLMs? Well, kind of. I think it can go in a couple of directions, depending on the studio and the situation.
If you’re any kind of professional writer you’ll know there are thousands of people out there with an idea they’re willing to ‘let you have’ so you can do the hard work of turning it into a story. Well, those people can now get an LLM to do this for them.
Screenplays, I think, are a relatively easy form for AI to churn out. They’re formally structured, they break into well-defined acts, and they’re much shorter than novels. A studio could easily train an LLM with from existing material featuring their characters and world and turn it loose. If you’ve spent any time with ChatGPT or Bard, you’ll probably notice that they already know a a lot about franchise characters.
Likely the studios will need a real writer to fix this output. The job becomes supervising an AI to generate franchise films, without authorship or ownership. Gig economy work.
Another possibility is that studios will use LLMs to homogenize original work. To make everything sound and feel like whatever is currently successful at the box office. Rewrite the dialogue to sound like it was written by William Goldman or Elmore Leonard or Paul Schrader.
I think the market for screen material will stratify. Original movies, created by humans, might become prestige content, while studios also pump out tons of cheaper, automated content just to occupy the space. If you have ever trawled the low-end of a streaming service’s children’s section, you’ll see there is already a lot of this, generated without the help of AI. Right now it’s mostly amateur-looking CG animation with very little dialogue (which requires writers and voice actors). AI might supplant this garbage with… their own garbage, featuring franchises characters and starring fakes of real actors.
I don’t know if this quest for scale is going work, but I’m sure the studios are going to give it a red hot go, should the WGA and SAG strike fail.
Traditional Publishing
Despite the fact the Department of Justice blocked last year’s attempted merger between Penguin-Random House and Simon & Schuster, the traditional publishing business is an increasingly difficult place for authors. The biggest writers are getting bigger, while midlist authors are shriveling on the vine. AI is likely to accelerate this.
The world’s best-selling author is former-advertising-CEO-turned-novelist James Patterson, whose model of success is based on name recognition and volume of titles. Patterson publishes many books a year, these days always co-authored with another writer. Sales of his individual books don’t fly as high as some other marquee authors, but that volume of new works keeps his income steadier and keeps his name prominent on shelves. Patterson, like a tech business, is all about scale.
LLMs are the perfect tool for this business model. Train one to write books that feel like a bankable author and it can scale the process up massively. Rumour has it that publishers are already negotiating contracts with certain writers, although I have not seen any such authors being named yet. In this model, a co-author’s job might be reduced to pushing an LLM to output a story that may (or may not) have been ‘designed’ by the name author, but which sounds like it was written by that person.
The result will be more books from big name authors on the shelves, with less room for everyone else. I don’t know how well this will percolate down to bookstores, especially independent ones, but regrettably, that is not where the battle is being fought. Bookstores are the best places in the world, but it’s only big box retailers like Walmart and Target who sell enough books to challenge Amazon, and they most certainly don’t care who is writing the books, or how.
Now whatever you think of Patterson’s business model, it’s worth noting that he makes sure his co-authors are credited and paid. When an LLM is the co-author, this isn’t going to happen and human co-authors likely back to gig economy work.
Indie Publishing
As a Gen-Xer, I love nothing better than to see an indie creator find success, be it in music or books or film, and the rise of indie publishing as a viable career path for authors is something I have been thrilled to see.
As traditional publishing has become more corporate and risk-averse, it’s been a delight to see writers finding success elsewhere, using tools and markets that traditional publishing has been slow to recognize or adopt. Many of the best books I’ve read in the last 10 years have been published by indies. Some of them have been picked up and re-published by big players. Some of them are written by authors who made their names in traditional publishing and struck out on their own.
There’s plenty of garbage, of course, but if you look today you can see a lot of indie books that are every bit as polished and as professionally produced as anything the Big 5 are putting out. Often these books are produced by the same people—the same cover artists and editors and designers and type-setters who work for the big publishers, turned to freelance or prosecuting a side hustle. Ten years ago I’d never have imagined I’d have a small press title with cover art by someone like Daniele Serra, who has produced covers for Joe Lansdale and Stephen King, but lo: it happened last year like it was no big thing.
In many ways, indie publishing has been a tonic for the cynicism of the big players—not just the independent writers, but the voraciousness of the readers who buy their work. Statistics say that the number of people reading for pleasure is falling, but you wouldn’t know it to look at the reader communities that have grown up around indie work. there are a lot of readers out there and some of them are voracious.
Indie publishing is a high margin, low volume business, compared to traditional publishing. I don’t have hard numbers, but anecdata suggests that the sell-through on book series yields authors the best results. Successful indie publishers release a lot of books, fast, in order to take advantage of this. They can move more nimbly without having to negotiate traditional publishing schedules, but they must also be prolific. Many of them have already turning to LLMs to boost productivity, using it to help craft blurbs, press releases and ad copy. (As I mentioned, I think it’s unlikely they are getting any worthwhile prose or dialogue out of an LLM, or at least not without substantial editing.)
But the other thing with indie publishing is that there are no rules. In addition to all of the amazing indie work I mentioned, and the mountains of poorly written, unedited garbage, there are also hordes of scammers—and already we are seeing this latter group use AI to work.
Kindle Unlimited (KU) is an Amazon program that offers subscribers and all-you-can-read experience for a small monthly fee. Authors must sell their book exclusively on Amazon to be a part of this program, in which they get paid from a shared pool of money an amount proportionate to the number of pages of their work read. This situation practically invites abuse, and already we have grifters generating garbage AI books and subscribing bots to read them. Here’s how it works:
Publishing a book on KU is free. Each bot reader account has a fixed entry cost, but can read unlimited amounts of material. The more books a grifter can publish, the more money they can take from the pool for lower cost (i.e. using fewer bots). Amazon has little incentive to stop this, because the pool of money is fixed—so it doesn’t cost Amazon anything. It just costs the other KU authors their fair share.
Comics
Comics is an interesting case, because Anglophone comics has for decades been mired in the kind of IP franchise world we are seeing now in other media. While Marvel and DC superheroes are still the mainstay of specialty comic store’s Direct Market, books like The Walking Dead and Saga have shown than strong genre work can outsell any of the corporate titles. And now we have a whole parallel market in which original graphic novels are sold through traditional book channels. There is increasingly some crossover between markets, but they are still largely selling quite different fare.
Comics are ahead of the trend we’re seeing in other media. With comics possibly maybe starting to recover from the entrenchment in the IP world, I think the impact of AI is going to be muted when it comes to comics writing—although there is a very different discussion when it comes to comics art.
Firstly, writing comics is architectural in a way that prose and writing for the stage or screen is not. In addition to wrangling plot, character, and dialogue, a comics writer needs to managed the flow of action and dialogue between panels and across pages, and be able to communicate this to an artist. This is a difficult overlay skillset to master and and I think it’s going to be a while before LLMs, which cannot count or reason, to learn visual storytelling.
Skilled artists might of course be able to solve this problem on the page, but I think artists are in greater danger of being automated out of the picture than writers, just from the point of view of cost. (I will cover AI art in a different post, so let’s leave that one there).
If we look at the bifurcated market, I can see the operators of franchise superhero and genre comics wanting to enlist language models to assist with plotting or dialogue, but I think this is less of a threat than it might be in screenwriting. With titles like Saga on the up, and the obvious sales trajectory of IP books written by A-list writers, I don’t think LLM-authored books is a great proposition for a business as precarious as Direct Market comics. There just isn’t the room to flood this sector with cheap AI filler. (Boy have we learned that the hard way.)
Works sold through the bookstore market are much less likely to be owned by some kind of an IP franchise. They are also more likely to be created by writer-artists who may not engage in a formal scripting process. I can’t see that anybody wants knockoff Raina Telgemeier books and I don’t see LLMs getting much, if any, adoption in this area.
LLMs will chiefly be useful to comics writers for helping with the scut-work: massaging pitches, summaries, synopses, blurbs and other necessary work that is not directly part of the writing process.
Theater
Theater is all about seeing live actors on a real stage. While we have seen some holographic performers recently (notably Ronnie James Dio’s posthumous tour as a hologram), audiences go to the theater to see live actors. Such a production with can only play on one stage at a time—there’s no scale-up to replicate the experience. Given that, I think it’s unlikely there’s going to be a need for AI-generated scripts to feed a horde of theater companies starved for material.
Playwrights might benefit from some of the utilities offered by LLMs, but I don’t see their livelihoods being taken.
In summary, different media are going to be affected differently. I don’t believe that LLMs can entirely replace human writers, but they will be used to scale up the amount of material generated in service of IP franchises or marquee authors, with real writers relegated to grooming their output. This will likely lead to a stratification of work in which more expensive, better quality ‘premium’ work is produced by humans and sold at higher prices, while greater volumes of lower quality work is used to try to drown out competition.
In a future post I’ll reexamine this logic and see if it holds up. I think if you look at last weekend’s Barbie/Oppenheimer showdown you can see where it all starts to come apart.
LLMs can still be productivity aids for working writers, and I expect this will become more widespread, especially once LLMs that are trained responsibly become available—also a topic for a future post.
Grifters will will continue to leverage LLMs to cheat the system. Even if Kindle Unlimited starts to police bots you can be sure that someone will find new scams: fake readers, fake reviews, fake books… these are just the beginning. But with that said, I believe indie publishing will continue to thrive.
Humans want good writing.
That’s it for this week. For those following the drama, I think that Bloody Waters is now properly available from Amazon, but honestly, who can say?
I have some more book news for next time, and of course I will continue to discuss generative AI, with at least one more post on LLMs to come.
Frankly yours,
— Jason