LLMs and the Business of Fiction: Trends (part 2)
A look into recent trends in the business of fiction and the ways that Large Language Models like ChatGPT will impact it.
Today I’m going to continue my discussion at the ways that I believe generative AI technologies like ChatGPT will affect the business of writing fiction by looking at some recent trends in media that have lead us to this point.
The ground is changing very fast as the technology evolves, vendors monetize, users react, companies adopt and governments try to respond. I can’t promise that what I write today will be accurate next week, but I’m doing my best to keep up with it. In the last fortnight I have spoken on this topic on a panel for the Horror Writers Association, a focus group with the Australian Society of Authors, and a presentation to the Records and Information Management Practitioners Alliance symposium. It’s great to see different groups engaging with the technology, but we are just at the beginning of understanding what it means, much less being able to forge a coherent and practical response.
If you missed the first post in this series, which talks about the technical aspects of Large Language Models (LLMs), their capabilities, and the ethical problems they present, I recommend you start there. You might also take a look at this earlier post, which is a primer on generative AI.
Alright. Let’s look at the trajectory that the business of writing has been on leading up to the so-called AI revolution. I’ll start with the big question:
Will LLMs replace writers?
No.
LLMs are not creative. They have no initiative or intention or experience. They cannot be relied upon for cogent or truthful information retrieval. They are very good at manufacturing convincing-sounding verbiage from just a few words of prompt, but they cannot create interesting or original stories. All they can do is regurgitate old ideas.
LLMs cannot replace human writers, but they can change the way that commercial writing is done. They can be useful productivity aids, but they can also harm our livelihoods. Both of these things are already true, but I think we need a more detailed look at the way the writing business has been evolving over the last couple of decades to get a better picture what this means.
Trends in the writing business
Let’s step back from fiction writing for a minute. About one billion years ago, when software came on CD-ROMs with paper manuals, I worked with a technical writer named Bob (actual name used). Bob was so good that customers would write to us enthusing about how good the doco was, which is like me saying that I once found a unicorn shoe in my lunch box, except for the fact it actually happened. Have a guess what happened to Bob.
Bob’s contract was not renewed because, management said, he was too expensive, and because nobody buys software for the manuals. Which is true, on one level. On another, good documentation reduces other costs to the business in terms of support tickets and thus customer churn. But these savings are difficult to measure against the bottom line. a billion years later, if you want a manual you have to buy one written for profit by a third party.
What happened to Bob is not unusual in writing work. We have also seen journalism jobs eroded in this way, even before the advent of AI. When my wife, a former newspaper journalist and magazine editor, moved to Australia, she took some freelance work micro-blogging for some travel companies, for the heartbreaking sum of $5 per article. This work is was pure SEO bait, written for the purpose of encouraging search engines to serve the company’s website, not for informing humans about hotels and travel destinations. With the advent of AI, it’s not just expensive guys like Bob who have lost out. These small gigs, already well below minimum wage, have already been obliterated. I am told that legal departments are next in line.
Trust in news media has plummeted as writers has been devalued and de-professionalized into the gig economy. It’s not easy to sort cause from effect, but factors include the disruption of legacy media by the internet, increasing political polarization, and the consolidation of media companies. Many of these companies own the digital delivery systems as well as the content, which allows them to measure all customer behaviour as well as the content itself.
These trends have set current trajectory for AI.
Consolidation of Media Companies
Publishing has seen the same kind of consolidation as news media.
In 2013, the merger of Penguin and Random House (PRH) saw the Big 6 publishers became the big 5. This would have become the big 4 last year, if not for an antitrust action last year by the US Department of Justice. Book retailing (well, all retailing) is increasingly dominated by online giant Amazon, and smaller booksellers on the back foot. (Amazon CEO Jeff Bezos is reputed to have likened the publishing industry as “a sickly gazelle” and Amazon to a cheetah). Only Barnes and Noble, another very large book chain, has recently managed to turn around its fortunes.
In the film and TV business, we have Warner Brother/Discovery/HBO and DC Comics all under the same corporate umbrella, while Disney now owns Pixar, Marvel, Star Wars, 21st Century Fox, and Discovery. You see the problem. The consolidated media companies have their own delivery systems (Amazon Kindle, Amazon Prime. Disney+, Marvel Unlimited, etc) and they have a vested interest in promoting their own IP over other work.
Wall Street is now in the thrall of Silicon Valley, and wants to run these media businesses like technology companies, which are all about scaling output. Studios like HBO, that turned a comfortable profit and were remarkable for producing top quality storytelling, are now expected to yield hockey-stick growth. When HBO’s name has been removed from the branding of its streaming service, I think it’s pretty clear that the first casualty of all of this is going to be quality.
Metrication of Consumer Data
The second trend here arises from the way that so much of storytelling is served online. Streaming movies and TV, video games, ebooks, digital comics.
Online delivery is convenient and saves on the mounting production and logistical costs of physical media. It also leaves an enormous data trail about the way that we consume media. Which movies we watch, and for how long. Which pages of which books we read. Which ads we click. It also leaves the media itself readily available to be analyzed, byte-by-byte. This data is a priceless commodity which can be used to predict what consumers want, and in what quantity.
Business analytics isn’t a new idea, but the volume and veracity of this new data, and the velocity with which it can be processed, is now unparalleled. Platforms like Amazon use these metrics it to drive us towards product that we are likely to spend money on—which in turn reduces visibility of new products that do not have the same historical data to give them an algorithmic boost. The only way to get that boost is to buy it, but it is very difficult for small players to compete with these massive conglomerates who may be receiving their marketing spend directly. It’s not a level playing field when your competitor owns the field and the ball and most of the opposing teams.
This is just old fashioned statistical bias, which will always privilege items that have more data points. “Going viral” is a prize because it means that an individual has beaten the statistical odds due to genius—or more likely, good timing and happenstance. But even this booby-prize is increasingly difficult to achieve as the algorithms tune smaller players out.
The Rise and Rise of IP Franchises
The third trend, is a natural extension of this bias: the continued growth of IP franchise media at the expense of the new. Star Wars, Batman, Harry Potter, Barbie. These all come with a quantifiably huge audience, which makes them less risky to invest in and more likely to achieve crazy profits. The algorithms reinforce themselves and anything else gets pushed out.
We have seen the same in the Anglophone comics industry as well, to the point where the entire distribution system was built around pushing Marvel and DC superhero books to specialty stores whose mainstay was selling superhero comics.
In this constrictive environment, traditional book publishing has also become increasingly risk averse. Publishers find it easier to sell books that resemble other books that are doing well, and they buy accordingly. They allocate marketing budget to books on their slate according to the sales history of their authors and other visible trends in the industry. When marketing spend is the only reliable way to predict a book’s success it becomes a self-fulfilling prophecy.
If this seems contradictory, it is. During the Simon & Schuster court proceedings last year, Penguin-Random House chief executive Markus Dohle admitted to using data scientists to try to predict trends, and while also claiming that success is entirely random.
Enter AI
And so we come to AI.
Imagine that you are a big company with the infrastructure to deliver content, comprehensive and granular data about what consumers want, and a slate of your own IP that you want to sell them. Like all big companies, you have two goals: maximize profit and minimize cost. AI looks like the perfect tool to help you figure out what to sell, maximizing profit, and if it can generate content then it can help you to minimize costs, too. Costs being the labour of human artists.
You can see the WGA is on strike about this.
The problem with this plan, of course, is quality. LLMs don’t generate good quality fiction. They will assuredly get better at what they do, but I don’t think they can entirely supplant the need for human writers. Even if the quality of text they generate becomes competitive with that of a human writer, they have no initiative. The ingredients for the stories, characters or tropes might be determined by data analytics, but someone needs to put the ideas together. Someone needs to guide them through the process of churning out books or screenplays, someone needs to edit this work for consistency. There needs to be a ‘human in the loop’, as the saying goes. That human is us—writers.
As we have seen with copywriting and news media, this is likely to drive job losses and falling wages. This kind of writing might become gig economy work and the already diminishing number of full time professional writers might collapse entirely.
But I don’t think it will.
What Happens Next?
Let me wave my unicorn shoe over my crystal ball.
These trends have pushed the writing business in a direction that is starting to feel pretty bleak, but I don’t think this will continue forever. There are some good indications that we are not bound for a future of art being drowned in a sea of AI-generated pablum.
To begin with, the US Department of Justice blocked the PRH/Simon & Schuster merger. While this still leaves us in a powerfully anti-competitive marketplace, it’s an indication that institutions are starting to recognize the issue. The fallout has been damning and at many of the industry’s rumoured bad practices were confirmed as true in the court hearings. Hopefully we will see further action on this front.
The big publishing houses have also, meanwhile, discovered that there is an audience for new and diverse books by authors from minority and marginalized backgrounds. The work is staggeringly good and has been selling accordingly. That is going to continue. This flies in the face of the idea of homogenized, algorithm-generated content. I think it is becoming obvious, even to the likes of Markus Dohle, that you can’t keep selling increasingly watered-down versions the same thing and expect it will keep doing numbers.
Fiction writers are not like Bob, the technical writer. Our names carry weight. Our backlists carry weight. Readers like knowing about us and will follow our work.
The comics business is, for once, a bit ahead of the curve, having been diminished repeatedly through decades of boom and bust cycles. The rise of the comicbook movie has not led to markedly better sales of the comicbook source material and times have been hard. Twenty years ago, the two big superhero publishers dominated the only distributor existed primarily to service them, but that hegemony is crumbling now.
Meanwhile, at the Hall of Justice, independent titles like the Walking Dead and Saga have shown that they can and will outsell the biggest corporate properties. Not only have the three biggest publishers have moved away from the Diamond distribution monopoly over the last year, but now traditional bookstore channels have built a parallel marketplace for graphic novels that are not corporate properties. The bookstore market is now, dollar-for-dollar, bigger than comics’ traditional specialty market.
In the world of film and television, small studios like A24 have managed to flourish with lower budget, profitable, critically acclaimed hit movies, and they have managed to secure approval to keep 24 productions filming through the writer’s strike by dint of not being assholes. Actors like Mark Ruffalo (the Incredible Hulk) are calling for actors to do more indie work and I hope they will.
Anything can still happen still.
It’s naive to believe that LLMs will go away, or that media companies—and individual writers—will not use them. We need to accept it and concentrate on how we respond.
I believe we can address the ethical issues in how they are trained and monetized. Legislation and legal action are already underway in different territories. Writers can use the tools can be used as productivity aids, without resorting to plagiarism. Today’s writers are drowning in an administrative burden, since we have to to spend additional, unpaid effort on promotion and other supporting activities that diminished publishers used to do for us. LLMs can help with a lot of that scut-work, assisting with writing the synopses and summaries and blurbs and drafting business emails.
Anything can happen still, but we need to make ourselves heard if we’re going to have a chance of shaping the outcome. The WGA is probably the most powerful writer’s body on the planet, but they only protect a subset of us. Writer’s organizations, big and small, need to make their presence felt.
Ok, my crystal ball is now fogged up and my unicorn is complaining that its feet are cold, so I think that’s enough for this week. Next time, I’m going to take a more detailed look at how LLMs are likely to affect writers in different media, and there will be at least one post following that in which I’ll try to lay out what steps to keeping this technology fair and equitable.
A final note about Bloody Waters. Several readers have mentioned to me that Amazon have not only failed to make the book available, but they have emailed them to see if they still want their pre-orders filled. This an issue between Amazon and the distributor, and I am told it should be fixed in a couple of days’ time. What was I saying about a level playing field, again?
It’s incredibly disappointing, given that pre-orders were open for months, and I thank you for your forbearance. If you don’t want to wait for Amazon, the book is available in paperback and all formats direct from the publisher, or Barnes and Noble has the ePub digital books for sale.
Franksly yours
— Jason