Image credit: Moor Studio / Shutterstock.com
Chatbots are already writing books. Predictably, they’re flooding Amazon as people try to make a quick buck. So far, though, they haven’t caught fire with readers. As the Washington Post notes, much of the A.I. generated text out there is “never seen by human eyes.”
Scams and grifts were inevitable with chatbots. Hell, Amazon was already overrun with plagiarized books and books compiled by web-scraping bots long before ChatGPT. The more interesting question is how readers will respond to attempts to write real books with chatbots. Personally, I’ve long felt that the audience will be more limited than many believe. Using chatbots for research or as Wikipedia hole replacement has problems—chatbots are still entirely unreliable for information—but I can understand the appeal. But who exactly cares what a computer program writes about being in love or going to summer camp or losing a pet or anything else it can never experience?
So I had an enjoyable amount of schadenfreude reading a Newsweek essay hyping A.I. book sales:
Thousands a month, you’re thinking. One could live on that! Well, not exactly. Though the “author” brags of “impressive returns,” the essay claims sales of “574 books for a total of nearly $2,000 between August and May.” Obviously Newsweek has been a joke magazine for some time, yet this is still a bit funny. He claims to spend an average of 7 hours making each book and has produced 97 books. So by his own math he’s sold an average of 6 copies per book and earned less than $3 dollars an hour. Not exactly the most lucrative hustle.
While Newsweek guy is running a transparent grift, the first interesting A.I.-authored (or co-authored) book dropped this month and seems to be having similar struggles. Death of an Author by Aidan Marchine is a collaboration between the writer Stephen Marche and a series of different chatbots. Marche has positioned himself as something of an A.I. art evangelist, lauding the technology in countless essays in places like The New Yorker and The Atlantic. I find his pieces somewhat credulous toward the Silicon Valley hype—and his misunderstands many critical positions, such as the WGA’s in the current strike—but Marche clearly cares about the artistic possibilities of A.I. in literature. His novella seems a serious attempt to prove his thesis that A.I. collaboration isn’t just artistically possible but is indeed “the defining skill for the next era of human creativity.”
A big claim. And I admire someone putting their money where their mouth is. Marche spent a lot of time on this book collaborating with multiple LLM programs. Perhaps even as much time as it would have taken to write a novel from scratch. Yet despite getting the review placement most human novelists could only dream of—WIRED, Slate, The New York Times, etc.—the book currently sits at single digits in Goodreads ratings. (To be fair, media coverage doesn’t drive sales like it used to…)
I read the novella’s excerpt on WIRED with real interest. Despite my many critiques of the A.I. hype cycle, I’ve always been interested in artistic constraints and believe there will certainly be real artistic possibilities with chatbots. I’m not sure we’ve reached it yet though. While the excerpt is more interesting than the generic chatbot essays I’ve had students submit, I can’t say it was particularly gripping. It still has that feeling of being written by a machine made to produce generic text, especially when it comes to dialogue:
“Anything to eat?” Julian asked.
“Yeah, I’ll have a plate of nachos and a beer,” Gus replied. […]
“The breakfast of champions,” Julian said with a smile.
Mostly the text seems to be lacking authorial vision in the prose itself, at least in the excerpt. There’s conceptual cleverness that the human came up with, but that fails to bleed into the sentences and paragraphs. It’s passable as prose, but there’s little personality. Compare this to the first few pages of a Raymond Chandler novel.
Of course, this is an early experiment. Perhaps future programs will sound less robotic and future artists will figure out more vital ways to use them. “A.I.” is such an intentionally broad an ill-defined term that it’s inevitable most authors will soon be using things branded “A.I.” (I expect Microsoft Word’s spell and grammar check to be renamed something like Bard A.I. Editor soon.)
But in my view, the artistic failure of chatbot writing stems from writers focusing on how chatbots can reproduce the kind of texts we already have. Almost every article about LLMs focus on how they can reproduce text that’s similar to text humans write each day. To which I say, so what? If a chatbot can produce a plausible noir novel with a passable plot and workable characters, that’s a technological achievement. It’s not much of an artistic achievement. There are endless noir novels to read already on the shelves.
Let’s assume a future chatbot can live up to the predictions and, say, scan every Stephen King book and then produce a new Stephen King-ish novel. Who does that serve, exactly? King already writes books and indeed he’s written more books than all but the most dedicated fans can ever read. And his dedicated fans aren’t likely to prefer a computer simulacra to the real thing. In general there is more quantity—and more quality quantity—to any given genre or style than anyone can read in existence today. We don’t need more quantity of novels that read the same as novels humans already write.
Instead, innovation will likely come from what chatbots can do that humans can’t.
The favorite metaphor of A.I. evangelists is photography and how it didn’t replace painting but created a new artform. Yet they seem to forget that it was a new artform. Photography blossomed with photographers embracing the unique things the medium could do, not by trying to make every photograph look exactly like a 17th-century still life.
So what can a chatbot do that a human can’t? One thing that stands out to me is individualization. A chatbot could theoretically customize a novel to infinite individual readers in a way no human ever could. Perhaps this would take the form of mere self-insert fanfic—almost certain that will happen with A.I. porn—but how about the book version of an open world or procedurally generated video game? I’m sure there are many readers who would love to choose-their-own-adventure style text romp through Westeros or Middle-Earth (or better yet a new fantasy world designed for this purpose).
I’m not saying this is terribly interesting to me. And it doesn’t address all the ethical questions about corporate-controlled programs that have been built on the work of human writers who had no say and got no renumeration. But this would at least be something new rather than just a computer program pastiche.
This post initially appeared in a slightly different form on the author’s Substack, Counter Craft, on May 16, 2023.