I have mentioned before, I think, that the vast mega Corp for which I sometimes work is heavily invested in AI. It is far advanced in its adoption compared to its peers and, as a result, I have benefited (if that’s the word) from a great deal of AI training. Regardless of our initial reactions, which varied from enthusiastic adoption to outright hostility, my sense is that most now view AI as a useful productivity tool, like e-mail or word processing, rather than the unstoppable ender of humanity the tech bros would have you believe. The tech bros, of course, need you to believe that. It’s how they raise money: by persuading CEOs and investors that AI will slash wage bills by ending employment as we know it. I feel that, for the most part, my colleagues no longer believe the hype. They simply want to be left alone while they commune with their agents.*
The publishing world is not so sanguine. Even if AI is simply another tech development like email and word processing, it is easy to forget just how disruptive those technologies were. Secretaries have largely disappeared; the USPS learned to make a living from delivering parcels rather than mail. Even if the world continued much as usual, whole classes of jobs vanished. When it comes to possible extinction, publishing feels that it is very much in the front line. After all, AI is simply a marketing term for Large Language Model, and language is what publishing is all about.
I have tended to take a relaxed view about this. AI writing, which apes the truncated language of the internet (now further polluted by AI’s own slop), is not a style well suited to long-form storytelling and, because it can only mix and match what already exists, AI stories are utterly derivative. Fiction, at least, is safe.
Or not. A few things have caused me to question that assumption recently. The AI companies, in search of a better writing style, are now stealing the intellectual property of authors by buying second hand books, scanning the contents and literally destroying the evidence. Undaunted by lawsuits related to their pillaging of books online, the AI companies continue to perpetuate their original sin: taking the fruits of human creative endeavor without paying for it. Regardless of whether this activity is legal or not, it is reasonable to assume that AI-generated texts are going to become less weird and stilted as a result.
Second, there is increasing evidence that, even today, readers are not put off by AI’s “bad” writing. A recent study found that people actually preferred AI short stories to the human equivalent, not least because they were less complex and easier to read. It is instructive to remember that so-called bad writing is often in the eye of the beholder, and that bad writing has never been a bar to publishing success. Consider the furor around Shy Girl, the high-profile novel that was pulled from the shelves following allegations that up to 78% of it had been written by AI. While it has been excoriated by some as a terrible novel, that rather misses the point. It was reviewed by professional editors and published, initially to great commercial success. Regardless of its alleged quality, people liked it.
With regard to Shy Girl, let’s assume the allegations are true and that 78% of it was written by AI. That’s still a long way from 100%. I say that because I generate a lot of AI-written output in my day job. It is verbose and stylistically terrible, but if I don’t care about that and the content is accurate, I send it out. No one reading it would believe that I wrote it myself—we have all become quite savvy about what AI writing looks like—but they assume I am vouching for its veracity and act accordingly.
If I do care, though, I edit the output so that it looks as if I wrote it from scratch. Now, here’s the thing: to get that effect, I doubt I change more than 5% of the words and I delete maybe another 10%. The effect is entirely disproportionate to the amount of effort involved. My guess is that Shy Girl would have been an utterly unreadable mess if it had been written entirely by AI. A short story is one thing; a novel is a different beast altogether.
I think it’s reasonable to assume that we are now in a world where decent short stories can be written entirely by AI, and decent novels mostly by AI. Thanks to the continued abuse of authors’ hard work, readable, engaging** novels written entirely by AI might not be too far away. What, if anything, is to be done about it?
I think it will depend on the reader. If readers really don’t care if a story comes from the mind of a fellow human being, billions of stories will be published every day—either in the hope that a few might make real money or that a couple of cents per occasional story when you’re putting out 10,000 an hour 24/7 will generate an income stream. Publishing companies as we know them will vanish, either changed out of all recognition into Amazon-style marketplaces offering automated curation or swept away altogether. In a world like that a small number of human authors might be able to make an Etsy-style living from the creation of old-fashioned hand-crafted goods—a version of today’s self-publishing ecosystem—but most of us will be writing books for the same reason we have always written books: we love to tell stories. That won’t change, not ever. We just won’t get paid for it.
If readers do care, and my guess is that they will, then the publishing industry will have to learn to be more honest about how books are written. I have said before that celebrity books written by ghost writers are tantamount to a fraud on the reader. If a book is ghostwritten, the ghostwriter’s name should appear on the cover: people should know if they’re reading the words of an actual celebrity, or words that a celebrity has merely authorized. In a similar way, readers buying a book for its “humanity” should know the extent to which the author made use of AI. I’m imagining books classified on a scale from one to ten, or violet to red, where the low end starts at none at all and rises through research, copy-editing, actual editing and so on until we get to the wholly AI novel. And if a scale like that was backed up by fines and jail time for misleading the reader, I would have no problem with that!
*AI agents, not the literary kind!
** Pure AI novels already exist, of course, but without human input they are neither readable nor engaging.