The New York Times has an interesting article by Kashmir Hill, about her reaction to discovering a 90-page AI biography of herself on Amazon. Here’s a link to the piece at The San Francisco Examiner. To say the biography wasn’t very good understates the case: just a collection of a few disparate facts and fictions cobbled together by AI. What makes a person make such a book? Ms Hill tries to locate the author of her biography but manages only to find another “writer”, Bill Jones, who publishes ten AI-generated books a month on Amazon (the most they’ll allow) — he now has 445 books available on Amazon. He is quite content to sell few copies of any one of them, even in several cases, no copies at all. Indeed he claims if he gets a really negative review of one of his books he takes the offending volume down. “I don’t take [the reviews] personally” he says, and as Ms Hill comments “Why would he? He didn’t, technically speaking, write the book.” He compiles his books as a pastime in his retirement “I wouldn’t characterize them as flying off the shelves” he said, “but it’s bar money at least.”
Inevitably these AI-generated books are not very good — though I suppose just like those monkeys typing Shakespeare, it must be at least theoretically possible that an AI-generated book could end up being a worthwhile book. But that’s not the point, is it? Nobody is pretending these books are “good”. Even Amazon, which is happy to overlook the “problem” for most of the time, will take down a book which is heavily criticized by customers. The point is, people will buy this sort of stuff. If you stand on a street corner waving dollar bills for long enough someone will soon enough appear willing to take the money even if they don’t have anything of value to swap for it. We keep getting bombarded by articles telling us that nobody reads any more, yet crap like this is being bought every day. Obviously people are willing and eager to read — and, sorry commentators, they clearly are reading, even if some of the stuff they’re reading isn’t great. (I suspect there is a fair bit of value judgement in those articles about how hoi polloi can’t read any more — only we, cultural commentators that we are, are smart enough to grapple with War and Peace!)
In so far as there’s demand for AI-generated slop, producing it is surely a perfectly legitimate endeavor. Moral judgement is beside the point. If some book buyers don’t care, why should not the enterprising entrepreneur satisfy the demand? It’s not as if the existence of bad books somehow imperils the quality of good books. You read what you chose to read. If you just want to buy a football-fan friend a birthday present of a book which you suspect they’ll never read, what does it matter that Bill Jones’ College Football Dynasties is “littered with mistakes throughout and the author tends to ramble on and be very repetitive”? It’s the thought that counts. Of course Mr Jones probably never rambles: ChatGPT’s the one who’s guilty as charged. But so what?
For more ranting on the ethics of AI writing, please see also “Shy Girl” fiasco.
Making Book 
Related, a just-released study: “Generative AI floods and dilutes the market for books”
https://arxiv.org/pdf/2607.20349
Thanks for this. A lot of math cracking a smallish nut! I’ll write about it soon.
Viewed one way, the AI slop book is merely an extension of the speed-written genre fiction book. Back when I was paying attention to discussions within the self-publishing world, I was intrigued by a blog comment. There had been a mention of second drafts. The commenter questioned the idea of a second draft as a waste of time. Finish the first draft and upload it to Amazon. Second drafts are the work of dilettantes. In fairness, this seems not to have been a widespread opinion (at least out loud), but there was no pushback. It was within the range of acceptible opinion. Presumably there was a market for these speed-written drafts, so why not? So long as the buyer knew what they were buying, there is no problem here. Similarly with a totally AI-written book.
But this is with a qualifier for purported nonfiction. Market a book as nonfiction and you are asserting that its contents are true. The essence of AI is that it can’t back up this assertion. OK, neither can we examples of fallen humanity. But we can make an honest effort, and own up to our failures. The purchaser of AI-written genre fiction is looking for nothing more than an hour or three’s diversion. The purchaser of nonfiction is looking for something more, and this is something that AI by its nature cannot provide.
In short, my objection to AI-written purported nonfiction is not that it is AI written, but that its marketing is inherently fraudulent.
But isn’t it possible to generate a non-fiction book and then carefully go through it and confirm that the assertions made are all true? That may not get you into the ten titles a week club, but I suspect it is how a whole lot of academic writing is now being done! Obviously using AI to draft a book is way beyond the use of spell checkers and even grammar checkers, but it lives, does it not, in the same general category of writer’s aids? I believe the fraud comes in in the acceptance of that first draft output, not in the whole concept.