I just saw a Ted Chiang article in the New Yorker describing current “AI” and why he thinks the fundamental nature of large language model -based software means it isn’t ever going to produce anything good. Which is in line with how I’ve been thinking about it - it’ll get more intelligible, but the way it works prevents it from getting better in terms of telling stories. While there’s plenty of places where intelligible-but-lousy text is useful to someone, published fiction just isn’t one of them (self-published, on the other hand…) That’s simply not their niche - if they cared so little about quality, they could have already done much worse without AI. (Plus, in the grand scheme of things, the writer is a relatively minor cost in publishing - there’s a lot of people who need to earn a living working on getting a book out, but the author isn’t one of them. Few authors make a living at it, but almost everyone working full time in the publishing industry must.)
I doubt it very much as well (especially with current technology), but hypothetically if it happened, the publishing model for fiction would just be broken entirely, rather than publishers cutting out writers from the process.
