Tulip

[Epistemic status: very low confidence in all predicting-the-future I do herein, to the point where it's probably better to treat it as worldbuilding than as genuine prediction]

It seems plausible that, some time within the next few decades, there will be machine learning systems capable of generating genuinely-good works of fiction, with levels of character consistency, thematic consistency, interesting worldbuilding, and so forth which make them viable competition to human-written novels. Once such systems exist, they're likely to thoroughly outcompete humans within the domains they know how to write in, given their vastly-greater ease of producing a high volume of content in a short time. Which isn't to say that humans will stop writing things, of course, but I expect it to make a major dent in humans writing with the expectation of becoming well-known or financially-successful.

However, it seems likely that their domains of effectiveness will start out somewhat limited genre-wise. In particular, I'd expect them to be very bad at making use of narrative devices and techniques which are neither common in the corpus of stories on which they were trained nor easy to invent just by combining those common devices together. So, for example, I'd expect them to take much longer to figure out Unsong-style TINACBNIEAC-isms, and especially to figure out how to actually tie them into the worldbuilding as Unsong does, than to figure out quasi-old-timey narration and soft magic as featured in Lord of the Rings and its hordes of imitators.

As such, I don't think that humans writing with fame or money as goals will be outcompeted all at once. Rather, they'll be driven from the more-formulaic genres into the less-formulaic genres as the machine-learning systems take over the more-formulaic genres. And that's going to make for a really interesting shift, when it happens. Even relatively early on, it won't be enough to just write weird genre mashups like magical girl pirate spy thriller, because "combine these disparate elements into a unified picture" is the sort of thing that machine learning systems are good at; instead, the name of the game is going to be inventing new story-elements (or at least making use of little-used ones which don't yet comprise a good pool of training data). Unusual plot structures, or novel worldbuilding elements, or distinct styles of narration, or new character archetypes; just, in general, finding things which the machine learning systems can't generate merely by permuting the data they already have.

Now, if I wanted to continue going down the speculation-about-future-society path, there would be a bunch of interesting stuff here. You'd get the traditional publishers collaborating to avoid releasing too many stories containing any given element, for fear that the machine-learning systems would learn to generate those elements given an overly-large corpus and thus hijack yet another chunk of market-space. You'd get new-style publishers specializing in generating particularly-good things with the systems. You'd get a rise in weird medium-bending stories like Homestuck whose formats make them harder to imitate. It's generally a pretty interesting scenario to think about.

But of far more interest to me is asking: what will those new AI-dodging stories be like? What new depths of story-space will be explored in the process of keeping the writing-for-money industry alive? I'd expect the pressure to lead people to come up with lots of new interesting story ideas, in a way that current incentives don't push towards to nearly the same extent. But, even if the incentive isn't there for us, it's still just fun to speculate: how might I put together a story which would (a) not be easily imitable by a story-bot and (b) remain actually good and not throw out the narrative baby with the bathwater?

(And, of course, if I get any particularly cool ideas, there's no need to wait for the story-bot apocalypse before writing them up; novel story ideas are good even outside of the hypothetical future where the publishing industry is incentivized to reward them.)