The death of democratic publishing

In other Eneasz-releated news, Eneasz recently blogged his reasons for believing that SF&F magazines (and, I'd assume, all print publishers) will soon have to stop taking unsolicited submissions, owing to the overwhelming number of AI-written stories being sent to them.

Comments

[silvadel]

AI stories are easy to pick out. They have a style to them. You can even ask AI whether it was written by AI. Also since anything done by AI has no copyright, you just have to limit things to the copyright holders of stories. True someone could commit fraud by saying something was their original work when it wasn't, but that would rise to the same level as plagiarism with respect to their reputation going forward. Most wouldn't risk it.

[Bad Horse]

5750001
Yes, but you still have to read enough of the AI story to pick it out. You can usually pick out an AI story or a bad human-made story within 1 or 2 paragraphs, but you still have to read those paragraphs. That's what takes most of the time of slush readers.

[Loganberry]

5750001

AI stories are easy to pick out. They have a style to them.

For now they're easy to pick out. That will get harder as AI gets better. As an admittedly highly flawed analogy, consider how easy it was to tell if a singer was using pitch correction software in that technology's early days. That changed. This will too, and probably quite fast. None of this is to quibble with the answer Bad Horse gave of course. I'm just fairly sure that the other argument you see a lot ("It's easy to tell AI writing as it's just a load of obviously artificial trash") is not one that's going to hold up much longer.
And that's before even getting to the idea that in certain genres, eg airport novels, a lot of readers really don't care about beautiful prose anyway. My hunch is that sectors like that will go largely AI relatively soon. Disclaimer in all this: I am absolutely no expert and could easily be missing something really obvious.

[Calipony]

5750011
I don’t agree, but that may be me and my conservative views. AI can be used to draw pictures or so-called artistic pictures, because everyone knows modern art is just bullshit so whatever sloppy, meaningless material you propose has a chance to slip in. This is not the case with literature, I guess. There’s no creativity in AI, it just blends up random chunks of already written sentences. Maybe it can infer new sentences out of deduced invariants — but I doubt it will ever be able to master some very basic concepts such as metaphoric writing or other figures of style (like, say, zeugma: he took offence and leave).

AI is a fad. It’s like string theory in physics. The era we’re living in looks desperately for theories or devices able to answer every single question we may ask. When the electrical grid will fail because we won’t be able to squeeze terawatts of power into highly inefficient silicon chips for measly results, the bubble will burst.

[mishun]

5750014

it just blends up random chunks of already written sentences

So... like humans?

[Herald of Opera]

5750033
There's a difference in theory, but the likes of Calipony don't usually comprehend theory.

[Calipony]

5750039
That’s right, I prefer practice. :p

[Wild Zontars]

The venerable publishers that carried the torch for so long will continue to be worn away by a million AI minds.

Assuming they don't all go bankrupt (after merging into one or two Mega-Publishers) first.

The only way I can see them coming out of this is to limit submissions to only authors proven to have existed before Generative AI was born,

If they go this route, the only question is whether they run out of living authors or readers first.

or who have become popular enough afterwards that their human existence isn’t in question.

In which case Indie wins the day, because once you're popular enough in an open market to be noticed by TradPub, a) do you need TradPub, and b) does TradPub want you and your wrongthink?

At least they've found a way to solve puppy-related sadness once and for all!

[Georg]

5750001 The AI detection of AI generated material is not flawless. From what I understand, it will flag anything that had been used as seed generation material as AI created, such as Monster Hunter International. (Yes, they stole seed material from just about everybody)

[Wild Zontars]

5750065
It also seems to flag anything with a particular "style". This is getting a lot of students in trouble, because if they write in a proper academic style, their "AI" score goes through the roof. (Amusingly enough, I had this same problem back when I was in school, teachers kept insisting I was writing too far above my grade level to be doing it myself.)

[Loganberry]

5750014

AI is a fad.

This is the key thing we disagree on, I think. I don't think it's a fad. I think aspects of current AI hype are probably fads, but I don't think AI as a whole is going away from the creative arts now -- whatever we here, or professional artists, or Hollywood writers, or anyone else might prefer. Of course it's a different thing and doesn't have any direct bearing, but I'm old enough to remember in about 1993 when it was fairly common for people to dismiss the web as a passing fad.

Not saying this applies to you, but I suspect that part of the reason so many people still dismiss AI is that their contact with it is through the free ChatGPT bot. That's nowhere near state of the art any more. A while back I asked GPT-4 and the (then-)current version of Claude to write a ponyfic for me, and though neither matched a high-quality human the Claude effort was very clearly better. (Here's the link, for anyone interested.)

I could of course be entirely wrong about this. Time will tell.

[silvadel]

Re: all pretty much...

The point is, you can have a tiered system and don't need anything more than an intern to pick out the stories that are AI generated from the rest. The actual slush readers would come after you used AI itself and the lowest of the low to sieve through the detritus. They would be reading, as you said, only a small portion of the submitted works just to see if they are AI or absolute trash.

That kind of tiered system is used all over the place, and yes it has some false positives, but if you accidentally toss 10 or even 30 percent of the legitimate stories, that isn't necessarily a bad thing (for the company, not the person who sent in something that mapped too closely to AI).

[Bandy]

The blog brings up one really feel-good point--that as it becomes increasingly difficult for writers to break professionally by getting published in prestigious spaces and utilizing the capital associated with publishing clout, those writers will increasingly revert to a DIY style of self-publishing and sharing with small circles of critics and friends (something, dare I say, we already do on this site ). However, the author fails to consider just how good big-name clout feels, and how much I personally want it. Big name publishers and beefy book deals are the only way to satisfy my vain ambitions. We must find a way forward without eliminating opportunities for me to flex on my friends and something something artists need to make a living too, whatever

[Viking ZX]

I'll chime in.

Yes, it's kind of bad right now. Most places have shut down submissions because they've been absolutely flooded by cheap, crappy, AI content. Overwhelmed, even. And for now, it's going to stay that way.

But it can't stay that way forever. The genie of cheap, crappy plagiarism of other's work can't last forever. Eventually laws and court rulings are going to be established. AI work (or even derivatives thereof) will likely, or at least hopefully, be required by law to declare itself as such, and that's going to bust the "bubble" of a lot of these submissions.

Because in full truth, 99.99% of these submissions are from actual authors, or even would-be writers. They're from the people always looking for a silver bullet, always talking about how "someday they'll write something" while never producing much more than a line here or there because at the end of the day they're not committed to the work. So when a "silver bullet" comes along that will 'write the story for them' they hit the "generate" button and then send it off without even bothering to ask if anyone actually cares to read whatever junk script they've produced.

AI-written works have serious flaws, something I addressed when I wrote a piece on it a month or so ago. The issue isn't that AI stories are going to drive authors out of creating actual creative content—the tech is just not there yet, and won't be for at least a few more years at the very quickest—but that what's currently going on is a flood of "get-rich quick" schemers moving to their next scheme, and those houses that take submissions are just being overwhelmed.

Yeah, it sucks. But it's not going to be a permanent problem. Those houses are going to start rolling out methods of streamlining. Some will probably more aggressively engage in only taking submissions from paid agents, a secondary form of gatekeeping that has already been gaining traction among traditional publishing houses (though for a purely financial reason: It allows the trad pub to save money and puts more of the publishing cost on the author). Others will start building "tiers" of submitters that grant ability to be recognize based on prior works (and I note that this also has drawbacks, but this is going to be the "what works" phase). Others will try using AI to catch AI submissions, but will run into issues of AI copying works that are then being submitted, creating new issues.

But the system will likely start to self-sort and correct. Axtara - Magic and Mayhem will be launching with a line on the copyright page that declares that no part of the text or cover was generated using AI. Recent issues such as AI-written mushroom picking books brought legal worry to storefronts who were selling AI-books that stated inaccurate things about mushroom picking, advising people to pick and eat some of the deadliest. That's massive legal liability, and a great solution is for the sellers to take a vested interest in making sure the AI buck is passed all the way to the "writer" of said book ... along with all the legal issues.

It's actually a more democratic publishing world that has allowed AI-writing to surge. The boom of indie publishers and authors year over year is what makes it possible for some lazy gold-hunter to publish a bunch of AI-written plagiarized junk. It's the filtering process that's having trouble as that flood of cheap, crappy content hits the industry, not the ability of anyone to engage in it. There was a similar flood of crappy music during the early MP3 years, when companies like Apple dropped products like Garageband and suddenly anyone could put their own MP3s out on Napster, Kazaa, or up for sale on storefronts.

But the system stabilized. AI-generated "writing" in the publishing world is a far larger flood, certainly, but it's not as though the industry wasn't undergoing a huge upheaval anyway with the advent of the internet. This is, effectively, just one more blow in the hurricane, and while it might capsize a few ships it's not going to sink publishing as a whole.

If anything, the establishments that are most at risk are the most entrenched, traditional publishers, because they've proven too immobile to adapt to sudden changes such as eBooks. They'll likely still be reeling from AI-content long after most other smaller pubs and indie authors have just continued onward.

[Bad Horse]

5750072

The point is, you can have a tiered system and don't need anything more than an intern to pick out the stories that are AI generated from the rest. The actual slush readers would come after you used AI itself and the lowest of the low to sieve through the detritus. They would be reading, as you said, only a small portion of the submitted works just to see if they are AI or absolute trash.

The tier is:
slush readers
editor

That's it. Because nobody gets paid for reading slush anymore. The slush readers are volunteers, because there's no money in publishing magazines.

I don't think there are interns except at major magazines, like The New Yorker, and even they don't get paid anymore IIRC. This helps keep middle-class people out of the respectable publishing industry, which is as important to the Manhattan set as keeping Jews out of the Ivy leagues.

[publiq]

I found this linked blog post from the OP to be enlightening on their view. I like the implication that mass-market entertainment will somehow become even more bland with AI that people will return to creating art dedicated to their friends. In a way, we're already doing that here by writing fanfiction.

Somewhat related, but the #1 reason why I'm often so antagonistic to anti-AI preachers in public discussions is that their message most often comes across as a combination of "git gud, scrub" and "stop having fun." In a world where Sable Diffusion and LLMA didn't exist, I'd probably view them as strategic allies so long as AI creativity tools required paypigging to the cloud. However, these tools can be run locally. I doubt Adobe or Getty will release their supposedly ethically sourced models to the wild. "Sure, you can have copyright-respecting AI output, but you'd better pay up." Information is meant to be shared.

5750184

which is as important to the Manhattan set as keeping Jews out of the Ivy leagues.

To be fair, the modern Ivy has primarily shifted its (former?) anti-Jew gatekeeping towards preventing an excess of Asians. Same mechanisms, different target. Still immoral.

[Herald of Opera]

5750065
Incidentally, that particular example is rather odd to this case; if it catches both AI and actual plagiarists then that saves even more time.