AI written books

[Critteranne]

Grimdark magazine also posted a statement on AI.

They say that this is going to impact their open submissions window. For now, they have paused open submissions and are going on a "solicitation-only basis." They also realize this sucks because they are looking for new writers in all parts of the world.

[gmw]

Originally Posted by Critteranne

Grimdark magazine also posted a statement on AI.

They say that this is going to impact their open submissions window. For now, they have paused open submissions and are going on a "solicitation-only basis." They also realize this sucks because they are looking for new writers in all parts of the world.

Thanks for that. So publishers are treating it seriously, which says something about either about the quality of the AI generated art, or about the quality of the approval and authentication processes. Not sure which, or maybe it's both.

[ekbell]

I think that this publisher is taking it seriously because it takes time to check over a submission, see that it's A.I. generated junk (or possibly less junky A.I. plagiarism) and toss it out. And even with better auto-detect we would still need humans for the 'Does this plot make any sense' check.

Open submission simply can not work if the junk to readable ratio is too high. Particularly when the junk uses proper grammar and spelling and makes some sort of sense on the paragraph level.

[Quoth]

Publisher and printer, in some countries, can be sued for plagiarism or libel. The LLM system combines content scraped from the internet from many sources, so is technically all plagiarism (though from combined sources) and could easily have libel.
The makers of these systems can't afford to human curate the input so that it's all PD and the users invoking the output are not usually qualified to spot plagiarism or libel.

So it's not just about reading it to spot if it's junk. The better it is and less like junk, the more likely it's going to infringe rights of existing people, or have real named people and apply plausible "junk" (i.e. lies) about them.

[gmw]

Originally Posted by ekbell

I think that this publisher is taking it seriously because it takes time to check over a submission, see that it's A.I. generated junk (or possibly less junky A.I. plagiarism) and toss it out. And even with better auto-detect we would still need humans for the 'Does this plot make any sense' check.

Open submission simply can not work if the junk to readable ratio is too high. Particularly when the junk uses proper grammar and spelling and makes some sort of sense on the paragraph level.

First, what's the point of a publisher if they're not going to check over submissions? These days one of their self-proclaimed reasons for being is to act as a gate-keeper and filter.

Second, I don't think that a 'Does this plot make any sense' check helps you separate AI from human - seems to me I've come across plenty of (assumed) human texts with nonsense plots.

Third, take a look at the the earlier critique of the ChatGPT example and tell me you've never seen the same criticisms of human text. Indeed that post also critiqued another post too, on not dissimilar lines, and yet I give Quoth the benefit of the doubt as regards being human.


Panic may be inappropriate, but there is certainly enough going on in this space to make it interesting to watch.

[Critteranne]

Originally Posted by ekbell

I think that this publisher is taking it seriously because it takes time to check over a submission, see that it's A.I. generated junk (or possibly less junky A.I. plagiarism) and toss it out. And even with better auto-detect we would still need humans for the 'Does this plot make any sense' check.

Open submission simply can not work if the junk to readable ratio is too high. Particularly when the junk uses proper grammar and spelling and makes some sort of sense on the paragraph level.

The graph included in the Neil Clarke blog post is an eye-opener. Some people said the A.I. submissions could be considered a denial of service attack because of the sheer number of submissions.

And of course, this takes away from the time they can spend on submissions from human writers. Clarkesworld is renowned for how efficiently they handle their submissions. Where other magazines can take weeks, Clarkesworld has been known to send a rejection (even a personalized one) in a day or two.

[j.p.s]

Originally Posted by gmw

I give Quoth the benefit of the doubt as regards being human.

I do the same for another member with well over 60,000 posts, but sometimes it's difficult.

[gmw]

Originally Posted by Critteranne

The graph included in the Neil Clarke blog post is an eye-opener. Some people said the A.I. submissions could be considered a denial of service attack because of the sheer number of submissions.[...]

Oh, now that's interesting. I imagine the peak will drop back somewhat after the novelty of the latest generations of AI has worn off and falls out of the newsfeeds, but it's not going to go away.

I was particularly taken by the following comment:

I think it’s one thing getting machine written text to give you some suggestions for outlines to get you over that “blank page” problem that most people struggle with, but to ask it to write the entire story that they obviously can’t even be bothered to read is just crazy and insulting to the person you’re sending it to.

So, according to this person, it's okay to use AI as one of the tools in your writing kit. But just how many "blank page" problems am I allowed to fix this way?

Writers had tools to help them fix their spelling and grammar (while theoretically redundant because your publisher should fix this stuff, but actually not redundant because you had to give a good enough a show to get seen by the publisher). Then there were various editing tools to how you pick up more interesting problems. Plus there are name (and even whole resumé) generators to supplement the old baby-name books ... and now we're apparently talking about using AIs to get us past writer's block?

So publishers like Clarkesworld are not just facing the pure AI deluge, they're also facing texts that are a varying mix of human and AI, human and plagiarism and AI.

Maybe it is time for at least some to panic.


P.S. Some hours later I looked over this and found various mistakes ... but opted to leave them in as offering verisimilitude toward my being human ... although, for all I know, some AIs may now be making deliberate mistakes in order to appear more human.

[DNSB]

Originally Posted by j.p.s

I do the same for another member with well over 60,000 posts, but sometimes it's difficult.

What do you have against the Grand Mouser aka PDurrant?

[j.p.s]

Originally Posted by DNSB

What do you have against Paul Durrant?

Nothing, and none of his posts look to be bot generated.

[gmw]

While not directly about AI generated books, I thought this article was interesting, published today on ABC News Australia: Replika users fell in love with their AI chatbot companions. Then they lost them

I'm not taking this as necessarily strong evidence that AI bots can be convincingly human (I can't really guess at that, having never tried an "AI chatbot companion"), but I do suggest it is evidence of the adaptability of humans. We can, given the right incentives, overlook many flaws in order to get what we (think we) need from a relationship.

And to bring it back to this thread topic, I think this suggests that AIs could well produce books that at least some readers will find acceptable, not because they're great stories by literary standards, nor because they are producing original material, but actually because they are reproducing patterns we find pleasing. I can easily see this happening, if not now, then before long. So while we may have various publishers scrambling to assure readers all their content is produced by humans, it may not be that far off to find some publishers deliberately and openly publishing AI text*.


* And that might be of particular interest because it is starting to look like AI text may not be copyrightable in at least some jurisdictions, and so we may get a live experiment of how no-copyright competes with copyright.

[andyh2000]

Originally Posted by gmw

While not directly about AI generated books, I thought this article was interesting, published today on ABC News Australia: Replika users fell in love with their AI chatbot companions. Then they lost them

I'm not taking this as necessarily strong evidence that AI bots can be convincingly human (I can't really guess at that, having never tried an "AI chatbot companion"), but I do suggest it is evidence of the adaptability of humans. We can, given the right incentives, overlook many flaws in order to get what we (think we) need from a relationship.

And to bring it back to this thread topic, I think this suggests that AIs could well produce books that at least some readers will find acceptable, not because they're great stories by literary standards, nor because they are producing original material, but actually because they are reproducing patterns we find pleasing. I can easily see this happening, if not now, then before long. So while we may have various publishers scrambling to assure readers all their content is produced by humans, it may not be that far off to find some publishers deliberately and openly publishing AI text*.


* And that might be of particular interest because it is starting to look like AI text may not be copyrightable in at least some jurisdictions, and so we may get a live experiment of how no-copyright competes with copyright.

I suspect a lot of 'genre' fiction will be able to be churned out by AI fairly soon. Cozy mysteries, romance, even my own beloved SF probably. Anything that as you say tends to follow a pattern that readers like consuming again and again in slightly altered forms.

Andrew

[gmw]

Originally Posted by andyh2000

I suspect a lot of 'genre' fiction will be able to be churned out by AI fairly soon. Cozy mysteries, romance, even my own beloved SF probably. Anything that as you say tends to follow a pattern that readers like consuming again and again in slightly altered forms.

Andrew

And your forgot probably the easiest target: erotica.

I can't help thinking of the whole million-monkeys-at-a-million-typewriters producing Shakespeare thing. Generators/publishers of AI material will create a lot of drek, but the Internet already does. However, we already have an the online population providing a free global filtering service, so all the publishers need do is put it up there and wait for the readers to "like" stuff stuff or not, just as happens now. The publishers with the best generators will get the most visitors to the most liked books and so the most revenue from advertising. The "more like this" links will have a whole new relevance because the AI will know exactly what "this" was generated from and be able to generate more using some variation of that base.

And not just the text. The covers would be a no-brainer assigned to AIs getting too old to earn their keep.

[Quoth]

Originally Posted by andyh2000

I suspect a lot of 'genre' fiction will be able to be churned out by AI fairly soon. Cozy mysteries, romance, even my own beloved SF probably. Anything that as you say tends to follow a pattern that readers like consuming again and again in slightly altered forms.

Andrew

But it will be plagiarised from multiple human written sources. That's how the LLM works.

[jhowell]

Originally Posted by andyh2000

I suspect a lot of 'genre' fiction will be able to be churned out by AI fairly soon. Cozy mysteries, romance, even my own beloved SF probably. Anything that as you say tends to follow a pattern that readers like consuming again and again in slightly altered forms.

It also creates the possibility of generating unique stories for each user, targeted to that particular person's personality and interests. It could churn out an endless steam of books you love with characters you connect with.