Some people will value that a story is “organic” and be will be willing to pay a premium for it. Other people will care less about authorship and more about price. Like any other broad market, different publications will cater to different consumers. That indicates a healthy market.
For comparison, look at the history of pen names. Authors will use different bylines, depending on their audience. It’s a type of branding. No reason an established author might not dabble with machine-written drafts or rewrites, and publish under a pen name. (I see the problem Clarkesworld faces as a matter of stories submitted under fraudulent pretexts, rather than machine vs organic authorship.)
If we get to the point where I can ask for a book to be written for me that I can set parameters as to plot, setting, characters and any other variables I care about, and it’s written in a few minutes for less than it costs to buy an author’s book, then that’s the way people are going to go.
We’re not there yet, but we aren’t that far away either.
I’m sure it already is. Unless there are guardrails to specifically prevent it from doing so, I’m sure people have gotten it to churn out some Penthouse Forum style stories.
When Chatbot and Dall-E 2 get together and start making movies, I expect the majority to be based on a rule 34 for something.
But the developers are not developing their methods to spam Clarkesworld, so I doubt they will customize their code to do so. The spam is a side effect. I’m having a hard time imagining a good business case to develop code specifically to sell AI-generated short stories.
Novels would almost be easier, since Amazon is not going to care how the books get generated and quality assessment is delegated to customers. But most self-published books sell a miniscule number of copies. A few million more bad books is not going to be a problem.
The point wouldn’t be to spam Clarkesworld, the point would be to improve their AI’s storytelling capabilities. Clarkesworld et al will have to develop tools to sort out the deluge of AI generated content, and those tools can be used to improve the AI’s story telling capability.
Have you tried asking ChatGPT?
But, you are thinkinging too narrowly. It’s not specifically to sell AI-generated short stories, it’s to improve everything the AI does.
With the image generator Stable Diffusion, anyone can train their own specialized extenstion to the dataset (tuned to generate their cat, for example) and compile it into a new custom dataset. All that is needed is a generalized chat AI that individuals can tweak with their own mods and you’ll get tons of people making custom versions for writing in specific genres.
This problem can be mitigated. OpenAI has already released an “AI text classifier” that has some utility, and I’m sure there will be other free and commercial versions. Publishers already have a set of automated plagiarism tools that significantly reduce the amount of man-hours required to filter that type of submission.
I’d assume the dataset includes published, that is relatively good, stories, so I’m not sure how including others would help the result. And the question I was addressing was an arms race to prevent AIs from recognizing the output of AIs. I doubt that the key to doing this is the quality of the piece, because then you’d get a lot of false positives that rejected real, if bad stories as AI generated. That’s something I think we can expect to see, but not what I was getting at.
I bet that an AI can be fairly easily trained to pick up the tells for AI generated stories, and bad quality might not be one of them.
Perhaps the answer isn’t to stop AIs from writing fiction, but to create AI editors that can examine submissions rapidly and flag the ones of a certain quality for human review, whether they are created by AI or not. Maybe we should focus on solvinng the spam problem and let AIs compete with humans.
We could use AI editors anyway. Even without AI submissions, sometimes the stack of manuscripts in the slush piles of publishers get so unwieldy that authors are given short shrift or have to wait months for rejection or acceptance.
This would work better as part of the AI training process. It’s a specialization of what is already done.
What I find frustrating about the AI submissions to Clarkesworld is that the intention is to abuse the system and make money with minimal effort. Must submitters don’t care about the magazine or if their story is actually good. It’s just another gold farming grind. But it ruins it for everyone else.
If a submitter was actually able to generate a good story, then they may as well generate a bunch of stories and publish their own magazine.
That is the immediate issue. The fact is that at this point AI fiction isn’t anywhere close to being publishable. Editors have said they can spot it in a paragraph or two. (This was always true of much slush fiction.)
It’s the volume that causes the problem. That many unsuitable stories slows down the process. Good stories get lost in the flood. Editors may have to restrict submissions to names they recognize, which is a disaster for new authors starting out.
Book publishers faced a similar problem and closed to unagented submissions, but agents don’t usually represent short fiction.
Another solution is to go back to snailmail submissions, but that locks out those who can’t afford postage, especially those living outside the US.
Any AI solutions are in the future, and the problem is occurring now.
People would sign up. If they choose to try to employ some sort of automated system themselves, then I don’t know there’s much to stop them from doing so, and it may turn out that crowdsourcing that helps develop detection tools as well.
And as they would get a random assortment of submissions, it is unlikely that they would be able to rank their own, and even if they did, they would not be the only ones to do so, and if their ranking is far outside of what others have judged, then the weight of their rankings may drop.
Who would be randomly assigning the submissions and organizing the volunteers?
This seems like a fair amount of work just setting up the review process. Organizing a sizable number of volunteers is tremendously difficult, and Clarkesworld isn’t exactly a huge money making operation.
Probably an automated system. Ask ChatGPT for help.
Admittedly, it’s been a while since I’ve gotten in depth into web development, and a lot has changed since the late 90’s, but I don’t see why it would be so hard to do.
I’d be surprised if there is no boilerplate system that would allow someone to read a piece of text and submit a ranking on it.
Playing with wordpress let me set up a system where people were presented a grid of dogs that people could vote on which was the cutest. That took me about an hour and didn’t tax my massively out of date skills.
For someone with up to date skills, something like my proposal should be fairly simple.
I don’t really see the technical side of this as being that much of an obstacle.
