(H/T: @RobinDLaws) I don’t know if The Verge has worked its way through the implications of the AI submissions problem yet. Right now, editors can detect AI-generated works of fiction, and they’re rejecting them out of hand*. Their problem? Well, let me put it another way: they’re teaching the AI how to successfully fool the editors. It will likely take some time, but the ultimate goal here is not to write the Great American Science Fiction story: it’s to come up with something that can get past the gatekeepers. That is a much more achievable goal.
It will happen. It probably hasn’t happened yet, but we’re still in the middle of the process. Worse of everyone concerned, it’s a process that will probably end up destroying the entire concept of the speculative fiction magazine — because once AI-generated text gets sophisticated enough to mimic C+ work reliably, why not just buy your own copy of the program, and have it churn out cookie-cutter content that’s specialized for you?
Anyway, all of this is going to be horrible for writers who aren’t at my skill level, a hopefully only mild annoyance for writers who are, and sheer Hell on Earth for publishers. You heard it here first: the days of slush piles are coming to a close. If you want people to notice your books, hustle.
Moe Lane
PS: My solution is to buy my books. They’re 100% meat-brain — well, I’m writing them on a computer, which will do things like catch basic grammatical errors for me,** but you know what I mean.
*They’re also in the process of irrevocably wrecking their current submissions model; which is both unavoidable, and another issue entirely.
**And demonstrate a subtly alarming ability to guess what the end of my sentences is going to be.
#commissionearned
Honestly, it hasn’t been a problem for us yet. Relatively few of our submissions appear to be AI-written, and those we think are aren’t very good, and frankly, we don’t get so many stories that we can’t read them all.
Clarkesworld’s problem has to do with their volume of submissions. I’ve talked to one of their first readers before, and they insisted that they can tell whether to reject a story after the first paragraph. I can’t do that. I give most decently written stories a few pages at least. However, I can detect a badly written story or an AI-written one after the first paragraph or so, and reject it quicker.
But, if you’re a Clarkesworld reader and only budgeting enough time to read the first paragraph for 90% of your submissions, then it doesn’t take appreciably less time to reject a AI-written story than a human-written story.
Since it takes me less relative time to reject an AI-written story than Clarkesworld, AI stories would have to take up a more significant portion of the slush pile to make any difference for me.
That’s good to hear about for your own shop. I’ve been talking to my wife about this stuff – it impinges on some of the things she works and thinks about – and she did point out that a lot of people are thinking about these problems generally, which is also good to know. I still can’t help but think that if Clarkesworld opens up submissions again, they’ll end up teaching AIs how to get past their filters.
Been thinking about machine learning applications on the help desk side – teaching the machine to send bright and cheerful “have you tried turning it off and back on?” instructions, or password resets..
I’m also bearing in mind that I better be able to put this thing down..
Mew
Mew
[i] And demonstrate a subtly alarming ability to guess what the end of my sentences is going to be. [/i]
Mine mainly has some weird thing for ducks.
(Looks innocent)