The rapid rise of artificial intelligence in both technical and creative arenas has been a two-edged sword. The U.S. Copyright Office is struggling with the idea that an art tool powered by AI cannot generate images without the direction of a human being and has so far declared that images created with AI cannot be copyrighted, while ChatCPT and the new AI powered Bing search engine have been doing everything from helping developers write software and doing students’ homework for them to professing its undying love for a human operator.
Now AI is helping humans in a brand new, unexpected way: it’s choking the submissions inboxes of publishers with AI generated trash.
Neil Clarke, editor of Clarkesworld, a science fiction literary magazine , wrote a description of the problem on his personal blog:
Since the early days of the pandemic, I’ve observed an increase in the number of spammy submissions to Clarkesworld. What I mean by that is that there’s an honest interest in being published, but not in having to do the actual work. Up until recently, these were almost entirely cases of plagiarism, first by replacing the author’s name and then later by use of programs designed to “make it your own.” The latter often results in rather ham-fisted results like this one I recieved in 2021:
These are the same sentences from the original story, “Human Error” by Raymond F. Jones, publiced in If (April 1956).
Towards the end of 2022, there was another spike in plagiarism and then “AI” chatbots started gaining some attention, putting a new tool in their arsenal and encouraging more to give this “side hustle” a try. It quickly got out of hand:
Graph starts in June 2019 and displays monthly data through February. Minor bars start showing up in April 2020. Mid-21 through Sept 22 are a bit higher, but it starts growing sharply from there out. Where months were typically below 20, it hits 25 in November, 50 in December, over 100 in January, and nearly 350 so far in February 2023.
(Note: This is being published on the 15th of February. In 15 days, we’ve more than doubled the total for all of January.)
I’m not going to detail how I know these stories are “AI” spam or outline any of the data I have collected from these submissions. There are some very obvious patterns and I have no intention of helping those people become less likely to be caught. Furthermore, some of the patterns I’ve observed could be abused and paint legitimate authors with the same brush. Regional trends, for example.
What I can say is that the number of spam submissions resulting in bans has hit 38% this month (. While rejecting and banning these submissions has been simple, it’s growing at a rate that will necessitate changes. To make matters worse, the technology is only going to get better, so detection will become more challenging. (I have no doubt that several rejected stories have already evaded detection or were cases where we simply erred on the side of caution.)
Clarke goes on to say that the tools for detecting plagiarized and machine written text are not only prone to false positives and missed negatives, they’re also too costly for any of the short fiction markets to afford.
According to Clarke, other editors are seeing the same problem, and just five days later his spam rate had risen from 300 pieces to over 500, more than tripling their spam rate over the previous month.
As you can see by his second graph, it’s nuts.
More importantly, it’s almost all unreadable garbage. No, the science fiction short story market isn’t dead — but until somebody figures out a solution, it’s going to be a lot lot harder for both publishers and writers to get anything done.