Clarkesworld, which is considered one of the top sci-fi and fantasy literary publications, has won several Hugo Awards. It regularly bans a small number of people from submitting works each month, mostly for alleged plagiarism. But as of Monday, it had banned more than 500 accounts this month, according to a blog post written by Clarke titled “A Concerning Trend.”
The magazine explicitly prohibits “stories written, co-written, or assisted by AI,” and Clarke said the latest deluge of machine-written submissions appeared to come from individuals outside the sci-fi and fantasy community. He blamed the flood on people trying to make money from “a side hustle” of selling AI-generated content. (The magazine pays writers a fee of between 10 and 12 cents per published word.)
The predicament follows much hype around OpenAI’s ChatGPT, an artificial intelligence technology that was released to the public in November and quickly proved surprisingly capable at a variety of tasks. It has written songs, sermons and sonnets and stoked fears of the death of the high school English essay and the demise of human creativity.
As of February, there were more than 200 books on Amazon that attributed authorship to ChatGPT, Reuters reported. Some have even started coaching aspiring authors on how to use ChatGPT as a “creative writing partner.”
Tools to detect AI-generated speech are available, but Clarke said they are “prone to false negatives and positives” and difficult to rely on. He said he has caught on to patterns that help him separate human and machine-written submissions, though he did not elaborate on his method for fear of “helping those people become less likely to be caught.”
Melissa Roemmele, a researcher at machine translation firm Language Weaver, said AI-generated text has “only recently started to superficially resemble human-written text.”
Machine-created writing and detection are “complementary challenges” — the better the text, the more challenging it is to detect — she said.
Clarke’s concerns go beyond the human-versus-machine debate. He said he is less worried that an AI-generated text is next in line for the Booker Prize and more that AI-driven spam could silence voices.
Clarkesworld has an open submission system, which makes it accessible to fledgling writers — and particularly vulnerable to a deluge. The magazine is always open to considering work and pays well, aspiring author Craig Shackleton wrote in a tweet.
Clarke was probably among the first publishers to notice the influx because he is “so on top of his submission pile,” Shackleton said.
An easy way to manage the flood would be to restrict who can submit work, but Clarke said such measures can marginalize lesser known and underrepresented writers. Requiring users to pay for submissions “sacrifices too many legit authors,” he wrote, and trying to use third-party identity-verification systems “would be the same as banning entire countries.”
Clarkesworld’s situation is not unique. Several academic journals, including Science and Nature, have instituted policies restricting the use of ChatGPT after the technology was listed as an author on papers. “Any attribution of authorship carries with it accountability for the work, and AI tools cannot take such responsibility,” Nature’s editors wrote in a post outlining their policy.
Such policies will probably become more common because more avenues to generate text via AI are on the way. Users recently started getting access to Google’s Bard and Microsoft’s Bing chatbot, while Chinese tech giant Baidu is expected to release a ChatGPT-esque bot called Ernie soon.
In the world of sci-fi publishing, a crackdown might involve shortening submission windows or considering only privately commissioned works.
“I worry that this path will lead to an increased number of barriers for new and international authors,” Clarke wrote. “Short fiction needs these people.”