Sci-fi magazine overwhelmed by hundreds of AI-generated stories

Science fiction magazine Clarkesworld has halted story submissions after receiving a growing deluge of AI-generated pieces. The magazine’s founding editor, Neil Clarke, says the problem has been created by people promoting surprisingly capable AI language models such as ChatGPT as a way to earn money from fiction publishing – despite the poor quality of the AI stories.

“The machine-written submissions we’ve received are far from publishable quality,” he says. “I’m sure there are some that are less detectable, but the majority we’ve received have been easy for me to identify.”

Clarke says that he has talked to other magazine editors who currently have the same problem, although he says they have been reluctant to speak to the press – as he was, until the problem grew to unsustainable levels.

The magazine normally has an open submission policy to encourage new writers, but took the decision to close submissions on 20 February after receiving 50 AI-generated story submissions that day. At that point, Clarkesworld had received 700 legitimate submissions since the start of the month and 500 machine-generated ones, with the rate of increase meaning that AI-generated stories would soon take over.

Clarke says the increase in “spammy” submissions began towards the end of 2022, the same time that accessible large language AI models were launched, and have increased month on month. “[Those submitting AI-generated stories] don’t care about their reputation in the field. That makes it more like malware or credit card fraud and faces similar challenges. We’ll try to minimise these instances and they will try to get around it,” he says.

Not all fee-paying science fiction magazines have seen the same problem, however. Djibril al-Ayad, editor at The Future Fire, says he has noticed no spike in spam submissions, but that may be down to being a smaller title. “I suppose the problem is if your submissions pile triples in size and all the AI stuff is abysmal but still has to be read, then it becomes a kind of denial-of-service attack situation,” he says, referring to attackers who knock servers offline by submitting huge numbers of requests for data.

“My take would be not to be particularly worried about the danger of accidentally publishing AI-generated fiction,” he says, as most will be bad. “If it isn’t, then well… cool.”