Sci-fi magazine has to halt submissions after receiving too much AI-generated fiction

In a Q&A, Clarkesworld founder Neil Clarke discusses his publication's futuristic dilemma
The robots want to write sci-fi now too
The robots want to write sci-fi now too
Photo: Frank Barratt (Getty Images)
We may earn a commission from links on this page.

Neil Clarke is one of the foremost editors of science fiction—and now he’s faced with a very sci-fi problem.

The monthly literary magazine he founded in 2006, Clarkesworld, publishes user-submitted sci-fi and fantasy. Since December, though, Clarke has been overwhelmed with entries that clearly have been written by tools that use GPT-3, machine language software developed by the company OpenAI.

The uptick coincided with the release of ChatGPT, the chatbot released by OpenAI in November. ChatGPT and tools like the digital image creator DALL-E have reignited public discourse about how humans and technology interact, as well as the ethics of using AI in school homework, journalism, and even professional art.

For Clarke, the problem is so severe that he has had to halt open submissions to his magazine, which typically garners 1,100 submissions per month. He had to ban more than 500 people from his submission platform in February alone, pointing toward an escalating trend.

Clarke spoke with Quartz about the emergence of generative AI and how he’s trying to root out the problem at Clarkesworld and reopen submissions.

This interview has been edited for clarity and length.

When did you start noticing that your inbox was being flooded by machine-written submissions? 

Back in the early days of the pandemic, we started seeing some oddities with plagiarism. People were using AI-ish programs to rewrite other people’s stuff to mask that it was changed. We were seeing an uptick towards the end of last year right before ChatGPT was announced. Things picked up in December, in January it picked up a lot more, and now February has just gone through the roof.

There are a lot of anti-plagiarism tools out there, but none of the anti-GPT stuff is any good. We’ve tried all of the ones that we’re aware of and they’re easily outwitted. It was getting to the point where it was interfering with our ability to process the stuff that we actually asked for. So closing submissions was the only path we could take.

What are these oddities you’re noticing? What are the indicators that this fiction isn’t written by humans?

Well, we’re not spelling out exactly what it is that we’re noticing about these submissions because they’re things that could be easily fixed—and we don’t want to make it any easier for them, or make it more difficult for us. But these are things that anybody who’s been reading submissions for any amount of time would look at and go, “This doesn’t make sense.” There are certain types of mistakes that are not made by humans, let alone at the volume with which we’re seeing it happen. A lot of these AI submissions are really beyond bad—I wish I could give you exact examples, because it’d be immediately clear to you.

I’m imagining problems with syntax, or weird capitalization, or—

But AI doesn’t do those things! It can make clunky sentences, but it doesn’t mess up things like spelling. There are other things. I have access to all the data behind every submission, and I know that there are three countries, for example, that make up the lion’s share of fake submissions. But we also have legitimate authors coming from those countries, so I am very careful about not saying the names of those three countries because I do not want to poison the pool.

Do you think that the use of this technology is an act of plagiarism, since it’s all likely derived from other people’s words?

We don’t know exactly how this stuff is being generated, but we know that with the AI art programs, there is almost certainly stealing. Some of the things that have come out are ridiculously close to people’s work. And I believe these companies are actively being sued. I feel like it’s suspect, which is why we will not accept submissions that have it. From an ethical standpoint, we’re not ready to say, “Hey, this is totally acceptable,” because we don’t understand where it’s coming from. We amended our contracts in November to ban the use of generative AI.

Do you think AI writing tools can serve a valid, assistive purpose?

In the future, possibly. But we have to make it clear that this is something that cannot be swept under the rug. If someone was working with a human, they would credit them a co-author, but if they’re working with an AI then they’re the only author. On the other hand, if I come up with character names using AI, that’s no different than putting your finger in a phonebook. We wouldn’t credit the phonebook, so we wouldn’t credit an AI for picking a name. It’s still a very gray area, but we want to stay very much on this side of the line until the industry has had more of a chance to figure it out.

You wrote in a blog post, “If the field can’t find a way to address this situation, things will begin to break.” What do you mean by that?

In science fiction and fantasy, short fiction is the heart of the field. It’s where a lot of the new ideas and new voices come from. That requires an openness to new people. And a lot of the solutions that I’ve seen proposed to deal with our problem involve shortchanging the people that are really going to move the needle. So if we said, “Forget it, we’re not going to do submissions anymore, we’re just going to talk to authors we already know,” then we’re going to stagnate. When you don’t have new voices coming in, it’s harmful to the field. The long-term health of the field is determined by whether or not we bring in new blood.

What are you doing in the immediate future to address the problem?

Some of our readers have offered to help—they work in different fields that are tangential to this issue, such as fraud prevention and AI development. We will probably make some adjustments to our submission software, reopen it, see what happens. A lot of the people who are submitting this stuff have blocked cookies, or turned off JavaScript, or done something to mask who they are. So if we require those things, we might find that they don’t want to bother with us. There’s lots of little experiments we can do.

We tend to pay better than most, and we’ve been around for a while, so we’ve made it onto certain lists that are being used by get-rich-quick people. My big hope is that either ChatGPT starts charging people, which would shut this down instantly, or these get-rich-quick people move on to a different side-hustle.

This is a very science fiction-y problem. You’ve probably read and edited more sci-fi than I will ever encounter in my life. Is there anything from the literature that comes to mind?

You know, I couldn’t really come up with any scenario where this lined up. I don’t think anyone anticipated using the robots to make art. This is one of those funny situations where everybody’s going, “Hey, should somebody have written about this?” And, nope—no one did. Maybe we’ll figure it out in like three months, some vintage story from the 1940s that was tucked away in the back of the back of some low-circulation magazine. Somebody will finally find it.

I’ve been thinking about Issac Asimov’s three laws of robotics, under which the robots of his fiction are programmed to avoid harming humans above all else. Right now, we’re pretty far from worrying about robots trying to kill us.

That’s the thing! They’re not trying to kill us. They’re just inadvertently burying us in all of their junk. I don’t think an AI is ever going to necessarily be any better than our best writer. The problem is not the quality. The problem is how quickly it can produce. It’s drowning out the other works, just by volume. That’s really what I see as the core issue we’re experiencing right now. It’s the quantity problem.