Clarkesworld, a popular fantasy and science fiction magazine, made headlines last year when it was forced to stop accepting stories from freelance writers because so many submissions were generated using artificial intelligence.
Clarkesworld’s publisher, Neil Clarke, was recently a guest of honor at the 46th-annual Norwescon – the Pacific Northwest’s premier science fiction and fantasy convention. In an interview at the event, held in the city of SeaTac, he was asked how he felt about the creation of AI, in particular the wildly popular tool, ChatGPT.
“As a technology person, I think what they’ve done is irresponsible,” he answered. “I think it is unethical. They have stolen work from people I respect – people I don’t respect. It doesn’t matter who.”
Artificial intelligence refers to the use of computer systems, software, and processors that collect and analyze vast amounts of data to identify patterns and make predictions that are typically associated with human intelligence. Generative AI – such as ChatGPT – draws on this rapidly-growing capability to produce content using a vast universe of text, images, video, and audio.
For example, a user can query ChatGPT simply to answer a question, or to synthesize the entire contents of a book, or to create a wholly “new” book based on content written by others. The responses often read naturally, as if they were written by a human.
Not surprisingly, writers, artists, musicians, and other creators of intellectual property are concerned about what this means for their future. There are few guardrails to ensure that original creative works are protected from misuse or outright theft by others.
In the case of Clarkesworld, a large number of freelancers hoping to get a story published simply pasted the magazine’s detailed story guidelines into ChatGPT; instructions such as, “Fantasy can be folkloric, contemporary, surreal” but zombies, sexy vampires, wanton werewolves, wicked witches, and demonic children are “hard sells.”
With about half of the magazine’s 1,500 to 2,500 monthly submissions being generated by AI, it is no wonder Clarke would “scream at the ceiling” every time he sees one, as he said.
Clarke eventually reopened submissions after finding a way to identify stories generated by AI; a method he is not revealing. Doing so would make it easier for potential offenders to game the system.
Clarke did explain one way the AI-generated stories stand out. “I describe them as having been bad in ways that no human has been bad before. They are boring. They are flat.” He says that because AI is a statistical model that only predicts the most likely next word, generated stories have no subtext or layers. But he has noticed they’re getting better with recent updates to ChatGPT.
Which begs the question; how long will it be until AI does match the capability of a human writer?
University of Washington Assistant Professor Léonard Boussioux, who is researching the collaboration between humans and AI, along with how AI affects the way we work, said in a recent interview: “To get something exceptional, I think we will have to wait at least a couple years, which may sound like not much. But it depends on what it is about.”
The AI Artist
Boussioux teaches information systems and operations management at UW’s Foster School of Business. He also holds an adjunct position in computer science at the Allen School of Computer Science and Engineering, and an affiliate faculty position at the Laboratory for Innovation Science at Harvard.
Boussioux personally uses AI as a creative outlet: “I have loved the advent of generative AI because there are so many things I wish I could show the world, but I don’t have the time or the skills to draw properly. And so, I leveraged the tools [of AI] to communicate things I could never show before.”
Boussioux displays some of his AI creations in his office. a simple one-sentence prompts that lead to a picture of a sci-fi city or a fantasy landscape.
Others take months. He has an idea in his head of an image he wants to produce. He’ll enter a prompt into the AI tool to see what it comes up with. Based on that, he refines the words in the prompt until an image matches exactly what he sees in his head. In one instance, he generated more than 5,000 images before he found exactly the right one. He calls this “the art of curation.”
As much as he enjoys the creativity of the process, Boussioux is concerned about the misuse of AI, so he personally does not sell his artwork. He produces it for himself and for educational purposes.
“I would not feel happy knowing that my piece, in a way, stole someone else’s art if I were to benefit from it,” Boussioux said. “However, I feel okay leveraging what they have done, and then I credit them, for it was inspired from this style. And I love doing this because I feel like I continue their artwork – I continue the journey.”
AI Detection Software
For Neil Clarke and Clarkesworld, the challenge is the avalanche of stories they still have to dig through to find the ones that are authentic and worth publishing.
“Imagine if you had to read your email and spam filters didn’t exist,” he said.
Clarke wants stories from everybody and is proud to rely on open submissions. He believes that is how you get the best stories, but at the moment it also allows AI-generated stories to flood the market.
Clarke and Boussioux agree that current AI detection software does not work very well. Savvy AI-users can create prompts that work around it. In other words, people find ways to game the system.
Boussioux is working on software that he hopes will enable users to evaluate the quality of an idea in an unbiased way. For example, he is having his students evaluate ideas for startup companies, and then using his tool to do the same.
If it works, the tool could be used as an alternative to AI detection by sifting out works of poor quality.
“We’re going to learn how we use AI to evaluate,” he said, “It’s becoming easier to just get some average ideas but we’re looking for the exceptional ones. How to detect an exceptional idea is challenging.”
Local Publishers
AI does not seem to be a problem for some other magazines and journals that rely on freelance submissions. For example, Image Journal is a Seattle-based literary magazine that publishes works related to art and faith.
Image’s publisher and executive director, Sara Arrigoni, said: “[We have] a fairly intimate community of contributors, many of whom repeatedly submit to us and have for years. I think because of this, so far our editors are not seeing work they suspect or know is AI-generated.”
Omar Willey, publisher of The Seattle Star, an online journal of politics and culture, called his journal a “poor target for AI. Most of the work I get is from already published authors, and those new voices who magically find us tend to know exactly who we are and what we do.”
The Future of Art
The way Léonard Boussioux talks about his creation process, it almost feels like you could replace ‘AI’ with ‘paintbrush’ and it would not be so controversial. Could AI, at some point, be considered just another tool for creation?
Boussioux described how many art forms coexist, evolve and are seen as legitimate: “Look at photography. Before, there was only painting. And now, we also have digital art.”
Could there come a day when authors who type manuscripts and authors who use AI are considered on equal footing? To be successful, AI authors will have to develop appropriate prompts, iterate their ideas, and put time into directing the AI to create the layers and subtext readers look for in a good piece of writing. If the content is truly original and the ownership legitimate, AI may be seen as just another tool to create.
That’s how Boussioux sees it. “My real philosophy about AI is that it’s a magical tool, but it’s a tool. I’m always the one driving it and I’m always trying to be responsible.”
Not everyone is so conscientious. Artists, in particular, are worried about AI upending how art is created. There is a possible future where AI makes creativity too easy and completely dismantles the world of writing, art, and other content creation.
Neil Clarke described what would happen if AI caused his piece of the publishing business to collapse.
“When short fiction falls, that’s a domino that cascades, because short fiction is a lab. That’s where a lot of the new authors – a lot of the new ideas – gets tested in that space… A lot of movies get made off short fiction. A lot of novel deals end up coming off of short fiction. A lot of careers are started there.”
AI is still in its infancy but things are moving fast. As Boussioux said, it may only be a couple of years before the technology is – well – beyond our imagination.
Interesting article, Taylor. I think what makes the issue so dispiriting is looking at what’s happened in creative sectors even before AI. This tells us all we to know about how a new “tool” will be used, who will make money from wielding said tool, etc.
As for literary magazine submissions, as someone (painfully) familiar with that world, it’s incredibly competitive and far from a cake walk in the first place. Submitting AI work is so far from the creative, grassroots spirit of literary mags, I’d hope anyone going that route simply loses interest and moves onto other hobbies. Not like there’s much glory in it anyhow.
Cheers,
Eric