Note from AWP: This article is free for everyone.
The summer I was twenty, I did an internship at a newspaper in a faraway town that had just broken its record for consecutive hundred-degree days, where I had no friends and no money and a bleak view of my chosen career. I was intensely unhappy. That’s a big reason I finally did what I’d been intending for years—grabbed a pen and set down the first few lines of a fictional story.
Something happened inside me almost immediately. A lovely blend of peace and focus slid in, as if someone had opened a grimy window in my mind, letting in fresh air and giving me a clear view of what was previously blurred. I felt entranced by what I saw there, which kept shifting and evolving. Time vanished, and I emerged with a feeling of disoriented exhilaration. The story was terrible, as I knew even then (and can confirm now), but it didn’t matter. Writing it felt like knifing into a turquoise pool.
There’s a reason for this. Neuroimaging studies have shown that moments of creativity recruit two brain networks that are usually in combat. The cognitive control network helps us focus, whereas the default mode network causes our minds to wander, but creativity seems to produce a unique situation that forces both processes into coordination. That might be one reason it can feel so difficult to enter a creative state, and why it feels so mystical and pleasant once we do.
Understanding this can help us see more clearly what our students stand to lose by offloading these moments of creativity to AI. Even if it could write their stories and poems better than them (a dubious prospect at this stage), letting it do so would deprive them of that lovely, mystical feeling, that cool turquoise swim, that place musicians disappear to when they improvise. The window in their mind remains shut, the images blurred, the air stale.
This isn’t just speculation. Researchers have been quantifying these effects recently, and I’m one of them. In the study I led at Oregon State University, we had thirty-one beginning-level fiction writers do the same creative exercise twice, once on their own, then again using AI. The primary goal was to measure how it impacted their creativity, and we found a couple interesting trends. For the less creative participants, creativity went up, while for more creative participants, it went down. But for everyone, there was a common feeling: dissatisfaction.
Participants’ explanations were different versions of the same idea—AI had kicked them out of the creative process and robbed them of what was meaningful about it. Other studies have produced similar findings, showing decreased satisfaction, ownership, and memory from participants who write with AI, among other cognitive costs. But that’s only because their participants, like ours, used AI in the most passive manner possible when left to their own devices, supplying the assignment prompt and letting AI generate their work. It doesn’t have to be that way.
After giving participants free rein in that second exercise, we provided lessons that demonstrated more active ways of using AI, and then we had the writers perform the same exercise a third time using the new techniques. This time, we saw significant jumps in creativity from both higher-level and lower-level writers. As importantly, participants from both groups reported feeling much more engaged and enthusiastic during the process. Many still expressed a strong preference for writing without AI, but all of them preferred the active techniques in the third exercise over their passive use in the second. They just hadn’t known these methods were possible until seeing our lessons.
“In our absence, students are left to fend for themselves in a wilderness of online advice from technologists, advertisers, and other mercenaries.”
That’s generally the way education works. But more than anyone else, those of us teaching fiction and poetry classes are the ones who can write just fine on our own—no AI needed, thank you. And so we’ve largely ignored an emerging frontier that desperately needs our expertise and guidance. In our absence, students are left to fend for themselves in a wilderness of online advice from technologists, advertisers, and other mercenaries.
To provide that guidance, creative writing faculty must first understand the series of cognitive phases that produce moments of creativity. The first phase is finding an opportunity for innovation or enhancement—that is, recognizing something like the setting as an opportunity for creativity in writing fiction. Next, you come up with possibilities to meet the opportunity, especially ones that are novel (rather than clichéd) and useful (rather than merely bizarre). Then you select the best option—maybe put Tiffany at the asbestos disposal site of the landfill rather than sending her to the neighborhood coffee shop. Finally, you find connections between your choice and the larger project—maybe breathing the putrid, poisoned air feels refreshing to Tiffany, compared with the atmosphere in Ron’s apartment after their fight.
AI might be best known for its speedy and tireless ability to generate options, making it ideal for the second phase of this process: brainstorming. Because it’s modeled on existing writing and generates responses by anticipating likely word sequences, however, it struggles with originality. That’s a problem if you expect it to give you solutions, but not if you’re kicking around ideas, then ricocheting off in more useful directions. Any new input can help trigger that process, including AI’s, as long as you approach it the right way.
While starting a new short story, for instance, I needed a profession for the main character’s wife. I consulted the AI platform Claude, and although I didn’t like its suggestions, my dissatisfaction gave me a clearer idea of what I was looking for—occupations that were traditionally female and involved care work. That alone was important progress. I asked Claude about those kinds of jobs, then kept explaining the flaws in its suggestions (too ordinary, too farfetched, too technological), until finally the answer occurred to me—a professional cuddler. Claude hadn’t given me the answer; it had just allowed me to explore my instincts until an answer emerged. That left me feeling energized, excited to write.
The other phase where AI excels is the first, finding opportunities for creativity—maybe not during the heat of composition, but certainly afterward. It can catch the opportunities that slipped beneath your attention or need better execution. This can be especially helpful for novice writers, who aren’t yet attuned to these opportunities, or don’t yet realize that their choices are clichés. Many beginners’ stories, for example, open by having a character wake up, smack an alarm clock, and eat breakfast. If you ask AI, “What’s one aspect of this excerpt that is bland, cliché, or unoriginal?” it’ll zero in on places like this. Ask it for more distinctive options, and you’re back in the brainstorming phase. These are the two ways we instructed students to use AI in our study.
But we also warned them that AI will never be able to spur them on this way during the third phase, choosing the best option. What’s best in these situations defies measure, explanation, even rationality. There was nothing objectively correct, or even superior, for instance, about the professional cuddler option. It was just some zing inside me, some recognition based on my interest and intuition and experience and good judgment. This apparatus operates beneath my conscious thinking, and trying to suss it out can destroy it. That leaves AI no way to perceive its contents, which is why it’s so bad in this phase, and why it can be so alienating to invite its participation.
The same is true for the fourth phase, when you integrate your choice into the story and draw connections to all the other choices you’ve made. More than any other phase, this one tends to happen on the page, when you’ve finished thinking and started writing. The associations you forge there are the crux of your creative exploration and the essence of your creative insight. That’s usually the most meaningful and enjoyable aspect of the whole enterprise—the trance, the turquoise pool—and if you’re like me, you won’t let AI anywhere near it. When writers do, they’re likely to have the same reaction our research participants reported when they let AI generate their exercises: dissatisfaction.
Letting students reflect on that dissatisfaction can be a powerful deterrent to AI use. That was the final stage of our study, and the attitudes we found from students were surprisingly conscientious and nuanced, even when AI improved their creativity. Our study was also limited in some important ways, including its small size and focus on novices. We worried that students might be tempted to off-load their writing to AI once they were using its interface, and many of the samples from the third exercise did indeed resemble AI output. These issues call for further research, oversight, or modification.
“Telling people to just say no isn’t working. We have to find better ways.”
Even so, offering students more proactive, critical techniques for using AI can help motivate them toward more responsible, transparent, and effective interactions with the technology, which is sure to dog them through courses other than creative writing and contexts other than college. ChatGPT now has more than seven hundred million weekly users, or about 10 percent of the world’s population. According to a survey from the Higher Education Policy Institute, 88 percent of undergraduate students are using AI for coursework. Telling people to just say no isn’t working. We have to find better ways.
For creative writing faculty, that might mean putting aside our preference to shoot the technology straight into the sun and meeting students where they’re at. Being honest about where AI can be helpful is a start in that direction, and it lends credibility to our warnings about its limitations, which are very real. Whatever has driven so many people to creative writing over the last two decades, AI can’t provide it. The real glory, the reason people like me keep coming back to it, is the creative act itself. Undertaking it on their own, even badly, will reward students in ways the best AI-generated content can never touch. AI can serve as a springboard for them, something to push off from and then leave behind, but they have to plunge into the water themselves.
J.T. Bushnell is a Teaching Innovation Fellow for the AI Literacy Center at Oregon State University. He coauthored “A New Muse: How Guided AI Use Impacts Creativity in Online Creative Writing Courses” as a fellow for the OSU Ecampus Research Unit and chaired the AI Advisory Committee for the School of Writing, Literature, and Film, where he has taught writing courses since 2007. His essays about writing appear in Poets & Writers, The Writer’s Chronicle, The Writer, Catapult, and Fiction Writers Review. His stories have been published in Iron Horse Literary Review, Passages North, The Mississippi Review, and other literary journals. His novel, The Step Back, was released in 2021 by Ooligan Press.