Texts that Slither
AI in Granta
Yesterday I read “The Serpent in the Grove,” one of five winning stories selected from 7,806 entries for the Commonwealth Short Story Prize. The winning stories were published in Granta, one of the most prestigious literary magazines in the world.
I read the story knowing that someone had run it through Pangram, which flagged the text as 100% AI generated.
I hated it. As I was reading, I recalled a comment one of my professors made long ago about certain theoretical texts. Not all theory (this was a professor in an English department who was well versed in literary theory), but some. She said that certain texts offer the illusion of cohesion. If you look at individual paragraphs, they are well written and seem to be making some kind of point. Transitions, too, seem smooth. One paragraph slides into the next. However, when you step back from the article, you realize that there is nothing to hold onto. The entire argument seems to unravel, or perhaps there was never anything solid beneath those phrases that so smoothly snake down the page. That is how I felt reading “The Serpent in the Grove.”
The text in question is, of course, a short story and not a work of theory. However, it still seemed to me to be full of slithering phrases that seem to signify but feel empty. The text is full of abstraction. There are concrete details, but these are overwhelmed by broad strokes, references to “patience,” “slowness,” “steadiness,” “exactness,” and “freedom.”
Some paragraphs overdo the use of simile. Take, for instance, these lines:
Sita moved quiet as if sound were taxed. Nineteen and brown like dust after rain, she turned roti dough with a rhythm that came not from joy but from endurance. Orphan was too kind a word. Orphans are sometimes cradled. Sita had been passed like a parcel from kin who were hungry for everything except another mouth; she learned to make herself small.
Three separate similes, all working with different imagery: “as if sound were taxed,” “like dust after rain,” and “like a parcel from kin.” Also, “dust after rain”? Isn’t that just called mud?
Then there are phrases that don’t make much sense: “Bush took him in – not like a mother, like a judge.” Judges don’t take people in, do they?
I read this story with a great deal of bias, however. Research shows that people tend to evaluate texts more harshly if they believe them to be AI generated. At the same time, it has become increasingly difficult to distinguish between work authored by humans and that produced by LLMs. A study by Tiffany Zhu et al. showed that readers rate AI-generated work highly if they believe it to be authored by humans. In some cases, the labeling matters more than the quality of the work.
“The Grove and the Serpent” has many features that indicate AI use: the artificiality, the not-X-but-Y construction, the nonsensical similes. Those qualities also make it poor writing. However, many of the readers pointing to those AI tells had already seen the AI-generated label from Pangram.
On X, writer Justin Murphy wrote
What if Jamir Nazir is just a sincere, passionate writer using technology to explore new approaches to literature, and Granta is now the most daring, avant-garde journal of contemporary writing? This emerging collective deference to Pangram among digitally cosmopolitan readers strikes me as more philistine than an Indian businessman trying to write poetry or tell stories with the help of his computer. Who here is really most guilty of outsourcing their taste to the machine? And if you find this man’s writing unworthy of the prize, then Granta’s mistake was trusting in a human to be the reviewer; had they employed an AI judge then his piece would have been rejected. The truth is that AI will challenge our viewpoints about the nature of high culture and high-cultural production much more deeply than people realize. Most of the emerging critical lenses shrink into irreducible contradiction if you think about them for more than a minute.
Murphy makes some good points. The harsh criticisms of “The Grove and the Serpent” are all over Bluesky, Substack, and every other platform where writers like to lurk. How many, however, would have read the story differently had someone not run it through Pangram? To what extent is technology shaping criticism, even for those of us who honed our skills long before the advent of LLMs and AI detectors?
After I read the story, I copied my most recent literary essay and pasted into Pangram: 100% human generated.
"And if you find this man’s writing unworthy of the prize, then Granta’s mistake was trusting in a human to be the reviewer; had they employed an AI judge then his piece would have been rejected".
There is a possibility Murphy has got the wrong end of the stick here. My suspicion (based on the blurb one of the judges wrote in praise of the story) is that AI was used by one or more judges to evaluate the entries, and this is how a bad, AI-written story came to win the prize. No human reader has ever bought the notion that any kind of walking has ever made benches become men, and nobody will ever make me believe otherwise.
Well done. These are fascinating questions. I completely agree that our reliance on AI to detect AI is one of the perverse ironies of the moment. I think Murphy is wrong about one thing, though: AI judges will award prizes to AI writing at a higher rate than human writing, and at an even higher rate to write written by the same AI model they use!