The scandal started the usual way. Readers noticed AI-like prose in a written work and took to ridiculing it online. Some ran the writing through an AI-detection platform that labeled it entirely AI-generated. The institutions involved in its publication scrambled to figure out what had happened.
The details in this particular scandal have to do with an all-but-unknown Trinidadian writer named Jamir Nazir. His story “The Serpent in the Grove” was among five regional winners of the Commonwealth Short Story Prize. The award came with 2,500 British pounds and publication on the website of Granta, a prestigious British literary magazine. Earlier this week, readers started gleefully tearing Nazir’s work apart online, posting screenshots showing canned stylistic patterns and a proliferation of weird metaphors: “Her hair is midnight rain; her laugh is bright as zinc,” read one line.
“A major milestone for AI, at any rate,” one person deadpanned. Subsequent sleuthing only reinforced the early suspicions: The photo of Nazir on the prize website was almost too slick-looking; his LinkedIn page was filled with florid posts about AI’s potential to change the world.
Before long, commenters were pointing fingers at two other winners of this year’s Commonwealth Prize: Malta’s John Edward DeMicoli and India’s Sharon Aruparayil. People posted screenshots from the same AI-detection platform; it flagged both stories as likely to have been generated using AI, DeMicoli’s in full and Aruparayil’s in part.
DeMicoli’s online footprint was minimal before his win and the subsequent scandal. But Aruparayil works in communications and, like Nazir, has posted about AI—at times using language that only a chatbot would appreciate. “I envision a future where decision-making is a seamless synergy between human expertise and artificial intelligence,” reads a blog post published with her byline.
[Read: How AI is creeping into The New York Times]
I reached out to all three authors using contact information I’d found online; only Aruparayil responded. She told me over email that she hadn’t used any AI tools at any point “in the writing, editing, or development process” for her prize-winning work. “The story has had only human hands and eyes on it, and I refuse to use AI in my writing,” Aruparayil said. She added that she had saved several time-stamped drafts, evidence of her active role in writing and editing the story, but she declined to share them. When I asked about her AI-promoting blog post, she said that she hadn’t written it. Rather, she said, an Emirati research foundation had attributed the post to her based on a project she’d done for it. The post has since been taken down; Aruparayil said that she’d requested the removal. (The foundation didn’t immediately respond to a request for comment.)
Much of the coverage of this latest scandal has focused on the possibility that two prestigious organizations unknowingly published AI-generated work. But that part shouldn’t be shocking. AI has also shown up in outlets including The New York Times and in books published by major houses. What’s different, this time, is what happened next.
In previous instances of suspected AI use, the authors quickly conceded that artificial intelligence had been involved. I wrote in March about a “Modern Love” column in the Times suspected of including AI material. Its author, Kate Gilgan, acknowledged to me that she’d turned to at least five AI products for “inspiration and guidance and correction”—in short, as a “collaborative editor.” Later, the Times introduced AI guidance banning freelancers from using AI in that way. Around the same time, the author of the horror novel Shy Girl—which readers called out for various AI tells after its U.K. publication—said that an editor she’d hired to help with an earlier, self-published version of the novel had used AI. Hachette, the novel’s publisher, discontinued the novel’s U.K. edition and canceled publication in the United States.
The current controversy is already playing out differently. Other than Aruparayil, none of the authors involved has spoken publicly this time. Razmi Farook, the director general of the Commonwealth Foundation, released a circumspect statement this week noting that “all shortlisted writers have personally stated that no AI was used and, upon further consultation, the Foundation has confirmed this.”
Yesterday morning, Farook clarified in a call what this meant: The foundation had asked the winners to confirm again that they hadn’t used AI, and the writers had obliged. Farook said that the Commonwealth Foundation hadn’t used AI detectors, because they’re fallible and because doing so would involve inputting authors’ work into an AI product without their permission. Farook said that she believes the prizewinners’ promises: “We feel very responsible for making sure they’re cared for and protected,” she said. She also acknowledged a pragmatic dimension: “Our legal parameters don’t allow us to contest the honesty of our writers just because a tool says that.”
[Read: The human skill that eludes AI]
In response to an email query, Granta’s editor, Thomas Meaney, wrote to me, “I am aware of this and we have been looking into it.” Later, Granta released a statement that was difficult to parse. Its publisher, Sigrid Rausing, said that the staff had asked the AI chatbot Claude about Nazir’s piece and that the chatbot had concluded that the story was “almost certainly not produced unaided by a human.” (Talk about bad AI writing.) This seems to mean that Claude suspected that AI had been substantially used, though it added that the work contained a “human core.”
The statement only further fueled online ridicule. Claude is a general chatbot, not a tool designed for AI detection; if purpose-built AI detectors can make mistakes in flagging AI prose, Claude could be expected to perform even more poorly. At the end of her statement, Rausing declined to render a verdict: “It may be that the judges have now awarded a prize to an instance of AI plagiarism—we don’t yet know, and perhaps we never will know.”
Where does all this confusion come from? A couple of possibilities seem worth considering. One is that the readers, and the tools they use, might have simply gotten it wrong this time. A much-cited Stanford paper published three years ago found that AI detectors had a higher false-positive rate for text written by non-native-English writers than they did for text from native English authors. Because the Commonwealth Short Story Prize is awarded to authors from all over the Commonwealth—an association of 56 countries—readers and AI detectors could be tripped up by language that is written by non-native-English authors or that deviates from American or British norms.
The Commonwealth Prize archives offer a useful data set for informally testing this theory. Since its launch, in 2012, the prize has been awarded to dozens of writers from all over the world. Pangram, the platform that detected AI material in the three prizewinners this year, is considered to be among the more accurate AI detectors. I asked Jenna Russell, a doctoral candidate at the University of Maryland at College Park and a research scientist at Pangram, to run stories from the past 15 years of the Commonwealth Short Story Prize through the platform.
She found that Pangram flagged almost none of the prizewinners. The exceptions included the three stories from this year: 100 percent of the text in Nazir’s and DeMecoli’s stories was flagged as likely to have been entirely AI-generated, along with 89 percent of the text in Aruparayil’s. There was also a fourth story from last year, by the Vincentian Canadian writer Chanel Sutherland, for which 88 percent of the text was flagged. (Sutherland didn’t respond to a request for comment sent through her website.)
[Read: The literary job AI can’t replace]
Unless those results are fatally flawed, which is not impossible in this early phase of AI detection, they point to another possible explanation for the prizewinning authors’ categorical denials. Knowing that detection platforms are fallible—proving AI use isn’t as simple as proving, say, plagiarism from another author’s work—writers could be discovering an enforcement loophole. As Farook explained, revoking a prize without proof is, morally and legally, no simple matter.
I pointed out to Farook that prizewinning stories full of AI-like style—the em dashes, the bad metaphors, the details grouped in triads, the not-X-but-Y sentence constructions—are probably undesirable whether they’re proved to be AI-generated or not. She acknowledged that judges might benefit from training in identifying those stylistic quirks. She added that, as AI detectors improve, using them in the judging process could become possible—though only with informed consent from those who submit stories. Still, none of that would definitively root out AI. Farook said that the foundation has convened a panel to “review the risks” related to AI. In the meantime, organizations would set a bad precedent by responding rashly to even reasonable suspicions.
Lucky for us, then, that the reading public doesn’t share that predicament. People on social media are free to fling whatever accusations they feel like flinging. Some will be reckless; others could turn out to be more discerning than prize committees, whose members might have far less experience than the highly online in sorting the real from the slop.
Aruparayil and Farook both told me they worry that online hunches could ruin an emerging writer’s reputation before they even get their career going. They’re not wrong. But critics have made a sport of skewering other people’s writing for generations, and fiction published under one of the most prestigious brands in literature should be fair game. What’s so harmful about applying literary judgment to expose writing that sounds machine-made? Maybe an author did use AI. Maybe their consciousness was just so influenced by AI that they started imitating it. In either case, a little public shaming might be warranted.
In their book, The AI Con: How to Fight Big Tech’s Hype and Create the Future We Want, the linguist Emily M. Bender and the sociologist Alex Hanna encourage people to resist AI by making fun of it. “Ridicule as praxis,” they call it. It will likely get only harder to definitively prove AI use in writing as would-be authors get more sophisticated about it, purging the obvious AI tells from their AI-generated drafts or even training AI models to imitate literary authors. Pointing out the ridiculousness of derivative, soulless writing—AI-generated or not—might deter writers from interacting with AI. It also has the added benefit of maintaining high collective standards for what real literature is, at a time when so much of our language has been colonized by algorithms and the powerful companies behind them.
Nazir still hasn’t commented publicly on the accusations involving his work. But his LinkedIn page offers a revealing look at his personal preoccupations. “The dominant anxiety is whether AI will replace jobs,” he writes in one post. “The real risk—and the one few are discussing—is that it will amplify bad judgment at a scale we have never seen before.”
The post This Literary AI Scandal Changes Everything appeared first on The Atlantic.
