‘Polished and confident, with a melodic voice that lingers long after the final line, Jamir Nazir’s prose pulses with a voice of restraint and quiet authority.’ So said Sharma Taylor, regional judge for the Caribbean winner of this year’s Commonwealth Short Story Prize, The Serpent in the Grove.
AI is very good at producing, at scale, exactly the sort of stuff that such critics will affect to like. Portentous cadences, look-at-me adjectives, solemnly poetic similes, the sort of thing that people produce when they’re trying to produce ‘fine writing’
But did it? Did it pulse with a voice of restraint and quiet authority? Really? On Monday, Nabeel Qureshi, who describes himself as an ‘entrepreneur writer, and researcher’, declared on X: ‘Well, this is a first: a ChatGPT-generated story won a prestigious literary prize (The Commonwealth Prize). ‘Not X, not Y, but Z’ sentences everywhere, the ‘hums’ trope, and plenty of other obvious markers of AI writing.’
Ethan Mollick, a professor at Penn University, chimed in on Bluesky: ‘In a Turing Test of sorts, it looks like a 100 per cent AI generated story just won the Commonwealth Prize for the Caribbean region.’ He added: ‘Pangram flags at 100 per cent but also, come on, if you know you know.’
At once, social media did its thing and we were away to the races. Hot takes spewed furiously from people who had previously been uninterested in, or entirely unaware of the existence of the Commonwealth Short Story Prize. A couple of hunches – ‘come on, if you know you know’ not being the standard of proof that most of us would recognise as a slam-dunk – were laundered into certainty. I saw Goody Proctor with the devil!
The story itself is not much good, I think. I could happily go a long time without reading such stuff as ‘Men who set traps plan for silence, not for the squeal. Shame is a substance he felt on his skin. It itches. It doesn’t rinse.’ or ‘First good rain after dry is a forgiveness the sky gives itself.’ Or, ‘One day – the hum loud as if noon had tuned itself – he brings his daughter. Her hair is midnight rain; her laugh is bright as zinc.’ But, y’know, if you like that sort of thing there’s a lot of it out there and most of it has been produced by humans.
So – and I hate to spoil a good panic – we don’t know for sure that this story was written by AI. Unless its author makes a confession, we probably never will. (And if speculation that his author photograph is itself AI-generated is true, he’s going to struggle to because he doesn’t exist.) A couple of blokes on the internet may have strong intuitions, and amid current AI paranoia that’s enough to cause a stampede, but that’s a little beside the point. Any news story where a discovery is credited to ‘internet sleuths’ is to be treated with scepticism.
The human response, though, is telling about that atmosphere of paranoia. Granta, which published the winning story on its website, issued a statement in the name of its publisher Sigrid Rausing, which opened: ‘For more than ten years Granta has published the winners of the Commonwealth Prize on our website. Granta editors have no control over the selection of the Commonwealth Prize stories, and nor are they involved in choosing the jury.’ That is: nothing to do with us, guv. Whether from principle or under the advice of lawyers, she concluded: ‘until the Commonwealth Foundation comes to a definite conclusion, we will keep these stories on our website’.
The organisers of the Commonwealth Prize, meanwhile, take the ‘what can you do?’ line in their own statement. There’s a bit of guff reassuring readers that ‘our judging process is robust’ and describing ‘multiple rounds of readers’, before they get to the nitty gritty: ‘Until a sufficient tool or process to reliably detect the use of AI emerges that can also grapple with the challenges pertaining to working with unpublished fiction [no, me neither], the Foundation and the Commonwealth Short Story Prize must operate on the principle of trust.’
And, wet though it may be, it’s probably the only conclusion that can be reached. This problem isn’t going to go away. ‘Robust’ judging processes still rely on critics, and very few fiction critics are any good at all. Only a tiny handful are excellent. Lots of people make careers in the literary world without being able to tell a good bit of writing from a bad one, still less to articulate the difference. They can, rather, recognise the kind of material that that they’ve been told by the ambient culture of the day passes as literary prose – and when they see it, they applaud it in vague and formulaic terms. Pulsing with quiet authority. That sort of thing.
AI is very good at producing, at scale, exactly the sort of stuff that such critics will affect to like. It produces, when you ask it to, material with the superficial texture of literary writing: portentous cadences, look-at-me adjectives, solemnly poetic similes, the sort of thing that people produce when they’re trying to produce ‘fine writing’. But producing bad fine writing is what bad literary writers have been doing for ages and what bad literary critics have been applauding for ages too.
The problem is intractable. AI detectors are not a means of establishing AI use to anything like the standard of certainty that would survive legal challenge. As has been exhaustively documented, they throw out false positives left, right and centre. And (these things being themselves AI) their workings are opaque and their findings no more subject to interrogation than the findings of a Magic Eight Ball.
The human eye? Some of us can spot, or think we can spot, AI slop when we see it. But it is complacent and self-regarding to suppose that, as the technology improves, this will be true in a year, or five years’, or ten years’ time even if it’s true now, which it probably isn’t. The features that people currently identify as ‘AI tells’ –antithesis, tricolon, and so on – are features of rhetorical structure that have been in place since Cicero.
All those gimmicky schemes publishers have hit on to certify ‘human-produced’ material, whatever branding and institutional weight might be behind them, mean nothing. They add up to nothing more probative than having writers declare: ’Scout’s honour’ and slapping a sticker on the front of a paperback. Operating on trust, in fact.
And trust is precisely what generative AI is attacking like the literary equivalent of necrotising fasciitis. The serpent, in other words, is well and truly in the grove.
Comments