Permit me, briefly, to assemble the evidence for the prosecution. The moment LLMs became capable of extruding clean, intelligible prose around late 2022, it was only a matter of time before people started submitting AI-generated stories to literary competitions. And no sooner had a “The Serpent in the Grove” by Trinidadian writer Jamir Nazir been announced as the winner of the Commonwealth Short Story Prize, than I, and thousands of other people who visited the page devoted to it on the Granta website, began to wonder whether the machines had scored their first victory.
Exhibit A: the quiet humming. From the story’s first sentence, “They say the grove still hums at noon,” to its last, “And now, at noon, when the wind turns kind, the hum sounds less like hunger – and more like the earth clearing its throat to speak the names of those who came back,” there is scarcely a sentence of “The Serpent in the Grove” that goes by without something softly humming, or quietly moaning, or sitting hushed in the stillness, or holding its silent breath. As any hardened AI user will tell you, this obsession with quietness is a tell-tale what you get when you try to programme an LLM to respond to keywords like “literary” with some kind of subtlety: a catastrophic confusion of content with form, culminating in an endless, tone-deaf bellowing about how quiet and subtle the whole scene is.
Exhibit B: the weird patterns of negation and assertion. By now, everyone knows that LLMs endlessly spit out the formula “It’s not X — it’s Y,” and Nazir’s story certainly does not skimp in that department (“The grove isn’t a ledger; it’s a mouth. It closes only when it’s satisfied.”). But what is less well known is the fact when you ask ChatGPT to write something “literary” or “sophisticated”, it tends to modify the structure slightly. New formulae creep in: “Not X. Not Y. Just Z,” and “No X, no, Y, no Z — only…” where X, Y, Z and everything are all florid metaphors that don’t really make sense. Ingeniously, Nazir somehow manages to combine this both technique and quiet humming: “No fan, no bulb, no hum – only the thin light slipping between warped boards and the breath of hills holding their heat like a secret.”
There were a few candidates for Exhibit C. I was tempted to cite the fact that there are no real characters in “The Serpent in the Grove”, or the fact that none of the endless similes tend to bear any relation to something a human being might perceive (“She had the kind of walking that made benches become men”). This, I believe, is at the root of the total, po-faced failure of irony which characterises all LLM writing – and which “The Serpent in the Grove” displays almost paradigmatically. But in the end, I cannot help but go for the obvious one: the fact that Jamir Nazir’s portrait, proudly displayed on the Commonwealth Prize’s Instagram page, looks to have been generated by AI, too.
The Commonwealth Prize committee’s stance on the whole debacle remains reasonably defiant. “We take these claims seriously and are committed to responding to them with care and transparency,” read yesterday’s statement, before reiterating that “Our judging process is robust.” Sigrid Rausing, the publisher of Granta, displayed less confidence, first pointing out, quite fairly, that Granta itself has no role in judging the prizes, and then by making the curious claim that “we showed Claude.ai the story and asked whether it was AI-generated”. According to Claude, while the story was “almost certainly not produced unaided by a human”, there are nevertheless several passages that “carry the kind of off-shape specificity that models still struggle to produce unprompted”.
I suspect there is some truth to this. After all, many of the sentences in “The Serpent in the Grove” are simply too bafflingly opaque to have been written by AI – which tends, for the most part, towards a kind of scrupulous grammatical correctness. Would ChatGPT write a sentence like “Sun on galvanise is a cruel instrument,” for example? Or “She looked at the plank mouth and the cutlass leaning casual, as if it had legs”? What is more, there is growing evidence to suggest that even those parts of the story that do exhibit the signs of AI writing might actually be produced spontaneously by human beings. A 2024 paper from the Max Planck Institute, for instance, found that since the appearance of ChatGPT, giveaway-words like “delve”, “commendable”, “meticulous”, and “realm” have exploded not just in published research, but in actual spoken, conversation. There is a distinct possibility, in other words, that Jamir Nazir is simply a slop-addled Don Quixote, who has spent so much time on Claude and Gemini and ChatGPT that their weird, inhuman rhetoric has lodged itself into his brain.
But how convincing a defence of Nazir, or the Commonwealth Prize, actually is this? The endless litigation of who or what wrote “The Serpent in the Grove” surely misses the point. The realproblem with the story is not that it was generated with AI, but that it reads like it was generated by AI; it shares the machines’ penchant for wooden rhetoric and tin-eared similes statistically improbable occurrences of the word “hush”. The fact that it won a competition – and was praised by the judge, Sharma Taylor, as “puls[ing] with a voice of restraint and quiet authority” – suggests a deficiency among humans. The UK’s literary establishment, it seems, has forgotten what makes writing good.
Indeed, the fact that LLMs have got quite good at impressing the kind of people who award literary prizes is pretty helpful to the cultural critic, since it indicates the very specific types of confusion that have come to characterise the world of books and publishing. Take, for instance, the form-content distinction. Every user of LLMs knows that the machines have a form-content problem: ask Claude to write something subtle, and it writes five hundred words on how quietly everything is humming; ask ChatGPT to write something that might make you laugh, and it’s liable to produce a long, rambling story in which literally every character ends up being tickled. This is because LLMs have no way of distinguishing between what philosophers of language call different linguistic functions: asserting, warning, promising, commanding, persuading, seducing, are all flattened to a single optimisation objective.
I suspect that many of the UK’s literary tastemakers and gatekeepers were displaying precisely this kind form-content confusion long before ChatGPT hummed its first quiet breath. In both non-fiction and fiction, books are increasingly sold and commissioned on the basis of what they are about, as though the only thing that might make a book interesting were its subject matter. I know one writer, who, upon handing in her carefully crafted period drama set in France, was asked if she would consider writing a gay romance loosely based on the ice-hockey TV show Heated Rivalry instead. I know several who have been told that, while the publishing house to which they have submitted loves their manuscript, they simply cannot find room for it, because they have already done a book featuring a protagonist of the main character’s ethnicity or sexual orientation. In the lightless depths of genre fiction, I am told, things are even worse: it is not uncommon for readers to order what content they want from a menu of tropes – friends to lovers, enemies to lovers, love triangle redemption arc, morally grey love interest, “there was only one bed”. Millions of copies can be sold, provided the writer obey one single commandment: thou shalt not make it new.
It is ironic that many of the most strident voices calling for the defenestration of Jamir Nazir, not to mention, in some cases, Granta’s entire editorial staff, seem to be reproducing precisely the same patterns of confusion in a different key. One of the main lines of censure online seems to be to feed “The Serpent in the Grove” into an online AI detector called Pangram, and then crowingly post the “100 per cent AI-generated” score it inevitably churns out. In other words, the most popular way of criticising Nazir for outsourcing his faculty of judgement to the machine seems to be to outsourcing one’s own faculty of judgement to the machine. In time, one suspects, the stamp of approval offered by software like Pangram will become just the latest proxy used by the industry to determine the quality of a work without getting involved in tricky formal matters. Already, the future is being adumbrated by the mob: a book industry that arbitrarily champions certain works on the basis of their artisanal, “human” provenance; a generation of publishers with the same relationship to works of fiction, roughly speaking, as the French Ministry of Agriculture has to certain types of cheese.
Most writers, of course, tend be pretty resilient to all this. After all, if you spend all day, every day moving bits of punctuation around on a word processing document, you tend, eventually, to acquire the conviction that form, rhetoric and syntax remain rather important. The problem, however, is that more and more literary prizes seem to be drawing their illustrious judiciaries from outside the writing profession. Judging panels are becoming an endless cavalcade of TV personalities, influencers, amateur aficionados of every stripe; after all writers tend to be a clammy, introverted, lot, and literary prizes – which often operate on the slimmest of margins – often have little choice but to invite more popular professions into the guild in order to attract publicity.
The good news, I suppose, is that if present trends really do continue, the gulf between the average writer’s command of form and rhetoric and that of the average person will soon be non-existent. Anyone will be able to have a literary career: you will simply have to plug in your demographic and educational credentials into the interface, go off and make a cup of tea, and by the time you have returned your chosen LLM will have extruded a nice, clean product that you can hand over to your publisher. You may well be able to make a living this way, provided you get the LLM to craft an accompanying “No AI was used in the making of…” assurance to aid in its marketing. You will almost certainly be shortlisted for a few literary prizes. But something will have been lost. One fine morning, you will wake up, open your laptop, and realise – with a quiet hum of horror, a hushed breath of silence – that you have become a machine, too.
[Further reading: AI will dissolve civilisation as we know it]
Content from our partners
Related
Join the debate
Subscribe here to comment