The spectacular fall from grace of Haitian-Canadian writer Thélyson Orélien in many ways resembles our own culture war over Jason Arday. The celebrated author, feted in TV studios and in the literary salons of Quebec and Paris’s Saint-Germain-des-Prés for his hard-hitting descriptions of the migrant experience, was accused on X last week both of serial plagiarism and of using artificial intelligence to write his latest novel. His accusers decried the soft bigotry of low expectations that caused the woke intelligentsia to turn a blind eye to mediocrity because their hero was a capital-B Black man. His defenders, just as in Arday’s case, tried to dismiss the original accuser as an extremist (the anonymous account responsible turned out to be run by Samuel Fitoussi, a columnist for the right-wing Le Figaro). They also attempted to cast the affair as the working of a racist lynch mob. The leftie magazine Les Inrockuptibles even came up with the thrillingly Stalinist line that ‘in this affair, whether the accusation is well-founded is not what matters most’.
Orélien has denied the accusations. Literary sleuths have pointed out that Pangram – the AI detector that has marked up swaths of the text of his novel C’était ça ou Mourir as non-human – is prone to false positives, wrongly flagging pages of classic French literature written long before the advent of AI. That, however, has not stopped the committee of the Prix Goncourt, France’s equivalent of the Booker Prize and not known as a hotbed of reactionary politics, dropping Orélien’s book from this year’s shortlist.
A former literary scholar at All Souls, Oxford, has called AI ‘completely brilliant’
But l’affaire Orélien poses a far more serious question for our times than the obvious one about race relations. What would it mean if a novel part-written by AI were actually excellent? In a fair world, should the author be sharing his prize with the Large Language Model (LLM) which may have toiled as his co-author? ‘This book is very good – if it’s AI that made it, we should use it!’ the Goncourt Committee member Éric-Emmanuel Schmitt told a talk show in the wake of the scandal. ‘No, I’m kidding!’ he quickly added.
But if Schmitt is kidding, other authors and thinkers are starting to engage with the real possibility that the world is approaching a singularity moment where AI is as good as humans at ‘thinking’ and writing. The remarkable Dr Iain McGilchrist, a former literary scholar at All Souls, Oxford, who re-qualified as a consultant psychiatrist, has called AI ‘completely brilliant’, able to offer advice with a sensibility ‘very like my brother, who is an extraordinarily civilised art historian’. Lower down the intellectual totem pole, Alain Damasio, known as ‘the Pope of French science fiction’, is sufficiently secure in his reputation to admit that he regularly uses AI in his writing. ‘I’m stunned by its creative capabilities when it comes to imaginary worlds,’ he said recently. ‘It’s almost on the same level as my own craftsmanship, and it’s only going to keep improving.’
The Orélien case is just the latest in a long line of acclaimed books that have been flagged as AI. Rie Kudan, winner of Japan’s 2024 Akutagawa Prize, admitted that she had used ChatGPT. AI-detector-triggering stories have won four out of the ten Commonwealth Short Story regional competitions in the last two years. An almost wholly AI written book was runner-up in a Chinese science fiction prize in 2023.
But there’s a catch. Anthropic, the creator of Claude, has bought up millions of books and scanned them to build the database of its LLM (and then pulped the copies, sparking outrage). In other words, the published works of most authors writing today are already in the belly, or the mind, of the beast. (I know this because Anthropic owes me and thousands of other authors involved in a successful $1.5 billion class action suit a shed-load of money for compensation.) That helps explain why a book by French author Sophie Jomain written in 2009 could come up as 94 per cent AI. A new text written by Jomain came up as 100 per cent machine-generated. ‘What does that mean?’ asks Jomain. ‘The more I write, the more my writing starts to resemble artificial intelligence? Or is it that the more I write, the more artificial intelligence feeds on it and starts to resemble me?’ In a further mind-bending twist, Pangram also offers to ‘humanise’ AI text for a fee, playing both gamekeeper and poacher.
Most people believe using AI to write is a kind of faking or cheating. But the question of whether the art it produces is automatically bad remains. The German artist Wolfgang Beltracchi painted hundreds of fakes of 20th-century masters that were validated by art historians, specialists at Christie’s and the director of Paris’s Pompidou Centre. They sold for tens of millions. Max Ernst’s own widow even certified one of the works. Beltracchi was exposed (and jailed) in 2011, and the value of his fakes obviously plunged – though his creations were objectively brilliant in conception and execution. Claude recently wrote a ‘Bach’ fugue that even music scholars describe as ‘workmanlike’ and non-experts like me find magical.
But Orélien – if he did use AI – wasn’t faking Ernst or Bach. He was faking himself. Moreover, people are still unaware that powerful new LLMs work best as a car you drive, not as a robot you send on errands. Try prompting an AI programme, even one as powerful as Claude’s Opus 5, to ‘write me a novel in the style of Thélyson Orélien’ and the result will be identifiable as AI slop – superficially convincing, maybe, but full of heavy-handed stylistic formulas and wooden characters. Amazon Books, whose content is estimated to be at least 40 per cent AI-created, is full of such formulaic genre-fiction drek.
However, if you, as an author, ask the machine to thoughtfully critique the prose you wrote yourself, it will likely do a great job. From there, it’s a fatally simple step to asking the LLM to go ahead and add its fixes and thereby improve your text. That’s the slippery slope on which today’s literary world finds itself. Samuel Fitoussi asks whether ‘the work of writing will change form… no longer about choosing each word, but about guiding one’s AI’. Could we soon be living in a world where ‘the one who produces the best novels will be the one who has invested the most creativity, intelligence or, at least, some kind of something in their AI’?
Fitoussi’s perhaps bleak vision contains chinks of hope for human creativity. In his prediction, universal access to AI does not mean anyone can write a novel. The machine could become ‘a cognitive lever, a talent amplifier’ for those with sufficient mastery to use it. Non-masters of the craft, meanwhile, will merely produce derivative rubbish.
And what about the critics and prize committees who claim to be the arbiters and gatekeepers of public taste? As C’était ça ou Mourir has shown, AI already has them thoroughly fooled.
Comments