Canadian publishers rely on trust as AI detection tools prove unreliable
A recent controversy over a Quebec author's novel has renewed focus on the publishing industry's struggle to verify AI use.
TORONTO · Canadian publishers say they are relying on author trust because technology cannot reliably detect artificial intelligence in manuscripts. The challenge has come into focus after a Montreal publisher, Les Éditions du Boréal, ended its relationship with author Thélyson Orélien this week. The publisher said an analysis of document metadata showed evidence that AI was used in writing his novel "C'était ça ou mourir." Orélien had previously denied using AI to write the book, stating he used it only for research. He did not respond to a request for comment after the publisher's announcement on Wednesday.
Alana Wilcox, editorial director of Coach House Books in Toronto, said the company's contract requires authors to affirm their work contains no AI-generated material. She said existing tools to identify AI use are unreliable and problematic. "As AI becomes more sophisticated, we don't imagine the existing tools will become more reliable, it will become more and more difficult to identify AI use," Wilcox wrote. She added there is no reliable way to police compliance and expressed no interest in supporting detection systems trained on stolen material, including books.
David Caron, co-publisher of ECW Press in Toronto, said their contracts include a warranty clause where authors guarantee their manuscript is original and wholly their own. "We have to rely on you to tell us what it is that you've done and what's here in your manuscript," Caron said. He noted they do their best but ultimately depend on the author to disclose what they do not know. Lisa Quinn, president of the Association of Canadian Publishers, said the industry views artistic creation as a human-centred activity. Quinn, who is also executive director of McGill-Queen's University Press, said there is a significant difference between using assistive AI for comprehension and using generative AI to produce writing.
READ MORE: Canadian literary prizes rely on publishers to bar AI-written works
READ MORE: Ottawa invests $193 million in Quebec aluminum to reduce reliance on U.S
"Those uses are radically different … where the AI is in fact a generative AI engaged in the act of writing," Quinn said. She shares concerns about AI-detection technology, describing it as predictive models built on "truthiness rather than truth." Quinn said a perfect answer is not forthcoming, and results only offer a likelihood, which puts writers and publishers in a difficult position. She said AI technology is evolving faster than the industry can set policies, leaving regulatory questions unanswered.
Allegations about Orélien's book first surfaced on social media in September. Prior to the controversy, the novel had been nominated for several major French literary awards and had won the Prix du Roman Fnac. It has since been withdrawn from consideration for France's Prix Goncourt. Quinn said that while such cases show trust may not always suffice, publisher policies must be founded on a relationship of trust with a clear human-centred principle. "This is where, I think, having clear policy as a publisher is helpful, but it must be built on a relationship of trust and with the human-centred principle very clearly a component of that trust," she said.
With files from The Canadian Press, BNN Bloomberg and CP24