Extract from ABC News
Artificial intelligence has taken over the book publishing industry, or at least that's the impression from the headlines. (Getty Images: Malte Mueller)
Is artificial intelligence an "existential threat" to traditional book publishing?
That's the impression we get from the headlines: online sellers flooded with AI "slop" books, publishing houses inundated with AI-generated submissions, book deals falling apart, and three literary prize winners so far this year being accused of using the technology.
Louise Doughty, a judge on the 2026 Commonwealth Short Story Prize whose winner was accused of AI use, thinks so.
"We're only at the beginning of understanding the depth of that threat, and how we can combat it,"she said.
Readers can still discern whether a text is AI-generated through certain "tells". (Getty Images: imaginima)
Readers struggling to 'keep up' with evolution of AI text
How do we even tell whether a book is AI-generated?
ABC Radio National's Critics Corner spoke to Matthew Rubery, professor of Modern Literature at Queen Mary University, trying to answer this big question plaguing some readers.
He said individual readers could still broadly discern whether a text was AI-generated, through "tells" including:
- "It's not X, it's Y" constructions
- Always giving three examples
- Laboured metaphors
But Dr Rubery said the signs of AI usage "change a lot" and it was "very tough to keep up with it".
Matthew Rubery spoke to Radio National's Critics Corner about AI disclosure. (Supplied)
One potential solution to the issue is a "human-authored" certification.
Many certification schemes are gaining traction around the world, using a range of methods including interviewing the author, asking for their edit history and using AI tools to analyse their work.
Trevor Woods, CEO of the human-authored certification company Proudly Human, said he expected the AI tools to get less and less effective over time.
In the meantime, however, the service has attracted a spike in interest which Mr Woods said could birth a "brand new industry".
Should authors disclose AI use?
He predicts the certification industry could be worth "hundreds of millions or billions of dollars" at its height, adding that individual authors, publishers and writers' groups would be footing the bill.
When asked whether big tech companies or governments should be paying for the schemes, he said that was not a realistic aspiration.
"It's like asking the bad guys that have robbed you of everything to then give you a little bit of money back to try and make you feel better about it,"he said.
Beyond the issue of writers bearing the cost of relieving readers' uncertainty around AI, Ms Doughty expressed concern about the emerging trend of using AI tools to detect AI-generated writing.
She said the tools were used to level accusations at this year's Commonwealth Short Story Prize winner, despite the prize's own investigations finding the winning work was human-authored.
"It's important for the industry to be aware of and guard against frivolous or malicious accusations of AI being used to derail a writer's career," she said.
Louise Doughty was a judge on the 2026 Commonwealth Short Story Prize. (Supplied)
She said the accusation "will always be easier to level at new authors" who did not have a body of work that AI tools could refer to, producing false positives and degrading reader trust.
"Absolutely, we must not end up with books written by AI, but we must also view AI diagnostic tools with the same level of scepticism,"she said.
Ultimately, Dr Rubery says, "there will be too many challenges for this to be successful" in the long run.
But advocacy groups have tabled another solution at last week's Joint Select Committee hearings on artificial intelligence.
Groups like the Australian Publishers Association say the government should enact laws requiring the disclosure of AI use in texts, among other improvements.
However, there are fears that the government is going the other way, favouring the interests of tech companies over authors and publishers.
Copyright changes to 'displace' authors from the book market
A proposal leaked last month revealed the government was considering watering down copyright law to create an "opt-out" system where tech companies could train their AI models on authored work without permission until rights holders told them to stop.
At the Joint Select Committee hearings last week, writers and publishers urged the government not to pursue the changes.
"AI companies seeking an opt-out model are asking government to reverse a basic principle: if you want to use someone else's work, you should first seek permission," Hachette Australia and New Zealand head Louise Stark said while fronting up to the committee.
Louise Stark spoke at the committee this week. (Supplied: Hachette Australia and New Zealand)
"An opt-out model shifts the burden onto the very people whose rights should be protected."
A submission by the Australian Society of Authors noted it was not a future issue.
Tech companies have already scraped vast amounts of authored work in what a Microsoft executive termed "the largest theft of labour in human history".
The ASA also cited a UK report which found 86 per cent of authors surveyed reported a drop in income due to generative AI.
Kylie Moore-Gilbert, author of The Uncaged Sky, attributes a drop in income to AI-generated summaries and knock-offs of her memoir posted online.
Kylie Moore-Gilbert says her memoir was scraped and reproduced with AI. (Supplied)
"If someone wanted to read a condensed version of my book in a couple of hours without paying for it, they could listen to this instead," she said.
"Each time a reader watches or listens to such a clip rather than reading or listening to my book is a sale I lose and royalties I don't receive."
Although more than 150 legal cases have been launched in the US over the use of generative AI, the courts have not yet provided definitive answers about enforcing such copyright infringements.
It means AI-generated books are free to "displace" writers from receiving compensation for their own work, says Australian Publishers Association head of policy Stuart Glover.
Tech companies train their large language models on books without permission. (Getty Images: imaginima)
Dr Glover, who also fronted the Joint Select Committee last week, said it was difficult to determine just how hard AI displacement had hit the publishing industry overall.
However, he attributes the softening of the non-fiction book market to AI.
He said instructional books such as travel guides and cookbooks were at "particular risk".
"Anything which has a hard reference edge to it I suspect is in a kind of terminal decline," he said.
"It's very much a mixed picture, and it requires responsible regulation by the government."
A crackdown on copyright would put more legal pressure on publishers to ensure their writers were not using AI.
"For copyright to subsist it must be created by a human author," Australian Copyright Council CEO Eileeen Camilleri said.
She said using an AI spell-checker or editor would not void an author's copyright, but "where you put in a prompt and get it to generate … copyright is unlikely to subsist".
Dr Glover said that was because "copyright is vested in the expression … the kind of words that you use to create a work".
"It's not vested in the underlying ideas," he said.
"If you generated an output through a machine, even if you had a really great underlying idea but you used the machine to write the idea, then you've got no way of protecting that work and anyone can copy it."
If an author has used AI and it is discovered after their book's release, the publisher may not be able to sell it exclusively.
As a result, publishers have become more vigilant about the work in their submissions inboxes — from updating author contracts and applying for human-authored certifications to getting cold feet about publishing new writers.
The Joint Select Committee on Artifical Intelligence's final report is due on November 30.
No comments:
Post a Comment