It was a matter of time before the phenomenon gained momentum. For months, a good number of authors – including George RR Martin, to whom we owe Game of Thrones — have already made their voices heard against the massive arrival of AI in the world of publishingdenouncing in particular the way in which their works were used without their consent to train ever more efficient language models. Generative artificial intelligence has just reached a new milestone, and is now entering the front door.
A publishing house sells its books to AI
On November 15, American author and screenwriter Daniel Kibblesmith shared on Bluesky a strange email from his publisher HarperCollins. We then discover that the publishing house wishes to obtain authorization “to integrate (his) book into a partnership established with a large technology company, aimed at using an extensive selection of non-fiction works to train generative artificial intelligence“. A decision that makes one cringe, especially when we know the conflicting relationships between authors and artificial intelligence.
$2500 to sell your work
In his email, the publisher specifies the details of this agreement: each author is free to refuse or accept selling his work for AI training purposes. If applicable, HarperCollins specifies that a fixed and non-negotiable sum will be paid to the rights holders, at the rate of $2,500 per work. In the case of Daniel Kibblesmith and his children's book Santa’s Husbandthe sum is therefore to be divided in two, with $1250 for the author and $1250 for the illustrator. “These conditions have already been negotiated and accepted by several hundred authors, so individual negotiation is not possible“, however warns the publisher, who demands “a global acceptance or refusal“.
Better than nothing?
Behind the proposal formulated by HarperCollins, the cynicism of the publisher is striking. Faced with a tense situation between the different players in publishing, he believes “This technology exists and is already used“. Understand, generative AI already uses the contents of published books to train its language models. It is better to be paid and willing, rather than to see your work exploited for free.
The fact remains that HarperCollins' promises are light, in the face of the announced surge of AI. If authors do indeed have the right to refuse to see their works sold to a large technological company – whose name we do not know, the agreement only covers three years of exploitation of the texts. It's hard to believe that after this time, these will not remain in the memory of LLMs.
For its part, the publisher highlights an attempt at innovation, aiming to offer new “interesting perspectives, while preserving the fundamental value of the works and shared income“. This is not the first time that HarperCollins has given in to the sirens of AI: last April, the company already announced a partnership with ElevenLabs, specializing in audio AI technologies, to produce audiobooks in foreign languages.
Conversely, other publishing houses are choosing to reject the perspectives offered by AI altogether, preferring to protect authors from a situation that is still in its infancy, and which is inevitably likely to escape them. This is the case of Penguin Random House, which modified the clauses of its contracts last October to explicitly prohibit his books from being used to develop AI language models.
???? To not miss any news on the Journal du Geek, subscribe on Google News. And if you love us, we have a newsletter every morning.
