Academic Publisher Sells Authors’ Work to Microsoft for AI Training
The international academic publishing company Taylor and Francis says “it is providing Microsoft non-exclusive access to advanced learning content and data to help improve relevance and performance of AI systems”.
That “learning content and data” is your writing and research.
The statement from Taylor and Francis, published in an article at The Bookseller, follows the recent circulation on social media of details about an agreement between its parent company, Informa, and Microsoft.
The Bookseller reports:
Informa will be paid $10m+ for “an initial data access” of the works it has the rights to, with a recurring payment of an undisclosed sum to be made over the subsequent three years.
When contacted by The Bookseller, Taylor & Francis said it is “protecting the integrity of our authors’ work and limits on verbatim text reproduction, as well as authors’ rights to receive royalty payments in accordance with their author contracts”.
One of the biggest concerns… is over whether it is possible for Taylor & Francis’ authors to opt out of the AI partnership with Microsoft. [Ruth Alison] Clemens [whose work has been published by the firm] told The Bookseller: “There is no clarity from Taylor & Francis about whether an opt-out policy is in place or on the cards. But as they did not inform their authors about the deal in the first place, any opt-out policy is now not functional.”
The Bookseller asked Taylor & Francis if it was possible to opt out and requested clarification over whether authors had been told about the AI deal. The publisher’s spokesman said he couldn’t comment further.
Taylor and Francis publishes many philosophy journals.
I sort of feel like the techbros are trying to turn us all into Marxists? I mean, could there be a more classic case of exploitation in Marxian terms?
“But you signed the contract!”… as if any of us had much of a choice.
Time for those of us not on the market to commit to open-access-only publishing (and possibly refereeing). Enough.
Presumably open-access publishing doesn’t need to be sold.
Disgusting. The end of paywalled, for-profit academic publishing cannot come soon enough.
Yep. And I think they foresee that it will, which is why they are squeezing money out of what they have at the cost of their reputation and accelerating authors’ exit
I’m not an expert in intellectual property and I don’t think legal doctrine on this is quite settled yet, but it’s worth noting that open access publications will—practically speaking if not legally speaking—be less protected from use in training sets. If people’s objection to this is grounded in an objection to their papers being used in training sets, the open access/for profit distinction is a red-herring.
One objection is that if the chatbot is trained on an academic’s paywalled work, people who are stopped by the paywall will not be able to access the original work, but the chatbot can charge less than the paywall and give a diluted or inaccurate version of their work.
So Open Access would be different, because if a work is Open Acces, the public will be able to get the original work the academic wrote for free. They won’t need to go to a chatbot for it (and possibly get a poor version of the research), and they won’t have to face a paywall to get the original version, either. So I believe Open Access will be different from for profit publishing.
Right, that is an asymmetry, Susan. I suppose I’d defend my parity thesis by curtailing it a bit: objections based on authorial rights seem to be insensitive to the for profit/open access distinction, as well objections to training such AIs in general. But if the concern is something like differential access as you say (and I can see public justification and epistemic arguments for it), then the difference matters.
Small point of clarification here: The contrast isn’t between open access and for-profit. Open access journals can be profit making, while paywalls can exist for non-profit journals.
No, I can hate both Microsoft et al.’s parasitic efforts to train their AIs, and T&F et al.’s rampant profiteering at the same time. It is really quite easy.
T & F is also, of course, the parent company of Routledge publishing. As an author who recently published with them (April 2024). I can confirm that I was not informed of this, nor offered any opportunity to opt out.
With respect to open access publishing, most of it is under a Creative Commons Attribution license (or stronger). That means that although AI can be trained on open access content, they are in violation of the license if they output verbatim passages without an accurate citation. It’s not clear that the material shared by Taylor&Francis has any such protection.
given the choice between ai incorrectly paraphrasing me or reproducing me verbatim, neither being paid of course, i’d actually take the latter. at least the point would be preserved.
“In Ian’s spectacular and thorough essay, he argues that large language models are the best thing that ever happened to the world” — ChatGPT
thesis of paper: large language models are a disaster in almost every conceivable way
FYI, I contacted T&F regarding articles published in Australasian Journal of Philosophy (in which I have published) and was informed that “the agreement with Microsoft includes books content only, and does not affect articles published in journals including the Australasian Journal of Philosophy.”
Authors are supposed to have ‘moral rights’, which include protecting the integrity of their work and their reputation as an author. It seems to me that using an author’s work to train LLMs doesn’t protect the integrity of that work — it uses their words, ultimately, to express ideas and meanings that are not present in what they originally wrote. It doesn’t matter that these new meanings are not attributed to the author; in fact it’s a second conflict with authors’ moral rights, that they get credit for their contributions to derivative works.