In this post I attempt to reveal inherent conflicts in our drive to be as open as possible, authors’ need to understand their rights, and a library’s mandate to provide their patrons with the enhanced discovery that comes with AI’s large language models (LLMs).

In my view, authors should be enabled to make decisions about how their content is used. In our current world of Creative Commons licensing, content can be reused, but not without attribution. What does this mean for publishers and libraries under immense pressure to provide LLM services, but perhaps not fully grasping the ramifications of their actions? Are there ethical considerations at play that may help guide our approach?

Many of us are not shy expressing our opinions on AI. Some of these opinions are borne out of fear – which is often a response to feeling a lack of control. AI seems to be a pressing inevitability, and we fear that we will be subsumed – our content, our rights, our jobs – all that we know. Of course, fear in response to the unknown is natural, and it is so much harder to grapple with potential positive outcomes from AI if we are collectively in a state of shock.

I work at the American Mathematical Society, a 30,000 strong membership society representing mathematicians — from student to researcher, from educator to professional. Mathematicians care deeply about their endeavor, and indeed are fueling much of the evolution of AI technology itself. They also are deeply invested in rigor – the rigor of a mathematical proof, the elegance of expression, and of course having their hard work attributed to them.

This is not just true of mathematicians of course. Janet Salmons’ recent guest post here in The Scholarly Kitchen (Supply Chain of Writing Fools), eloquently made the point from the perspective of a writer and qualitative scholar. “…I found out I’m just a kink in generative AI’s content supply chain.” Her books have been “swallowed up” by LLMs without her consent. As Salmons puts it, “For scholarly writers, the issues go beyond compensation. For us, the irreparable loss is to the integrity of our work.”

So where lie the priorities? It really depends on where you sit. Salmons describes well the point of view of a researcher, and in her case, one who understands the nature and value of a Creative Commons license. This is not true for many authors, who happily embrace a CC BY license without understanding the deeper nature of copyright and licensing. As an academic institution, you care about your faculty, and you care that authors retain their copyright. Many publishers no longer require the transfer of copyright. Even when employing a Creative Commons license, the author still retains copyright over the work (although depending on the license, some rights are effectively softened). An institution also cares that they provide services to their faculty through their library, and this is where things get a little sticky. Many libraries are embracing AI as an extension of text and data mining services with a desire to use AI internally to enable discovery. Beyond that, they want to provide faculty at their institution with ever richer resources to enhance teaching and research.

When a publisher such as the AMS talks to a scholarly library about purchasing content (and here perhaps I will focus mainly on eBook content), we all agree that it is essential that we move to accessible formats – which is a complex endeavor in mathematics, but one which is rapidly evolving. We also collectively understand that we are selling our content for use by a library’s patrons in teaching and research. On top of this there is an implicit agreement that authors of this content are recognized as contributing to the richness of a library’s scholarship, and indeed they are rewarded for this with academic recognition and, of course royalties.

At the same time though, many libraries are talking with publishers about allowing AI licensing. And this is where here there appears to be some real dissonance among the stakeholders. To some extent this has ballooned into suspicion of publishers’ motivations in holding back on such AI licensing clauses in content deals. But there are reasons for this: for example, here at the AMS, authors routinely retain copyright on their books, granting the AMS an exclusive license to publish their book. It is therefore hard for us as a publisher to then turn to a library customer and grant an AI license in which attribution is absent, or partially achieved by a non-legally binding method – although there have been efforts within the Creative Commons license to address this issue through Preference Signaling. It will of course be interesting to see how the many fair use/copyright litigations in play will be resolved.

Authors need to decide the fate of their work. As publishers, it is our collective responsibility to inform and educate authors on copyright, and the implications of licensing arrangements so they may be as informed as possible.

Thus the dissonance lies in the real conflict between the ideals of openness that many authors, publishers, and libraries embrace, and the benefits of providing AI services to users, libraries, and publishers alike — not to mention the outsized potential of material gain from corporates in the AI sector who appear loathe to share those rewards with authors.

A perhaps tangential, but in my view relevant, concern for all stakeholders is the question of what company authors keep when their material ends up in an LLM. Do authors really want their ideas to appear alongside pirated content from outfits such as the Books3 Database, which most LLMs use? (A 2023 article on the Books3 Database in Wired magazine entitled The Battle over Books3 Could Change AI forever is worth looking at here.) Do librarians and authors really think such content should be mixed in with their legitimate content? And how does this interaction affect the integrity of the scholarly record, not to mention author, library and publisher integrity?

There is no doubt in the potential value in library AI services for faculty and students. There is no doubt that copyright and licensing matters. There is no doubt that an author must publish with attribution to their work, and their content published and sold with the understanding of who the rewards accrue to, the integrity of content, and the importance of acknowledging author contributions.

How do we solve for this dissonance?