Inside Higher Ed | Moor Studio/Getty Images
Truth is, university faculty and librarians interact infrequently, perhaps to wrangle obscure research material or to provide newbies with a library introduction. But now that collaboration must deepen on behalf of anyone who cares about authentic knowledge.
As a faculty member at the University of California, Irvine, I was recently shocked to learn that AI slop had breached our carefully curated—and expensive—library holdings. Most shocking was that the incursion wasn't marginal. It came from the heart of our holdings: Springer Nature. Apparently, since May 2021, Springer Nature has been publishing “AI-based” literature reviews, book-length texts that the publisher describes as incorporating “a mixture of human-written text and machine-generated literature overviews.” And since the California Digital Library purchases Springer’s front list of electronic books for University of California campuses, these “AI-based” works are now available to all UC faculty and students.
Springer Nature’s machine-generated literature overviews come with a disclaimer in the metadata. Not the best scholarly practice—which should be driven by human expertise—though perhaps not tragic in more technical areas like Skeletal Muscle Physiology, where certain types of nonhuman data analysis can be uniquely useful. But seriously: a “book” like this costing $139.99 on its own, now incorporated quietly into the CDL? A book drawn only from Springer Nature content, not from the best sources available, and still admittedly experimental as an academic publishing business practice?
A $159.99 hardcover, Criticism and Critical Theory: A Machine-Generated Overview, “authored” by Laxman Jogdand, starts with a disclaimer of sorts that explains the book has been generated via an extractive summarization approach and then “carefully edited” and introduced by Jogdand. Whether the introductions are themselves AI-generated or not, they might as well be, as they string together banalities about how, for instance, “literary readings have reached new heights with the advent of a scientific approach” and that classical authors like Longinus underscore “the timeless relevance of classical literary thought in understanding the complexities of human creativity and expression.”
But perhaps when you are extracting such humanistic literature from only the Springer Nature bullpen, that’s the type of introductory slop you get (a thorough review of one such title is available here). And I can only imagine how such resource hoarding looks from the perspective of small publishers, like Duke University Press or Verso, where they actually publish good critical theory books.
In addition to the explicitly labeled “machine-generated” titles, there is also the case of Mastering Machine Learning: From Basics to Advanced, which Springer Nature retracted only after Retraction Watch reported that the title had multiple references to works that did not exist. Everyone makes mistakes, and corporate publishing is no different?
Springer Nature, no doubt like other business-minded publishers across the globe, is floating trial balloons in order to figure out what university arbiters of knowledge will tolerate at the bleeding edge of a business model. How widespread is this? For example, another major publisher, Elsevier, appears to be pushing at the boundaries of authenticated knowledge less aggressively, but they—like everyone else in the serious writing world right now—struggle to define where the current boundary holds. Where their Q&A asks, for instance, “Can I use AI-generated text directly in my manuscript?” the answer is essentially yes. Sure, the author remains responsible for the text that is generated. But at the same time, the author gets partially subsumed into their new role as a quality-control inspector.
University faculty and librarians have to say no now, collectively and loudly. Our role as the highest arbiter of authenticated knowledge is at stake, and there are no other institutions—including the legal apparatus protecting intellectual property—that could step into the void; this quality-control issue is only partially covered by the law or by federal institutions like the National Institutes of Health, which themselves are intertwined with those same businesses managing expert peer-review publications, like The Lancet (Elsevier) or the deservedly prestigious Springer publication Nature.
How did we get here? From a publishing business perspective, of course, the more in-house generated machine-learning and AI material a publishing house can produce without expert authors to compensate or universities to vet, the better for the bottom line. At the risk of sounding overdramatic, we are just now at the cusp of a global battle to re-establish what authenticated knowledge means in the first place. And there are very well-funded and lawyered entities now trying to push those boundaries aggressively in a direction that departs from identifiably human expertise while embracing new forms of knowledge, or pseudo-knowledge, defined not by human expertise but by the technical and legal—and financial—firewall that separates published material from the human experts at their origin.
At this rate, we will soon be living in a world where expert peer review persists mostly at the top of the knowledge food chain, while the knowledge-production machines controlled by commercial entities chip away at actual human knowledge-production practices from below.
What are the next steps? First, broadcasting. Recent tactics out of the knowledge business suggest they are trying a slow boil that shifts sensibilities around knowledge without anyone protesting. And as the UC Irvine Library’s head of collection strategies, Becky Imamoto, and I discovered as we presented this material to our campuswide AI Advisory Committee, faculty are typically not aware of the situation five years on, though they find it shocking when informed.
Second, collective action. As the landmark 2020 open-access agreement between the University of California and Springer Nature demonstrates, when called out, our knowledge businesses are in fact eager to negotiate these basic issues at scale, rather than let the knowledge system they help curate crash more dramatically.
Third, we should selectively support publishers who continue to hold up the highest standards, for instance (in my recent experience) another global titan, Cambridge University Press, which has poured significant material and intellectual capital into its forthcoming five-volume Cambridge History of Rhetoric project, involving the efforts of about 250 recognized area experts, supported by at least three rounds of editing on the scholarly side, plus another four careful rounds on the production side. (I’m co-editor of volume 5.)
For librarians, the message is this: University research faculty care deeply about you in your role as the most immediate gatekeepers for what counts as knowledge. It’s time for a new collaboration along these lines when the stakes appear, for us both, profound. For others, the message is to consider seriously what alternatives are left if the universities and their libraries fade.
