Conferences, even those in the South Carolina autumn sunshine at 26 degrees Celsius (79 degrees F), are a good opportunity to look at what issues are exercising the minds of those in scholarly communications. The Conference attendees were asked in a quick poll to identify the main themes, and they chose, inevitably, AI inevitably, trust, impact – and MCP.
AI
AI is transformative, and perhaps that’s what creates an undercurrent of disquiet. I thought I heard more negative references than positive mentions of AI during the conference. Quite by chance, I received this week an invitation to participate in a survey of “AI Fatigue in Libraries” (presumably fatigue among librarians, rather than among patrons). There were people talking about the benefits of AI, but you had to hunt for them.
How search and discovery is changing – for the better?
Each year at Charleston, you can observe how the tools introduced by the AI startups are slowly integrated into the major paid aggregation services, such as Science Direct, Web of Science, or JSTOR. This year, JSTOR provided a rapid walkthrough of their new “dynamic” search interface – not, in fact, very dynamic, because it takes around 15-20 seconds to carry out an AI-based search. The dynamic search appears to use genAI-type features to execute a natural-language search, and then enables the user to modify their search in light of the results shown. With many options available to the user, It all looked quite elaborate, and my immediate response was that the typical JSTOR user will not be a very expert searcher, and might not even realise (or want to use) the plethora of additional refinements available, such as “themes” (related results), or “topics” (which can’t be clicked on), and “search suggestions (which could have a narrow or broad scope). I also noticed a “go to item” button next to the results, which, explained presenter Ashank Rai, “because we want you to go to the article” (which suggests the user may not go near the article itself). I can’t help thinking many users will use JSTOR Discovery the way they use Google AI features: going no further than the AI summary results, and not clicking on the hits themselves.
MCPs
These were listed by participants as key topics, which was intriguing. MCPs were talked about, notably by Gary Price, at last year’s Charleston Conference, and by 2026 they have become widespread. Model Context Protocols are a way of managing machine-based interaction between LLMs and other services. Perhaps the appearance of “MCP” in the trending topics list indicates simply how terms move towards the mainstream over time. This year, they are simply part of the scholarly environment.
The common good
Rosalyn Metz, of Emory University, the keynote presenter, talked about setting up the common good for a community, and cited Astra Zeneca’s partnership with the University of Oxford to produce and distribute a Covid vaccine as an example of an institution working successfully with a private corporation. I remember, however, that while Astra Zeneca’s vaccine was sold at cost, other pharma companies created for-profit vaccines and made considerably more money. This contrast was reinforced by a later presentation about earnings calls (see below). She also talked about how to prevent your own content being used by genAI tools, although I could image a case being made for the common good to be genAI having access to everything, so it can answer all your questions. The common good principle sounds wonderful, but I wasn’t clear from this presentation on the specific actions she was recommending, apart from charging for access to your content, which violates the idea of open-access provision.
Follow the money
Doug Way, in another well-attended session, shared what he had learned from joining earnings calls for the six largest publishers, including Clarivate and Elsevier. When you heard how the publishers presented their activity to the financial analysts, academic publishing looks like a very different animal. Open Access is a way for publishers to increase their revenue – there is a steady growth in the number of articles published, and one journal alone has an estimated revenue of $138m per year from APCs. Subscription journal make even more money than transformative agreements. The largest publishers, notably Clarivate, are moving towards recurring licence fees for content, rather than perpetual access (following the Netflix / Spotify / Amazon Kindle model). None of this was unexpected, but all those zeros in the numbers Way quoted made people in the audience jump.
Trust
There were many references to trust in the Conference, but not much that made me feel more trusting. We must be, stated one consortium staffer, “Continually earning their trust … our members are the most important things we have”. More specific, Toby Green talked about TrustMarc, a way to “make existing trust signals visible” by providing machine-readable tags for every publication. It’s a great idea, but one that will only work if universally adopted. One questioner asked if we can trust the AI providers, and currently, we have no way of knowing.
Impact: The case of the disappearing clicks
When attendees stated “impact” was one of the key topics, they meant “lack of perceived impact”. Eugene Garfield all those years ago created citation counting as a metric for influence back in 1955. The internet started counting clicks in 1994 (the first banner advert with click-through rate). Over time, each has started to look inadequate. It’s common knowledge that the number of clicks, as a measure of usage, has gone down dramatically following the launch of genAI. We know about this from Google, whose advertising revenue has decreased because people only read the AI summaries., but it clearly many other subscription information services, such as BMJ Best Practice. BMJ’s presentation was labelled, rather plaintively, as “If they don’t click, does it still count?”, since, right at this moment, it doesn’t count. Many eyes turned to COUNTER at this point to see if COUNTER is measuring AI usage, and Tasha Mellins-Cohen revealed in a later session, that COUNTER is currently talking to genAI providers to obtain stats for usage of their AI tools, and thereby imply usage of a product. How and if this can be done in a reliable fashion remains to be seen.
Conclusion
None of the above issues was entirely a surprise. What did astonish me was the way that libraries press on with their agenda, despite the worrying developments in the macro-environment. For me, the best of Charleston was summed up by the University of North Georgia Press, with just three full-time staff, presenting at the Conference alongside the combined might of Stanford, Princeton and Johns Hopkins University Presses, yet creating more OER textbooks (over 40 to date) than all three University Presses together.
Leave a Reply