Championing transparency in an AI world
2026年6月16日
Linda Willems
As the number of library users turning to AI grows, librarians are looking more critically at the tools and technologies they use.
In her role as Vice President of Global Library Relations and Partnerships at Elsevier, Emily Singley is a regular speaker and attendee at academic library conferences.
“I see it as a crucial part of my job — it gives me the opportunity to learn from the library community and better understand how we can partner with them,” she explains. “And after years of attending them as a librarian myself, they feel like a second home!”
Perhaps unsurprisingly, AI and its impact on the institutional mission now features in many of these conferences’ schedules. It is also proving a hot topic during coffee and lunch breaks. Emily has found that while librarians see the potential of AI to benefit research, education and even library services, they are also aware of the risks. For example, an Elsevier survey of librarians found that 92% worry AI will be used for misinformation, while 88% point to ethical considerations.1 They are not alone in these concerns. While researchers are increasingly using AI, only 22% trust existing tools.2
In this article, Emily shares her four key takeaways from recent North America library events. We also talk with a member of the development team behind LeapSpace, Elsevier’s new AI-informed workspace, to understand how they are addressing these concerns.
1. Librarians are cautious about AI
“What I’m hearing is that librarians feel a lack of agency — that big tech is making AI happen and there’s nothing that they — or their institution — can do about it,” Emily reveals. “But this is proving a problem because many currently struggle to trust AI.”
According to Emily, most of these librarians report working with students who are using AI inappropriately. “This is sparking concerns about plagiarism,” she explains, “but also about loss of critical thinking. If the thinking portion of research is outsourced, they are worried about what that means for researchers, particularly for early career researchers. And what about the next generation?”
These observations were reflected in our librarian survey, with 84% of respondents worried that AI will erode critical thinking.1 And they may have good reason — more than 70% of researchers we surveyed for a 2025 report say they regularly use ChatGPT for work, with usage highest among early career researchers (83%).2
While many of the librarians Emily spoke with are focused on navigating the new situation, others are taking a different path. “I heard stories about librarians whose resistance to AI is spilling over into their roles. For example, at one conference I was introduced to the concept of “AI shaming” — apparently, some librarians are giving library users who ask about AI a really hard time!”
Emily suspects this kind of behavior is driven by a fear that libraries — and what they stand for — are under threat. “There’s concern that people may stop engaging with content, and then what would that mean for education, for research and for libraries?”
Librarians are also worried about “slop” — a term used to describe the high volume of low-quality content that AI currently generates. “There are real concerns about quality of information and how it is getting diluted. And that all makes perfect sense when you think about where libraries sit.”
2. Librarians realize they have an important part to play in responsible adoption of AI
Emily isn’t surprised by any of the concerns she’s hearing. “Librarians are information professionals. They understand search and retrieval, they understand publishing and they understand content in ways your average researcher doesn't even think about,” she explains. “So, they should be saying, wait, what is that source and how was it vetted? And is what the AI says even in that source? At several of the conferences, there were some really good conversations about the importance of carving out this role as the “critical voice” of AI.”
For many, this includes being a “safe space” for library users to pose questions about AI; questions they may be embarrassed to ask their advisors. “Libraries have always been a trusted and neutral information source, and I think that librarians are embracing that, along with AI literacy, because they see a real need for a place on campus that holds that expertise. Right now, at many institutions, it’s unclear where that knowledge sits and it’s often diffuse.” This is borne out by the results of our 2025 survey Researcher of the Future — just 32% of respondents said their institution provides good AI governance and only 27% believe they have adequate training in using AI.
Emily says: “That's the opportunity gap for libraries and I think it's something they are really well positioned to take on. But with AI moving so quickly, it’s a real challenge for them to keep pace, so you also see librarians looking for opportunities to upskill and learn.”
3. Librarians don’t just want to be told AI is transparent... they want evidence
Emily points to the fact that if companies want librarians to feel confident in responses generated by AI tools, they need to show them how those tools work. “Libraries are always going to question — that's their job. So, it's not ‘cool, you put that little button there,’ it’s more like, ‘cool, but how does it work and what does it actually do?’ They want to know how the AI decided which information to bring back and where it got it from. And librarians are very interested in metadata fields — understanding which ones a tool searches, and those it doesn’t. Transparency in processes and results is key to building librarian confidence.”
4. While new technologies remain an enigma for some librarians, understanding is growing
At most events she has attended, Emily has witnessed a rising awareness of AI and an appetite to learn more about the new technologies emerging. These include agentic AI, a sophisticated AI system that combines the data extraction and content creation abilities of GenAI with a reasoning engine, which simulates logical thinking and decision-making. This enables agentic AI to draw up and execute multi-step plans and use multiple tools – ideal for open, complex or multidisciplinary questions.
Emily notes: “It’s still early days with agentic AI. Few people, including librarians, have a good sense of it yet. Even so, it was pretty clear to folks at these conferences that this is where the future of AI lies.”
She adds: “But as AI grows more sophisticated, so does librarians’ need for transparency. They work really hard to curate and get the best and most trusted content for their researchers, so it’s vital that they can evaluate tools and their sources.”
Building a research-grade AI
Adrian Raudaschl, the Senior Product Director behind Elsevier’s AI workspace LeapSpace, understands why librarians are concerned.
“Most AIs are designed to sound persuasive, not to help you think and show you the evidence. But I trained as a doctor in Glasgow's NHS where evidence-based medicine was drilled into me. We didn’t just believe something because someone said it, we checked the source and looked at the confidence level. That's how researchers think too and that’s guided how we have built LeapSpace.”
He explains: “LeapSpace is not a people pleaser. It’s there to help you make informed decisions, even if that means going back to the drawing board. We're not trying to create a passive reading experience, we show you what's uncertain, what else you could try, what's emerging, to sharpen critical thinking and judgement.”
Adrian points to the fact that when the agentic AI reasoning engine at the heart of LeapSpace gets to work, it shows you the steps it takes on screen. And responses are referenced so that claims or assertions can be traced back to their peer-reviewed source.
And for Adrian, the quality of those sources is one of the most unique aspects of the tool. “Everything in LeapSpace is grounded in Scopus—the world's largest curated research database. Over 100 million records, with external content advisory review. On top of that, 15 million full-text articles from multiple publishers. Not the open web, but peer-reviewed, curated content. That's the foundation.” And that content base is continuing to grow – Elsevier has signed licensing agreements to index selected full-text journal content from Emerald Publishingopens in new tab/window, IOP Publishingopens in new tab/window, NEJM Groupopens in new tab/window, Oxford University Pressopens in new tab/window, and Sageopens in new tab/window, with further content partnerships in the pipeline.
Adrian adds: “Then there's how we surface that content. We have Trust Cards, so that you can easily evaluate what we show you. The Trust Card link to Statement shows how closely an AI-generated claim aligns with the full text or abstract that inspired it. We also highlight the specific section of the source LeapSpace has used. Trust Card’s Claim Radar feature takes this a step further — it scans the broader literature for each claim and shows the proportions of articles that support, contradict or are ambivalent about it.” Elsevier is also in the process of recruiting an independent AI Advisory Board, whose role will include monitoring LeapSpace’s algorithms for transparency and publisher neutrality.
All these measures are not only designed to build user confidence but to maintain human oversight. And for Adrian, these steps are vital if LeapSpace is to deliver on one of its core goals — to help researchers accelerate innovation. “The burden of knowledge keeps growing and we’re running faster to stay in place. But here's the thing — many ideas are already published; they're just not connecting. And exploring a new connection is risky because it may take months or lead nowhere. That's one of the things that LeapSpace can change. We make it easier to explore —the connections, the gaps, the adjacent ideas — and we can help researchers do that with a critical approach and confidence.”
貢獻者
