Parallel Session 10 – AI and Library Service Development II
Date: Thursday, 2 July 2026, from 09:00 to 10:30
Moderator: Liisi Lembinen, University of Tartu, Estonia
Location: R7
10.1) Navigating Change: Academic Libraries and AI
Presenter: Mauritza Jadefrid, Linnaeus University, Sweden
Since the release of ChatGPT in November 2022, discussions about Artificial Intelligence (AI) in academic libraries have expanded. Many academic libraries are undergoing transformation in everyday work practices, interacting with technology. At the same time as generative AI (GenAI) continues to develop, academic librarians are trying to figure out when and how GenAI is relevant for the library. Academic libraries are considered reliable and trustworthy, and implementing AI into the library comes with many questions. The gap between traditional library roles and modern knowledge environments creates tensions in keeping core values within the libraries and maintaining necessary innovation.
In my doctoral study, I explore how four Swedish academic libraries navigate change. The purpose of the thesis is to investigate how academic libraries engage in and contribute to shaping understandings of AI as part of their activities. I want to understand in what ways, if any, artificial intelligence is becoming something to be implemented in the activities of academic libraries. I also want to understand to what extent and in what ways artificial intelligence is being shaped and negotiated in academic libraries.
In this presentation, I will share the work in progress from my doctoral project. The study has a qualitative approach and is based on a sociomaterial perspective on technology in organisations. The starting point for the methodological study is conversations with participants about their practices in relation to AI, based on their roles and tasks at the library. Semi-structured interviews, follow-up conversations, and field notes from participants’ meetings and workshops constitute the methods used in the thesis.
I have four questions I want to discuss:
- How does AI manifest itself as a strategic issue in the context of academic libraries?
- How do library staff describe their role in relation to AI implementation, and what issues arise regarding ethical responsibility when AI is implemented in the information environment of academic libraries?
- How are issues of learning and competence expressed in relation to AI in academic libraries?
- How does AI take shape in library work processes and activities, and how is the technology reconfigured through staff and organisational interactions?
These four questions could be described by the concepts of strategy, activities, ethics, and competence. Strategy deals with issues of power, direction, values, and ethics. Activities encompass action, application, and the technology used. Ethics concern’s identity, ethical issues, and professional culture. Finally, competence covers issues related to change, knowledge, and reflection.
10.2) From Complexity to Clarity: Enabling Transparent and Responsible AI-Enabled Data Use at WUR Library
Presenter: Cristina Huidiu, Wageningen University & Research, The Netherlands
At Wageningen University & Research (WUR) Library, we are reimagining how data-driven tools and AI can empower librarians, researchers, educators, and students by making complexity approachable. Our work responds to a fast growing change in academia: how can we engage with powerful AI technologies without compromising transparency, trust, or institutional control over sensitive data?
We present a portfolio of initiatives that translate this principle into practice. First, we explore “Talk to Your Data” prototypes that allow users to interact with research output metadata using natural language. This approach demonstrates how generative AI can unlock data accessibility—without transferring any data to external models. In this context, we share our implementation of a privacy-preserving architecture where models are brought to the data, enabling question-answering and summarization tasks while safeguarding compliance with European data requirements.
A second focus is our investment in intelligent data quality pipelines. These combine automated validation, enrichment, and feedback loops to reduce manual metadata management. The result: improved metadata reuse across existing and new usecases without an exponential growth in cost.
The third layer focuses on advanced analytics use cases powered by shared, well-curated data products. These support research evaluation and strategic decision-making across departments: from standard bibliometric analyses to Standard Evaluation Protocol (SEP)-aligned reporting, with additional usecases on our roadmap centered around the estimation of ROI on research infrastructure and generating trust indicators to help researchers navigate the publishing landscape responsibly.
Our approach is grounded in the academic values of openness, reproducibility, and user empowerment. We design with non-technical users in mind, ensuring that tools support explainability, human oversight, and multilingual usability.
Through this presentation, we aim to spark conversation on how libraries can lead in embedding AI in ethical, accessible, and human centered ways as well as discuss the challenges and successes, both technical and human, we have learned from along the way.
10.3) Using LLMs to Design an Indicator for Clinical Relevance
Presenter: Silvia Fattori, Vrije Universiteit Amsterdam, The Netherlands
Working in a research intelligence team within our university library, we often support research departments in medicine to explore and interpret trends in their research output. Increasingly, these research departments want to have more insight into the relevance of their academic work for application to societal issues. In the field of medicine, clinical relevance is a key component of such societal impact. Our current measures of clinical relevance include authorships of and citations in clinical guidelines. These measures are valuable, but cannot convey important gradations of relevance, and only measure relevance ex post. In collaboration with researchers from the Cancer Center Amsterdam (CCA) and with a researcher experienced in computational modeling, we propose a new framework where we use the capabilities of modern LLMs to categorize academic publications into user-defined levels of clinical relevance and apply this to the output of CCA . Our approach can potentially be adapted to other research areas, providing a more nuanced measure of societal impact. This work could also contribute to the development of new research metrics and inform research policy decisions.
In the session, we will discuss the following points.
– How we conceptualized and validated the different levels of clinical relevance and how these levels were operationalized to work with the computational model. Our levels range from basic research using cell cultures to clinical work with human participants. These levels were validated by experts in cancer research.
– How we experimented with different LLMs and explored various prompt engineering techniques to improve the quality of the indicator, and how we evaluated the outcomes
– Preliminary results, limitations, ideas for future improvements, and other applications of the method in the context of academic libraries.