I have been contemplating a visiting period somewhere in Southeast Asia for a while, so today brought a small first: my first conference in the Philippines, courtesy of De La Salle University in Manila, talking about libraries, librarians, and artificial intelligence. I hope to be back next year — ideally with less bandwidth and more aeroplane involved. For now, greetings from Oxford.
Back home, or
into the footnotes?
The Event
- Conference
- CLIS 7th Conference on Library and Information Studies
- Hosted by
- * De La Salle University;
- Where & when
- Where — 16-18 June 2026
- Official Links
-
* Conference
* De La Salle University
Research summary - Beneath the Surface of Search: AI, Uneven Retrieval, and the Librarian as Mediator
AI can speed up research while making some knowledge much easier to find than others. As academic libraries expand their role from guiding searches to supporting the wider research process, this imbalance becomes a central concern. AI research tools tend to favour highly visible, well-indexed sources and dominant languages, especially English. Their answers can narrow the range of perspectives researchers encounter and make summaries increasingly uniform. What surfaces readily can begin to look like what matters most. The central issue is uneven retrieval: some knowledge appears with little effort, while other relevant material remains difficult to uncover through ordinary interaction. This has particular consequences for underrepresented languages and regions, including parts of Southeast Asia.
The study examines these inequalities through critical analysis of library guidance, policy and governance discussions, and professional debates, alongside comparative readings of outputs from large language models – AI systems trained on extensive collections of text. A decolonial perspective guides the analysis, asking how inherited structures of power shape whose knowledge is visible and whose authority is recognised. The comparison points to three connected patterns: AI tools favour already prominent sources; locally grounded knowledge may require repeated, informed prompting to surface; and differences across languages and regions reflect unequal training data, validation practices, digitisation and research infrastructure. The difficulty therefore extends beyond missing information or biased content. Relevant knowledge may be available to a model without appearing in its default answers. Recovering it can require users to recognise the very gaps they are relying on the tool to help them explore.
These patterns give librarians a crucial role as mediators of uneven retrieval. Their expertise must combine technical understanding with cultural and interpretive judgment: recognising what an AI answer leaves out, helping researchers find locally grounded sources, and understanding how different communities establish credibility. Library governance must also consider how AI tools shape discovery, summaries and judgments about trustworthy knowledge, alongside their benefits for efficiency. Because the study uses interpretive comparison rather than a large-scale benchmark, its conclusions offer a conceptual framework rather than a universal measure of AI performance. Its central argument remains clear: inclusive research support depends on librarians who can identify the limits of default AI responses and help bring overlooked knowledge into view.