AI search is very good at finding what you asked for. The harder problem, the one nobody has fully solved yet, is finding what you did not know you needed.
That problem has a name in our field. Serendipitous discovery. And the profession that spent a century designing for it has something real to offer the engineers working on it now.
A well-designed library does something most people never consciously notice. A browsable stacks collection places organic chemistry next to biochemistry next to medical history. A well-built catalogue lets a researcher follow a subject heading into territory she never planned to enter. A reference librarian who really listens will occasionally come back, not with the answer to the question asked, but with something that reframes the question entirely. None of this is accidental. It is deliberate design by people who understood that the most valuable discovery is often the one you were not looking for.
Alexander Fleming was not looking for penicillin. He was looking at a contaminated petri dish. The conditions that produced that observation were not optimised. They were open.
AI systems today are built to optimise for relevance. That is the right starting point and the gains are real. But pushed far enough, a relevance-first system gives every researcher exactly what her query predicted she needed and closes off everything else. The 2026 CHIIR Workshop on Generative AI and Academic Search, which brought together leading researchers in human information interaction and retrieval, raised this directly, examining how GenAI reshapes the cognitive and strategic dimensions of academic search, including how it affects exploratory browsing and the open-ended discovery that has historically driven original scholarship.
This is not a criticism of AI. It is a design gap. And the information profession has spent a century thinking about exactly this gap, building classification systems, browsing structures, and discovery environments around the principle that the unexpected encounter has value.
That thinking, translated into how AI systems surface tangential results and allow for open-ended exploration, is the contribution our profession can make to tools that are already impressive and could be genuinely transformative. We are not asking AI to do less. We are pointing toward what it needs to do next.
Reference:
Liu, Y., Hoeber, O., Arguello, J., Liu, C., Rieh, S.Y., Sinnamon, L. et al. (2026). Report on CHIIR 2026 Workshop on Generative AI and Academic Search (GAI&AS). Proceedings of the 2026 Conference on Human Information Interaction and Retrieval. ACM. doi: 10.1145/3786304.3787953
Full text (open access): arxiv.org/abs/2606.08936
ACM Digital Library: dl.acm.org/doi/10.1145/3786304.3787953