
Orland Hoeber (University of Regina, Canada) has spent much of his academic career looking at why and how we search for scholarly content. Should we have one search interface for everything? Clearly not, but what kind of searching is most appropriate for academic tasks? Hoeber has created a new search interface to deal with the kind of searches – and capability of this tool to remember what you have searched for.
But it’s not just remembering searches, it’s what you search for initially. Google-type simple search, for a name or a date, covers as much as 99% of our searching, but that remaining 1% requires a completely different approach. Traditionally, this was handled by the reference desk in the library, but you don’t often see these nowadays. Hoeber calls the skill “library search literacy”. To an extent, RAG refines and improves a simple search, but it doesn’t help when the user doesn’t know what they are looking for – it may even be a drawback. Memorably, in a recent interview, Hoeber summed up the process when he said “we’re providing answers where the result might be knowledge rather than answers”.
The obvious question is, does GenAI provide knowledge?
Retrieval augmented generation and a lot of the GEN-AI type approaches to searching are very much focused on “let’s turn every query into a question and let’s turn every response into an answer”. And from a learning perspective, if I just gave my students, “here is a question and here’s the answer”, and we jump completely over the process of learning how to get to that answer, how much of that answer is going to be retained after the class? You know, very little.
Hoeber has created an alternative model, called a “multi-workspace”, that enables several search results, that would normally be held in separate tabs, to be combined. This doesn’t involve a rejection of AI, rather a use of generative AI but keeping control, thereby avoiding the danger of “cognitive offloading”, letting the AI reach all the answers without your involvement. The result can be trialled in a new product, Dilex Search (although currently only available on ExLibris and Primo), Among other things, it provides cross-session searching (remembering what you searched for earlier), as well as cross-device searching (continue on your laptop what you started on your phone).
One intriguing observation about Dilex Search is that it is not prioritised to find the fastest results, but to enable the user to think through the problem better. This raises the question: “What are those spaces where we need thoughtfulness and what are those spaces where we need fast?”
Dilex Search is worth playing around with, even if you have no intention of changing the way you search, simply because innovations in search are rare, and accessible working examples you can play with are even rarer. We can all benefit from exposure to different search tools.
Longer-term, Hoeber envisages a system that builds a profile for the user of what they know, based on their search activities over a long period. I find intriguing the idea of a system being aware of my progressive growth in understanding as I become more familiar with a particular domain. The goal, describes Hoeber, is not to have the human in the loop, but to have the AI in the loop of human intelligence: in other words, AI in the loop.

Leave a Reply