In the not-too-distant past, my library adopted a new discovery system for our catalog access. Right away, my colleagues and I noticed weirdness: some keyword searches would pull up seemingly unrelated items (where a search through the bib record revealed no clues) or an exact title search would display the title we were looking for several items down. I’ve been stumped for why this is happening, and got no good answers from the vendor.
Which is why I’m super grateful for Matthew Reidsma’s recent, very excellent blog post, “Algorithmic Bias in Library Discovery Systems.” It’s long, but worth reading the whole thing. In fact, I’d say it’s a necessity for all librarians—go read it, I’ll wait.
Welcome back! So, to summarize: Reidsma tested Summon’s Topic Explorer function and discovered some odd, leading things, which are the function of how the tool’s algorithm works. For example, a search for “muslim terrorist in the United States” led to a Wikipedia result about “Islam in the United States”—a distressing conflation.
This is why it’s really *not good* for us to present the whole “Internet=bad, Library=good” dichotomy that is so easy to fall into. In many cases, the library search isn’t great either: the algorithms that run it are created by people, and are certainly not perfect, and may reflect the biases or simple not-thinking of the creator. So no matter what the tool is that we use to find information, that question of evaluation is critical: is this actually a reflection of what I was looking for, or does it take a leap (e.g., terrorists –> Islam)? And even if all of the results do seem related, we need to interrogate how they are being displayed: most people are only going to look at (maybe) the first ten results: do they represent a certain viewpoint? These problematic issues exist in ALL library discovery systems, not just open web products like Google.
When I go back to teaching in the fall, I’m going to be sure to teach evaluation outside of the context of a specific search engine—I want students to feel comfortable questioning all information they see, and not to elide that skillset because they implicitly trust that a certain search (i.e., the library search) is more reliable. If our job is to teach students to be critical consumers and creators of information, I’d say that it’s incumbent upon us not to take the easier path, but to surface the way these systems are constructed and the potential for bias and leading that such systems create. Many thanks again to Matthew Reidsma for his excellent article, that highlights the problematic nature of discovery systems.
How will you approach this in your teaching?
For more related research on algorithms, you should check out Safiya Noble’s work on how commercial search engines represent gender and racial identity. It’ll make you stop in your tracks and rethink how you approach teaching searching. Promise.