Friction and Finding - The Human Condition

The semester has started and students are back on campus. Mentally I'm like this turtle on a rock in Shinzen Garden in Fresno, California.

The new school year has revived the discourse on generative AI use for teaching and learning. What should the role of generative AI be in higher education? How should instructors use generative AI in their classes? Should students be permitted to use generative AI in their work? To what degree? Approaches fall across a spectrum from encouraging students to use generative AI throughout the course (either due to fatalism or boosterism) to not permitting any use at all.

I've seen many faculty take the measured approach that focuses on students' learning and developing necessary skills - don't use generative AI for activities that prevent building the necessary skills and learning material that are the bedrock of education. We go to school not just for the credential a degree confers, but also for the critical thinking and reasoning. I guess it's what many consider the "scaffolding" necessary to really use generative AI.

Essentially, "Don't use AI to synthesize your readings, don't use AI to do your thinking, and don't use AI for your writing."

Common permissible applications of AI included light editing, certain kinds of data processing, and search and information retrieval.

Of course as a librarian, that last one irked me.

It's really easy to take for granted how to construct a search query. We do it several times a day and our brains are wired to interact with the world that way. We think in keywords and different sources. But this is a skill that has been shaped over time through experience and learning.

Which is why when I talk to (grad) students about researching for a project or literature review, I spend time getting them to interrogate their topic through a power analysis. How does their topic intersect with other fields? What information, research, or data is available? Who else is in the field? Where might you find relevant stuff? These are all things many of us do to some degree when we go looking for information. We think of the appropriate search platform and which sources are credible. And that's a crucial part of the education experience - developing these skills which are an important aspect of critical thinking. Outsourcing search and information retrieval to an LLM isn't much different than using an LLM to brainstorm or write an outline for you.

There's also the crucial issue the LLMs are just the wrong tools to use for search. They are effectively walled gardens that might have somethings you're looking for, but also likely weren't trained on other material that would be relevant. So your results will vary by design. Adding the complication (especially in academic settings) that many content platforms are inaccessible for the LLMs to train from, instead offering their own service (which they want you to subscribe to). So using ChatGPT, Claude, Gemini or the like for your research might seem fine until its not. But if students don't have the knowledge to recognize that limitation, there's a problem!

And the friction of searching - trial and error, dead ends, etc. - is where a lot of learning takes place. Rather than trying act like this stuff is easy, librarians should use this as an opportunity to engage with our communities about why everything is harder than it should have to be. Especially since AI is making access so much worse.