In the Summer of 2024, Memorial University Libraries put together an AI Strategic Project Team. The Muse sat down with Chelsea Humphries, the Chair of the Team, to discuss AI and its impact on the libraries and students.
The team was developed “to support learning and information sharing both within the university as a whole around AI” says Humphries. It helps library staff and students “to learn more about AI, how it functions, and how we can make more informed decisions.”
Humphries says that MUN Libraries itself doesn’t “necessarily [use] AI, but [they] do a lot of work around AI literacy. [They] provide resources and instruction to both faculty and students.”
However, there are some specific tools that have new AI features. There are currently “a number of databases that the library subscribes to that have brand new generative AI features in them.
Some examples include Scopus and Web of Science. These are database subscriptions which use generative AI as a research tool that acts as a sort of “super-powered version of predictive text,” but the AI is an “add-on” rather than a requirement to use the database.
These AI add-ons are large language models (LLMs). Scopus and Web of Science are LLMs that are trained on private data, whereas LLMs like ChatGPT are trained on public data. There are some important differences between the two that students should know about.
Firstly, Scopus and Web of Science are “drawing from the scholarly material actually in [their] databases, instead of ChatGPT drawing from the entire scraped internet,” says Humphries.
This makes for a better research tool because of what Humphries calls “retrieval augmented generation (RAG).”
What RAG does is make sure that the output of the AI is rooted in the private database it was trained on. This roots it in the scholarly material and makes the AI less likely to “hallucinate.” These LLMs are also better at providing citations.
Secondly, LLMs like ChatGPT collect users’ data, whereas Scopus and Web of Science do not. “The default setting if you’re using ChatGPT is that it will retain anything you type into it, and anything it spits out. It will use that to continue training the tool for other people.”
“We in the libraries encourage anyone using these tools to refrain from sharing anything super personal. If you’re doing research, consider [it] your own intellectual property. You don’t necessarily want [your work] to be saved in a big generative model.”
Humphries also states that users of these tools should be cautious in giving copyrighted materials to generative AI. Amid ongoing legal battles, “we don’t have any definitive rulings on whether copyrighted content [is] legally allowed to be used… or if that’s copyright infringement.”
This is to say that there could be copyright implications for sharing such material with AI.
Humphries argues that while AI can be used as a useful research tool, it doesn’t replace having actual human input on a topic.
If students were to instead “ask a librarian, there’s a good chance you’ll get a really robust answer that won’t have the same risk of leaving information out.”
“We are all still the experts and the authorities on the topics that we are doing our work on and the things that we are investigating.”
