AI Evaluation in Action: Discovering and Combating AI Hallucinated Citations

Students and faculty find themselves in a precarious position, as they are both encouraged to integrate AI tools into their research and teaching practice, but also remain wary of the accuracy and reliability of their outputs. As conversations shift to the importance of evaluating and verifying information provided by AI tools, we need to put these practices into habit. This presentation will explore a key element of AI literacy: the evaluation process. In this session, attendees will learn about the very real issue of AI hallucinated citations and walk away with a verification process to use with patrons turning to LLM AI tools as research partners in locating scholarly sources.

As researchers increasingly turn to large language model tools to try and discover information on a topic, they are encountering hallucinated citations masking as real, legitimate sources. Hallucinated citations are citations provided by large language models that do not actually exist. They may be made up of real and imagined elements, but do not correspond with a real world source. Reference accuracy is a vital element of scholarship. There are already documented cases of hallucinated citations making their way into peer reviewed scholarly literature, legal briefs, and, worryingly, into Google Scholar. If not caught, these hallucinations can propagate and amplify misinformation, poison scholarly databases, erode trust in the research process, and lead to reputational consequences for scholars. 

This webinar will help librarians understand what AI hallucinated citations are and how they can go about helping patrons verify sources provided to them by AI tools. Attendees will be taken through a step-by-step verification process that they can replicate with patrons to ensure that sources provided by large language models are real, legitimate, and relevant. While many librarians are coming to understand the limits in scope and ability that popular AI chatbots offer as research tools, this workshop will also encourage them to hold critical conversations with patrons as they work to assess AI outputs for research.

Presenter: Natalie Ornat-Bitting is the Humanities Librarian and Associate Professor at UNC Charlotte.