AI in the Library:
An Exploration of New Research Tools

Kyle Feenstra - Coordinator, Learning & Instruction Support

Justin Fuhr - Science & Engineering Librarian

Mê-Linh Lê - Health Sciences Librarian

Session Overview

• Warm up activity
• Large language models (LLMs), and natural language searching
• UM Libraries' AI Research Assistant / Google Scholar Labs
• Elicit
• NotebookLM
• Questions & Discussion

Warm Up Activity

Large Language Models
& Natural Language Searching

...are fundamentally changing information seeking behaviour and related learning processes. 

How do learning and research change when Generative AI is used:

• for information retrieval?
• to summarize or clarify texts?
• as an editor or writing assistant?

• UBC study - How are UBC students using generative AI?

Impact of Large Language Models
& Natural Language Searching on Libraries

Information retrieval processes that have relied on:

• Boolean logic
• Keyword matching
• Subject / Author authorities
• Standardized classification systems and indexes 

Are becoming more automated as they are replaced by:

• Natural language processors trained on Large Language Models (LLMs)
• Algorithms 
• Machine generated indexes

From Advanced Search

to Advanced Algorithms

UM Libraries
AI Research Assistant

Retrieval
Augmented
Generation

(RAG)

Pre-trained Language
& Knowledge Base

Readable Library Collections

LLMs

• Enables the LLM to search library metadata, abstracts, and some full text resources.
• Reduces hallucinations

UM Libraries AI Research Assistant

• Can respond to questions posed in natural language.
• Generates a summary drawn from five academic sources from the Libraries' collection.
• Provides a link to a complete set of search results. 

Limitations:

• Cannot retrieve information from print materials, JSTOR, Elsevier, DataCite, news media sources. 
• Does not search the web.

Google Scholar Labs

• Provides a list of relevant sources from Google Scholar with short summaries for each.
• Helps refine the research question if it is too broad.
• Allows follow-up questions.
   

Limitations:

• We don't know how it works
  (i.e. how it determines relevance).
• Cannot search closed library catalogues or closed databases. 
• Experimental stage.

• What is UML AI Research Assistant good for?
• What is UML AI Research Assistant not good for?

• AI-powered research assistant
• Discovers and summarizes academic sources (methods, results, conclusions)
• Free account (must log in; 10 sources screened per question), with additional paid features.
• Useful for literature reviews, research scoping, and research trends. 

Elicit
https://elicit.com

• Research: find relevant papers and extract findings
• Teaching: prep literature summaries, help students learn research skills.
• Student support: model effective literature review practices.

How faculty can use it

• Ask research question -> screens sources -> extracts data -> creates research report.
• Article summary ("Paper chat").
• Citation and source quality (taken from databases).

Features

Use it as a starting point or to complement your existing literature review search.

• Start with a specific research question.
• Generate report.
  • Review AI-generated summaries against original sources.
  • Check extracted data from original sources. 
• OR view a test search and research report: https://tinyurl.com/catl-elicit

Try it out! https://elicit.com

• What is Elicit good for?
• What is Elicit not good for?

NotebookLM

https://notebooklm.google.com/

• Personalized research assistant and note-taking tool
• Source-grounded: works off documents, files, videos, audio recordings, etc that the user uploads
• 'Chat' with your uploaded content to get grounded information and automatic citations to where in your sources it has gathered the information from

NotebookLM

Privacy and Copyright

• Uploaded information not used to train models ('walled off garden')
• Your queries and data not used to train unless you provide feedback via thumbs up/down  
• Recommend only adding Creative Commons material to abide by copyright and licensing agreements
• Requires Google account; free version has 100 notebooks, 50 sources per notebook, 50 chat queries per day, 3 audio overviews per day

NotebookLM

Features

• Sources can be transformed into learner-friendly resources such as podcasts, mind maps, flash cards, quizzes, etc
• Incredibly helpful for users with different learning styles

NotebookLM

How Faculty Can Use It

Research:

• Thematic analysis across resource types
• Summarize dense text (e.g., policy docs)

Instruction:

• Course preparation
• Create discussion prompts
• Generate quiz questions

Student Support:

• Modify content for different learners

NotebookLM

Access my Notebook

• tinyurl.com/AI-UML-25

OR 

Create your own at notebooklm.google.com 

• Choose a topic you know well
• Upload options
  • 3+ open-access articles 
  • Course lecture notes and/or slides
  • 3+ websites or multimedia links
• Generate content in the Studio and assess its quality and veracity
  • How good are the quiz questions?
  • Does the audio recording properley represent the topic?
  • How accurate is the info provided when you ask questions on the topic?

• What is Notebook good for?
• What is Notebook not good for?

AI Essentials Workshops

tinyurl.com/AI-UML-26

AI Fundamentals March 10 at 12pm

AI in Research: Using AI For Literature Reviews - March 17 at 12pm

AI in Writing: Citations and Attributions - March 24 at 12pm

Questions?