For the HBLL AI Learning Task Force

How BYU-Idaho Library is approaching AI

Mat Miles · BYU-Idaho Library · milesm@byui.edu

Thank you for reaching out, and for the work your task force is taking on. I'm glad to share what we're doing at BYU-Idaho.

This page follows your four questions in order. The lesson plans, assignments, and reports linked below are yours to download and adapt however they help. They are PDFs here; if you would like the editable Word or PowerPoint versions, email me at milesm@byui.edu and I will send them.

1How we keep staff trained and capable

You askedWhat is your staff doing currently to maintain training and capability in the use of AI?

We work at a few levels at once, from library-wide coordination down to one-on-one help.

  • CoordinationA library AI group coordinates our overall approach and keeps our efforts pointed in the same direction.
  • Staff trainingsWe hold regular trainings for library employees so everyone has a working familiarity with the tools and the issues.
  • In the classroomAt a faculty member's request, librarians co-teach AI-integrated research sessions inside their courses. Each one pairs a library session with a student assignment. We take the professor's existing assignment, keep its requirements, and weave in critical use of generative AI.
  • One-on-oneI work as a hands-on consultant and mentor for staff and faculty. Much of that is helping professors integrate AI into their curriculum and use AI to build curriculum.

Four principles I teach

  1. How generative AI works.

    Large language models are prediction engines: neural networks whose weighted connections are strengthened during training. I teach this with a three-round prediction game.

    • A nursery rhyme: "Twinkle, twinkle, little star, how I wonder…" Everyone finishes it instantly.
    • An open-ended prompt from the course topic, where answers spread out.
    • "I, Nephi, said unto my father…" (1 Nephi 3:7). The class and the AI both complete it. When asked for the exact reference and the next few verses word for word, confidence drops, and the AI may quietly change small words while sounding just as sure.

    I also show that an LLM is not a deterministic program. A simple math program gives the same result every time. An LLM given the same question gives similar answers, but never exactly the same answer twice.

    A program A = 3, B = 5
    C = A + B
    → C = 8, every run
    An LLM Same question, asked again
    → similar answer,
      never the exact same one

    Finally, I show how AI has moved from pure prediction to problem solving. Newer reasoning agents, such as Gemini Deep Research, still predict the next word underneath, but they work in a loop. They make a plan, search, read what they find, notice a dead end, change their search terms, and try again. A prediction model guesses the whole path through a maze at once; an agent walks it and backs up when it hits a wall. The loop makes AI more persistent, not more truthful, so students still trace every claim back to the original source.

  2. A sliding scale of subjectivity.

    The more subjective the topic, the more critical you need to be. On objective topics like glycolysis or gravity, AI is usually on solid ground. On contested issues, its output is one viewpoint among many. Numbers and citations always need to be checked.

  3. Have AI verify itself.

    Ask it to check its own work. Then verify it yourself against the original source.

  4. Have AI interview you before it completes a task.

    Let it ask the questions it needs answered first. The result fits your situation much better.

2What resources we have

You askedWhat kinds of resources do you have available?

On campus

AI-integrated assignments

These are real sessions we've taught. Each comes with an instructor session plan (timing table, talking points, demo prompts, checklist, rubric, and troubleshooting) and a student assignment.

County health data + AI

Public health course · PolicyMap + Gemini · 90-minute session

Students evaluate a PolicyMap county health report with Gemini.

Finding and assessing original research

ESS 111 Introduction to Health Careers · NotebookLM + Gemini · 50 minutes

Students locate original research articles and use AI to help assess them.

AI in research

Political science course · Gemini or ChatGPT, Gemini Notebook, Library Search, EndNote Fusion · 90-minute class session

Seven principles for using AI in research, practiced in one pipeline. Students map a problem with a chatbot, search the library, save sources to EndNote Fusion, read their PDFs in Gemini Notebook, and build a first bibliography. Includes a self guided student workbook.

Faculty consulting

A few examples of the AI work I've done with faculty:

Measuring Student Learning in the Age of AI

Research report · prepared for the BYU Provo Library Committee

  • AI-assisted systematic review screening for a sports science research manuscript, with a documented AI audit trail and an AI-use disclosure.
  • AI-supported Canvas course redesigns.
  • AI-assisted drafting of an open textbook chapter with a faculty author.

3Where I look beyond the university

You askedAre your resources mainly based at BYU-Idaho, or do you also include trainings from other organizations (Meta, Google, ALA, SAA, etc.)?

My answer here is personal. I don't rely on formal vendor or association courses. What actually keeps me current is this:

  • YouTube. This is where I learn the most, especially for seeing new tools and techniques in action.
  • Ethan Mollick's One Useful Thing. Consistently practical thinking about how to use AI well.
  • Current reporting. I follow AI coverage in the Wall Street Journal, The Atlantic, and the New York Times.

I read about AI trends in general, not only in education. The most useful ideas for libraries often come from seeing how AI is changing work everywhere else.

4My advice for your task force

You askedAny advice about where we could look for further training focused on library staff and faculty beyond university-provided training?

  1. Learn by doing real work. Pick a task you actually have this week and use AI on it. That teaches more than any overview course.
  2. Tie training to real assignments and workflows. Staff learn fastest when the training is built around something they already do, rather than a generic course.
  3. Follow a few good thinkers and the news. Look beyond education sources. How AI is changing other fields will tell you a lot about where libraries are headed.
  4. Give people someone to go to, and teach verification alongside tools. A hands-on person or a small community of practice keeps momentum going. Pair every tool with habits like the four principles above, so people learn to question the output as well as produce it.

I'm happy to talk through any of this or share more. You can reach me at milesm@byui.edu.