A common refrain about artificial intelligence in education is that it’s a research tool, and as such, some school librarians are acquiring firsthand experience with its uses and controversies.

Leading a presentation last week at the ISTELive 25 + ASCD annual conference in San Antonio, a trio of librarians parsed appropriate and inappropriate uses of AI in a series of hypothetical scenarios. They broadly recommended that schools have, and clearly articulate, official policies governing AI use and be cautious about inputting copyrighted or private information.

Amanda Hunt, a librarian at Oak Run Middle School in Texas, said their presentation would focus on scenarios because librarians are experiencing so many.

“The reason we did it this way is because these scenarios are coming up,” she said. “Every day I’m hearing some other type of question in regards to AI and how we’re using it in the classroom or in the library.”

• Scenario 1: A class encourages students to use generative AI for brainstorming, outlining and summarizing articles.

  Elissa Malespina, a teacher librarian at Science Park High School in New Jersey, said she felt this was a valid use, as she has found AI to be helpful for high schoolers who are prone to get overwhelmed by research projects.

  Ashley Cooksey, an assistant professor and school library program director at Arkansas Tech University, disagreed slightly: While she appreciates AI’s ability to outline and brainstorm, she said, she would discourage her students from using it to synthesize summaries.

  “Point one on that is that you’re not using your synthesis and digging deep and reading the article for yourself to pull out the information pertinent to you,” she said. “Point No. 2 — I publish, I write. If you’re in higher ed, you do that. I don’t want someone to put my work into a piece of generative AI and an [LLM] that is then going to use work I worked very, very hard on to train its language learning model.”

• Scenario 2: A school district buys an AI tool that generates student book reviews for a library website, which saves time and promotes titles but misses key themes or introduces unintended bias.

  All three speakers said this use of AI could certainly be helpful to librarians, but if the reviews are labeled in a way that makes it sound like they were written by students when they weren’t, that wouldn’t be ethical.

• Scenario 3: An administrator asks a librarian to use AI to generate new curriculum materials and library signage. Do the outputs violate copyright or proper attribution rules?

  Hunt said the answer depends on local and district regulations, but she recommended using Adobe Express because it doesn’t pull from the Internet.

• Scenario 4: An ed-tech vendor pitches a school library on an AI tool that analyzes circulation data and automatically recommends titles to purchase. It learns from the school’s preferences but often excludes lesser-known topics or authors of certain backgrounds.

  Hunt, Malespina and Cooksey agreed that this would be problematic, especially because entering circulation data could include personally identifiable information, which should never be entered into an AI.

• Scenario 5: At a school that doesn’t have a clear AI policy, a student uses AI to summarize a research article and gets accused of plagiarism. Who is responsible, and what is the librarian’s role?

  The speakers as well as polled audience members tended to agree the school district would be responsible in this scenario. Without a policy in place, the school will have a harder time establishing whether a student’s behavior constitutes plagiarism.

  Cooksey emphasized the need for ongoing professional development, and Hunt said any districts that don’t have an official AI policy need steady pressure until they draft one.

  “I am the squeaky wheel right now in my district, and I’m going to continue to be annoying about it, but I feel like we need to have something in place,” Hunt said.

• Scenario 6: Attempting to cause trouble, a student creates a deepfake of a teacher acting inappropriately. Administrators struggle to respond, they have no specific policy in place, and trust is shaken.

  Again, the speakers said this is one more example to illustrate the importance of AI policies as well as AI literacy.

  “We’re getting to this point where we need to be questioning so much of what we see, hear and read,” Hunt said.

• Scenario 7: A pilot program uses AI to provide instant feedback on student essays, but English language learners consistently get lower scores, leading teachers to worry the AI system can’t recognize code-switching or cultural context.

  In response to this situation, Hunt said it’s important to know whether the parent has given their permission to enter student essays into an AI, and the teacher or librarian should still be reading the essays themselves.

  Malespina and Cooksey both cautioned against relying on AI plagiarism detection tools.

  “None of these tools can do a good enough job, and they are biased toward [English language learners],” Malespina said.

• Scenario 8: A school-approved AI system flags students who haven’t checked out any books recently, tracks their reading speed and completion patterns, and recommends interventions.

  Malespina said she doesn’t want an AI tool tracking students in that much detail, and Cooksey pointed out that reading speed and completion patterns aren’t reliably indicative of anything that teachers need to know about students.

• Scenario 9: An AI tool translates texts, reads books aloud and simplifies complex texts for students with individualized education programs, but it doesn’t always translate nuance or tone.

  Hunt said she sees benefit in this kind of application for students who need extra support, but she said the loss of tone could be an issue, and it raises questions about infringing on audiobook copyright laws.

  Cooksey expounded upon that.

  “Additionally, copyright goes beyond the printed work. … That copyright owner also owns the presentation rights, the audio rights and anything like that,” she said. “So if they’re putting something into a generative AI tool that reads the PDF, that is technically a violation of copyright in that moment, because there are available tools for audio versions of books for this reason, and they’re widely available. Sora is great, and it’s free for educators. … But when you’re talking about taking something that belongs to someone else and generating a brand-new copied product of that, that’s not fair use.”