The role of AI in library services
7 février 2024
Par Susan Jenkins, Borui Zhang
The application of machine learning and other techniques in research offers opportunities for libraries to expand their services.
The AI librarian
Dr. Borui Zhang wasn’t thinking about a job in library services after earning her PhD in theoretical linguistics with a minor in computer science at the University of Minnesota in 2021. While there, she focused on applying machine learning to modeling low-resource languages. Chance intervened when her housemate, an academic librarian, shared a job listing for a new role at the University of Florida George A. Smathers Libraries S’ouvre dans une nouvelle fenêtre and encouraged her to apply.
"Though my title is Natural Language Processing Specialist, my nickname is ‘the AI librarian’. But the nature of this job is really connected to my linguistics training."
Borui ("Bri," as she’s known at UF), was hired as part of a campus-wide initiative by the University of Florida to invest in building AI skills and education across all faculties. The Smathers libraries put forward two positions for the initial hiring campaign, and Borui was the first hire.
Her role strengthens the library’s Academic Research Consulting Services S’ouvre dans une nouvelle fenêtre team, or ARCS. A dozen library professionals with a wide range of information sciences expertise, the ARCS supports students, researchers and faculty from all areas of the university. They consult on areas such as research data management, digital humanities projects and spatial information services, and now, natural language processing and machine learning applications that can empower a variety of research processes.
Making AI approachable for everyone
The library’s approachability makes it an ideal location for offering these services. As Borui says, "It could be any level of student up to postdoc, researcher, or even a faculty person. We’re open to all comers." For students or researchers that don’t have funding for their projects or lack prior experience with AI techniques, the library’s low entry threshold means they can easily explore applications of AI for their research, making it accessible across a range of disciplines.
The techniques of machine learning have become "a great tool for a lot of new researchers to test out research methodologies" that leverage natural language processing. As a result, Borui finds she is supporting researchers in a wide variety of fields. For example, she consulted with a medical research lab to develop a language model that predicts health outcomes for infants, based on using NLP to structure clinical notes.
She has worked on an educational design project with a graduate student in the College of Education for a "hackathon" conference submission and in museum practice with a graduate student-faculty member team from the Florida Museum of Natural History, leveraging large language models (LLMs) and computer vision (CV) techniques to enhance the interpretation of image collections. This diversity makes her job, like many library roles, perpetually interesting because she sees how language — "textual data" — connects her expertise with so many domains.
Borui also participates in the team managing the HiPerGator Sponsorship at the library, a fund named for the university’s famous supercomputer. The sponsorship allows students to request an allocation of computational resources necessary to finish their research projects with AI or data science operations. The graduate student’s hackathon project was a prime example. "We can enable this very junior research team to finish their project so they can present at a conference, just for their professional academic development."
Getting to know AI terminology
Below are some key terms to know for developing competency in AI technologies, and some of their applications.
Machine Learning (ML): Enables computers to learn from and make decisions based on various forms of data including images and text.
Neural Networks: A type of machine learning that mimics the human brain in analyzing and learning patterns from data. Can be applied in learning and predicting user behavior, and for large dataset analysis.
Natural Language Processing (NLP): Computer-human language interaction. Can support search functions and text analysis such as topic modeling, sentiment analysis, named entity recognition.
Language Models: Algorithms for understanding human language. Can enhance library information retrieval, text summarization, and support user assistance chatbots.
Computer Vision: Machine learning for interpreting visual data. Can be applied in digitizing library materials or for advanced image-based searches.
Topic Modeling: A statistical model to discover topics in text-based documents. Useful for organizing digital collections, enhancing metadata, and improving scholarly search tools.
Generative AI: Algorithms, such as ChatGPT, trained on a model of language (text), audio or visual data that creates new text, image, or audio outputs, based on user instruction (prompts). Language-based generative AI assists in summarizing, revising, translating, or developing new texts, while image or sound-based generative AI can be applied to create illustrations, data visualizations, or synthesized music. Multi-modal generative AI algorithms use text to create images.
Building awareness through teamwork
In the centralized UF library system, communication flows easily between different library services, which helps connect students and researchers to the ARCS and Borui’s specialty. "If we have a more complex problem, the structure helps us to work together very quickly to solve it." Liaison librarians are especially key. "When there’s a possible link with AI techniques, the liaison might say, 'Oh, this could be a question for Borui, let me check with her if she's interested.'"
Awareness of the library’s AI services benefits from a strong campus awareness of the library’s resources overall. "Before my arrival, I'm pretty sure researchers have had a lot of collaboration with other folks here. I think there are multiple ways people find me. The library website is also important. If you search, 'I have an AI question', and maybe my name or my department, my profile comes up."
Borui also teams up with colleagues to teach workshops and credit courses for faculty and students to help establish baseline knowledge of how machine learning and language models work. She’s currently engaged in helping an interdisciplinary professional development committee create video-based AI introductions for faculty members from the Health Science Center. "These are geared towards community members who do not have an AI background, but just want to know a little bit more about what AI can do for them, especially generative AI and ChatGPT-related models."
While the library cannot offer credit courses on its own, it does partner with faculties to create new courses, which led to Borui co-teaching a 3-credit course in AI techniques to sophomores with an instructor from the College of Education. "He's also a recent AI hire. We have 68 students from 10 different colleges in the course." This came about when her previous department chair saw a call for new course proposals. She said, "Borui, since you have already connected with AI faculty members, why don't you see if anyone is interested in working with you on this?’ So I reached out to my current collaborator and he happily agreed."
The future for AI in academic libraries
The next hire at Smathers is likely to be a specialist in computer vision, the branch of AI machine learning that trains systems on identifying forms and patterns from images and the visual environment. It has applications across multiple research processes from morphological modeling in geosciences to improving accuracy in medical diagnoses. The ARCS already receives requests for support with this type of machine learning, from researchers processing images. Borui thinks it’s smart for libraries to be involved in providing these services, to ensure they are evolving with the rest of research into the twenty-first century. "In general, AI is just very powerful. It seems to be the trend for every life aspect in the future." She offers the following ideas for how libraries can begin incorporating AI technology in their services:
Learn and assess the AI-related enhancements already appearing in your existing research applications. "I see a lot of liaison librarians helping students and faculty with literature searches, and you can see those search databases have new AI features."
Investigate applying available tools to current tasks. For other types of academic library research support, or even the traditional liaison role, they can begin to adopt AI into their existing domain. "The power of AI can very likely be suitable in all these areas."
Introduce students and researchers early by including discussion of AI research techniques in library orientation programs.
"If we're all very open-minded and willing to try new things to solve the existing questions, I think that's the practical way" to incorporate these capacities into library services.
The connection between libraries and AI literacy is growing. Read more about the role of libraries in the AI transformation on Library Connect.
Dive deeper into AI topics for librarians:
Key components of developing AI literacy at your institution covers the fundamentals you need to begin learning more about AI and support your user's education.
Addressing bias in AI-fueled knowledge systems discusses the way librarians and researchers are working together to combat the biases in AI models.