Artificial intelligence (AI) is transforming how knowledge is created, accessed, and preserved. As the University of Southern California takes up these questions this week at its first USC AI Summit, the USC Libraries stand at the center of the conversation. Business librarian Benjamin Hall will take part in the summit, presenting on “Information Literacy in the Age of AI,” one example of how USC librarians are helping the university community navigate this rapidly evolving landscape. For centuries, libraries have helped scholars navigate each new information revolution, from cataloguing the printed book to mastering the digital database. Librarians and library personnel aren’t just observers of this latest transformation; they’re essential partners, helping students and other researchers think critically about the technologies revolutionizing life and learning today. We asked several of them to share their perspectives —and how to approach these technologies thoughtfully.
What’s the difference between using a library database and asking ChatGPT a question?
Kelsey Vukic (head, Social and Behavioral Sciences Library Services): Generative AI systems such as ChatGPT are trained on large language models, which makes them adept at recognizing patterns and “guessing” how text might unfold. While they might be able to answer some questions, they typically are not transparent about the data used to generate their responses. Library databases, by contrast, provide access to research materials and other publications that present evidence or findings from verified research. While you cannot ask a library database “why is the sky blue?,” you can, using appropriate search terminology, find publications that explain why the sky is blue and cite their evidence, and follow the ongoing scholarly conversation on the topic.
How do I pick an AI tool?
Clarissa Moreno (social and behavioral sciences librarian): Keep in mind that while USC offers access to some AI tools such as Copilot and Gemini, the USC Libraries do not endorse any specific tool, and most are not designed for research purposes. To choose the right tool for you, first consider what you want to achieve—feedback, organization, or something else entirely. Then evaluate the tools available to you, factoring in access through USC, cost, usage limits, and features. It’s also important to see whether the tool fits your workflow or integrates with your existing software. Finally, try your top choices on a small task to see how they suit your needs. Each tool interacts differently with its users, so there’s no one-size-fits-all solution—yet.
Are there AI tools in the libraries that I can use?
Moreno: At this time, the USC Libraries do not officially endorse any AI tools. While some of the databases are beginning to offer AI or “research assistant” features, it is worth noting that AI cannot replace the critical role of the human researcher. Before using any AI tool for research help, you should be confident in your own research skills, as research is a very personal process and only you can truly understand what it is you are looking for.
How do I cite AI tools like ChatGPT or Claude in my papers?
Vukic: APA Style recently updated its guidelines for citing generative AI. Previously, it was standard to cite the AI tool, not the individual chat session. However, since many AI tools now make it possible to share individual chat sessions, APA now allows both in-text citations and references when doing so benefits the reader. Follow the author–date–title–source format. See the APA Style Blog for sample citations.
How are USC librarians working together to better meet the challenges and opportunities related to AI?
Michaela Ullmann (head, Instruction & Assessment): Many librarians at USC bring expertise in the use and ethical implementation of AI, offering workshops and instruction sessions on the critical use of AI in research as well as on discipline-specific AI tools. Beyond their individual and subject-focused initiatives, librarians and staff collaborate across the USC Libraries—through groups such as the USC Libraries AI Taskforce and the AI Information Literacy Working Group—to develop guidance for students and scholars on the responsible use of AI in research. These collaborations address topics such as the ethical and legal implications of training large language models on copyrighted materials, the review and assessment of AI research assistants in vendor products, and the creation of tutorials and lesson plans on the critical use of AI for information literacy instruction. USC librarians also contribute to university-wide and national discussions on AI literacy, helping shape how libraries prepare students and researchers for an evolving information environment.
Can AI discover connections across disciplines I might otherwise miss?
Danielle Mihram (digital scholarship librarian): Yes—thanks to its ability to analyze massive datasets, identify complex patterns, and bridge disparate fields of study at a scale and speed impossible for human researchers alone. While AI is a powerful tool for discovery, human judgment remains crucial for interpreting AI-generated insights, providing context, and guiding the meaning of the discoveries. The partnership between human researchers and AI systems creates opportunities for major scientific and creative advances that neither could achieve alone.
One example is the intersection of AI and historical research, which is rapidly evolving and transforming how researchers access, analyze, and interpret history. A 2025 article in Historica notes that AI tools—from large language models to advanced data-processing pipelines—are no longer experimental; they are actively integrated into workflows, enabling historians to tackle complex questions, process vast multilingual corpora, and uncover connections that might otherwise remain hidden. The project “Mining the Dispatch” exemplifies this trend. It explores the dramatic and often traumatic changes in the social and political life of Civil War Richmond, analyzing nearly the full run of the Richmond Daily Dispatch from the eve of Lincoln’s election in November 1860 to the evacuation of the city in April 1865. Its principal methodology—a computational, probabilistic technique—uncovers categories and discovers patterns across the text corpus. For a comprehensive coverage of AI and the Digital Humanities, see my research guide on the subject.
How will AI be used to preserve cultural heritage collections at libraries and archives?
Sam Gustman (associate dean for library IT and chief technology officer): We're likely to see AI agents monitoring signs of degradation over time in important cultural collections. For example, we know that tape recordings start to rot after 20 years and show age-related damage. AI agents can assist with monitoring large-scale audio and videotape collections at libraries and archives and issue warnings based on the ages of individual tapes. For digital collections, AI agents can do something very similar. We give every digital video file in our USC Digital Repository a unique fingerprint that helps us monitor its integrity. An AI agent could monitor fingerprints across a massive, petascale digital collection to look for early signs that individual files are starting to no longer match their preservation fingerprints.
How do archivists use AI tools to make archival collections more accessible to students and other researchers?
Bo Doub (accessioning archivist): As an accessioning archivist, I use AI chatbots primarily as data transformation tools to enhance descriptions of archival resources. When the USC Libraries acquire archival collections, each acquisition often includes notes and inventories describing the contents to varying degrees. My job is to contextualize and structure these descriptions into published finding aids that allow researchers to discover and use the materials in their work. I use ChatGPT to transform unstructured inventories—often legacy Word documents written in narrative form by collections’ creators or their family members—into structured data in Excel. Once the data has been organized into a standardized template that separates descriptive elements such as folder titles, creation dates, physical descriptions, and extents, I import the resulting metadata into the collection's published finding aid. This AI-assisted workflow streamlines accessioning and processing by reducing manual data entry while also expediting researchers' access to newly acquired collections.
Should libraries preserve large language models as cultural artifacts?
Two librarians offered different perspectives.
Bridgid Fennell (education librarian): Libraries can contribute to the large language model ecosystem, but preserving the full corpus of training data is not the best use of our resources. The Internet Archive already collects publicly accessible digitized media ingested by LLMs, while libraries collect or subscribe to high-quality, user-centered resources that have been appropriated outside of the copyright protection. The utility of LLMs lies in proprietary algorithms; Stanford’s Foundation Model Transparency Index illustrates AI tech giants' opaque practices and data source acquisition. Projects like the Public Interest Corpus—a collaboration between the Author’s Alliance and Northeastern University—invest in AI technologies that value intellectual and creative producers while serving the public good, ensuring that corporations don’t exploit knowledge production or monopolize innovation.
Deborah Holmes-Wong (director, USC Digital Library): Everyone is thinking about LLMs when they say “AI.” AI is more than just generative AI and LLMs. There are other types of models like computer vision which is used for facial recognition, finding tumors, and "reading" license plates. There are models used in security systems to identify anomalies and sound alarms; by banks to approve loans; by universities to identify students; and by employers to select job candidates. These systems use probability to produce results. Biases or errors in either the model or training data can harm large groups of people by denying them loans, education, and jobs. These models, along with their training data, need to be archived so that their impact can be assessed in the future.
We hear a lot these days about critical AI literacy. What unique role can library professionals play in fostering this kind of literacy?
Curtis Fletcher (director of the Ahmanson Lab at the USC Sidney Harman Academy for Polymathic Study): Effective critical AI literacy requires building on the principles of information literacy as practiced in academic libraries. Library professionals offer a unique model of information literacy, one that is focused on the full life cycle of information in order to foreground the social, cultural, and technical systems that determine how it is collected, accessed, and granted authority. Concepts central to this work, such as data provenance, collection bias, and even epistemic power, provide crucial tools for understanding how AI systems are built, how they function, and what assumptions they carry. Amid today’s relentless AI hype and adoption, this perspective is essential to developing critical AI literacies that surface the ways in which technical infrastructures, social hierarchies, and cultural narratives shape the data that feeds contemporary AI systems, the broader knowledge environments within which they now operate, and the power dynamics and epistemic authority we increasingly grant to the corporations that develop them. Put simply, library professionals are specially trained and uniquely equipped to think and teach about socio-technical information systems—and that is precisely what current generative AI models are.
How can AI help students engage more deeply with digital collections and other library resources?
Mike Jones (director of web and automation technologies): AI has incredible potential to help students and libraries foster deeper, more meaningful engagement with library collections and resources. By approaching AI as an assistive tool rather than a means of outsourcing research, students are empowered to explore the USC Libraries' collections in a more intuitive and accessible manner. For the libraries, AI opens opportunities to develop new services and tools, such as our Interactive Interviews, that lower barriers to discovery, creating unique and personalized learning experiences that align with each student's individual interests.