Harvard Library Tests AI Tools to Help Researchers Navigate Collections

Harvard Library is beginning to integrate artificial intelligence into its systems in an effort to make its vast collections easier to search and analyze.

The effort is being led by Stuart Snydman, the managing director of library technology, who is now overseeing new tools designed to help students and researchers navigate Harvard’s extensive archives and special collections.

“AI can help libraries make complex collections easier to find and use — without lowering standards or compromising the trust you have in the library,” Snydman wrote in a statement.

One of the library’s major projects is Collections Explorer, a developing system that uses semantic search to allow users to find materials across Harvard’s special collections. Unlike traditional keyword-based searches, the tools allow researchers to search using natural language and can surface relevant materials even when specific terms do not appear directly in catalog descriptions.

The system is still in development and has not yet been widely released. Since early testing began, the library has worked with more than 150 Harvard students to evaluate the platform and gather feedback on its usability.

Another initiative focuses on integrating library search capabilities directly into AI chat tools. A technology known as model context protocol could allow platforms such as ChatGPT, Google Gemini, Claude, or Harvard’s AI Sandbox to link users directly to Harvard Library materials when answering questions.

The goal, Snydman wrote, is to ensure that researchers who begin their work with AI-powered tools can still access reliable and verifiable academic sources that inform AI responses.

“We think that’s important for trust and for ensuring people can verify what they’re finding,” he wrote.

Beyond improving search, AI could also expand how researchers work with historical materials. New systems are making it possible to automatically transcribe handwritten documents — work that has traditionally required extensive human labor or crowdsourcing efforts.

Still, Harvard’s experiments with AI remain largely in early stages.

“It’s important to say that much of this work is still in development or pilot phases,” Snydman wrote. “We’re intentionally experimenting before moving anything broadly into production, and we’re highly focused on checking in with and testing with students and faculty to make sure we deliver the highest value as we integrate AI.”

Even as Harvard explores new digital tools, Snydman emphasized that the library’s core mission remains unchanged: connecting scholars with reliable information and preserving historical materials.

The AI systems currently being developed at Harvard only interact with digital collections and metadata, according to Snydman.

“Our goal in enhancing our systems with AI is, in part, to help people find and use the physical materials themselves,” Snydman wrote.

Harvard has also long been a leader in digital preservation. The University’s Digital Repository Service, launched in 2003, was one of the first systems of its kind designed to ensure that digitized materials remain accessible. The repository is now undergoing a major upgrade as the library prepares to incorporate new AI-enhanced tools.

For Snydman, who has worked in library technology, for nearly 25 years, the rise of AI represents the latest in a series of technological shifts.

“I’ve watched wave after wave of digital tools reshape the academic process, from personal computers to mobile devices to the internet, to Google, and now AI,” Snydman wrote.

Over the longer term, he wrote, the most important challenge for libraries will be maintaining their core values while adapting to new technologies.

“My decade view is less about predicting a particular tool or innovation,” Snydman wrote, “and more about ensuring that great libraries like Harvard keep innovating, while staying grounded in the values and expertise that help scholars and students do great work.”

—Staff writer Shalini N. Ramchune can be reached at [email protected] and on Signal at @shaliniramchune.03. Follow her on X @ShaliniRamchune.