Layers of trust: AI assisted data management, transcription, and discovery in an audio digitization project.

Cooperstown Graduate Program and Milne Library recently digitized a large collection of reel-to-reel historical recordings. The resulting collection posed a scale problem that no small archival team could manage by hand: metadata in inconsistent formats across systems, thousands of untranscribed recordings, and for many items, incomplete information of what the tape actually contained. This case study examines how AI was built into that workflow, not as one tool applied uniformly, bust as a set of capabilities matched to different levels of trust.

For metadata already known to be correct, hallucination risk was unacceptable. AI was used to write Python scripts that moved and reformatted metadata between systems, but the scripts, not the AI itself, touched the data at runtime, keeping every transformation transparent and auditable.

For the larger problem of undocumented audio, we turned to AI transcription, testing two local AI models that provided privacy and security because the model was running on our own infrastructure. An AI transcript with accuracy around 95% isn’t perfect, but it’s a dramatic improvement over not knowing a tape’s contents at all, and it makes systematic human review realistic for the first time.

At our lowest confidence tier, we experimented with AI-generated abstracts and metadata for items with no human created descriptions. These will be useful as a starting draft and for discovery, but require a human review gate before anything enters the permanent record.

This talk will explore each tier, the reasoning behind it, the tools used, and where each one succeeded or fell short. The larger argument is that responsible AI adoption in research and cultural heritage work depend less on a single “use it” or “don’t” policy and more on matching AI’s role to how much verification a given task allows. Using a case study will allow participants to start to develop their own mental model for evaluating where AI belongs and where it doesn’t.

Speaker Biography

Ed Beck is the Open & Online Learning Specialist at SUNY Oneonta, where he supports faculty and students in designing engaging, accessible digital learning experiences. His work spans online learning, open educational resources, artificial intelligence, and digital literacies, helping educators adopt new technologies in practical, ethical, student-centered ways.

He is co-founder of SUNY Create, an initiative helping campuses use open-source tools to support digital portfolios, course projects, and public-facing scholarly work, with a focus on building authentic learning opportunities and durable, transferable skills in students.

Ed brings a collaborative approach grounded in openness, innovation, and care for the learner experience, and regularly leads professional development on digital pedagogy, emerging technologies, and inclusive teaching. In 2021 he received the SUNY FACT2 Award for Instructional Support; in 2026 he begins his first term as FACT2 chair.