Collaborating for Change? Assessing Metadata Inclusivity in Digital Collections with Large Language Models (LLMs)

This research explores how Large Language Models (LLMs) can be contribute to human expertise and ethical judgment to support reparative archival description practices. We assess the interpretive abilities of three LLMs (Command R+, GPT-3.5 Turbo, and GPT-4o Mini) in assisting humans with metadata inclusivity evaluations. Our testbed comprises a small metadata subset (369 records) from the Robert Langmuir African-American Photograph Collection at Emory University. Despite limited task-specific training and no access to the digital objects associated with the metadata, the LLMs demonstrated notable capacity in identifying gaps, harmful language, and latent contextual elements. By integrating computational methods into descriptive workflows, this research advances Computational Archival Science (CAS), demonstrating how LLMs can enable connecting computational techniques with archival practices in tackling complex, human values-driven challenges.