Efficiency Gains in Rare Book Assessment: Evaluating Generative AI as an Adjunct Approach

Abstract

Months of environmental fluctuations in the rare book storage area compelled the Preservation Department to conduct a condition assessment. To limit the possibility of inattentional blindness and to test the quality of generative AI systems two different generative AI systems were tested against well-trained humans. The test consisted of one-hundred books which were evaluated by humans for damage to the text-block, the spine, the margin/gutter and the paper. The areas were photographed, those photographs and uniform text were input into Gemini Pro and ChatGPT Pro. The human results were compared against Gemini Pro and ChatGPT Pro. Considering generative AI was not designed to perform rare book assessments, the AI systems performed better than expected evaluating the text block, the condition of the paper and suggested preservation actions.