Manifest for Generative AI Rare Book Assessment Data Set
Master_Assessment_List.csv
ChatGPT_Comparison.csv
Gemini_Comparison.csv


State of Data: Cleaned
Instruments used to create data: Excel, Gemini, ChatGPT


Processing Steps:
100 rare books chosen at random
Each book was photographed six times in the following areas:
1. Front cover and spine
2. Front endpapers (both)
3. First random set of pages
4. Second random set of pages
5. Back endpapers (both)
6. Back cover and spine


The following data was collected about each book:
1. Barcode
2. Condition of the text block
3. Gutter margin width
4. Paper type and condition
5. Binding type and condition
6. Suggested preservation/conservation actions
7. Damage prior to assessment (Y/N)
8. Foxing Present (Y/N)
9. Prior Water Damage (Y/N)


A human experienced with assessing the condition of rare books evaluated each book.  Six photographs were taken of each book and were fed into Gemini and ChatGPT to assess for the same assessments that the human performed.  
The prompt that was used:
“Using the uploaded images as a guide generate a preservation assessment report using the following parameters:
Condition of the text block
Gutter margin width
Paper type and condition
Binding type and condition
Suggested preservation and conservation actions for this book”
Explanation of column headers:
The column headers are identical to the ones identified above, and each should be self-explanatory


Software required to view:
Any that can open comma separated value files


Licensing:
Open access, free to use


Funding Source:
None


Contact Information:
Mark Coulbourne
kcoulbou@umd.edu