Abstract
We propose that document engagements are palimpsestuous - meaning emerges through layered interactions between viewers and documents rather than inhering within documents themselves. A photograph of a coffee mug prompts the question "Who's Blake?" for one viewer while evoking decades of family ritual, community relationships, and funeral flowers in ice cream containers for another. Traditional cataloging systems capture surface-level nouns (coffee mug, ice cream parlor) while systematically excluding the epidata and proximity relationships where meaning lives.
Through conversations with Claude.ai about photographs - including a Lewis Hine child labor image, everyday objects, and landscapes - we demonstrate that AI systems excel at identifying visible elements but cannot access deeper contextual layers. Words imposed on photographs represent arbitrary viewer reactions rather than extracted native elements. We quantified this distinction using Delta E calculations on pixel variations: a uniform blue sky (low information content, high meaning to an anxious farmer) versus a complex bonfire image (high information content, meaning dependent on knowing it's a football rally ritual). Whether visual complexity is minimal or maximal, the principle holds: description ≠ understanding. As document systems become increasingly AI-driven, preserving pathways to these palimpsestuous layers - the human contextual knowledge transforming description into understanding - becomes crucial.
Recommended Citation
O'Connor, Brian C. and Bonnici, Laurie J.
(2025)
"Artificial Intelligence and the Palimpsestuous Nature of Image Documents,"
Proceedings from the Document Academy: Vol. 12
:
Iss.
2
, Article 11.
DOI: https://doi.org/10.35492/docam/12/2/15
Available at:
https://ideaexchange.uakron.edu/docam/vol12/iss2/11
Digital Object Identifier (DOI)
10.35492/docam/12/2/15