Operationalising the subjective: Coding visual art with AI.

The use of Artificial Intelligence (AI) to study socio-cultural phenomena in visual art is a rapidly growing field, yet it introduces fundamental methodological challenges around measurement. Drawing on recent debates surrounding measurement theory frameworks, this article examines the challenges of systematisation and operationalisation when using AI to analyse visual constructs in artworks. We argue that the shift from well-defined art historical attributes to subjective and contestable socio-cultural concepts marks a critical methodological break, demanding explicit attention to what is being measured, separate from how it is measured. Further complicating matters, ethical and validity concerns around AI inference of sensitive attributes mean that a seemingly well-intentioned research question may give rise to problematic measurement instruments. We discuss these challenges and how researchers may navigate systematisation and operationalisation with AI for visual art.