Operationalising the subjective: Coding visual art with AI

Area of Science:

• Digital Humanities
• Art History
• Computational Social Science

Background:

• The application of Artificial Intelligence (AI) to analyze socio-cultural phenomena in visual art is expanding.
• This field encounters significant methodological hurdles, particularly concerning measurement theory and its practical application.

Purpose of the Study:

• To examine the challenges of systematisation and operationalisation when using AI for visual art analysis.
• To address the critical methodological break from traditional art historical attributes to subjective socio-cultural concepts.
• To discuss ethical and validity concerns related to AI inference of sensitive attributes in artworks.

Main Methods:

• Drawing on contemporary measurement theory frameworks.
• Analyzing the process of systematisation and operationalisation in AI-driven art analysis.
• Investigating the distinction between 'what' is measured and 'how' it is measured in AI contexts.

Main Results:

• A critical methodological shift occurs when moving from objective art historical attributes to subjective socio-cultural concepts.
• AI's inference of sensitive attributes raises ethical and validity concerns, potentially creating problematic measurement instruments.
• The transition requires explicit attention to the definition of measured variables.

Conclusions:

• Researchers must carefully navigate systematisation and operationalisation when applying AI to visual art.
• Explicitly defining what is measured, distinct from the method of measurement, is crucial for valid AI-driven art analysis.
• Addressing ethical implications and validity concerns is paramount for responsible AI research in art studies.