Humanist-in-the-Loop: Machine Learning and the Analysis of Style in the Visual Arts

This chapter examines theories about artistic style and explores the extent to which they can be used support the computational modeling of painted surfaces. It is argued that style comprises more than the visible features of paintings and that this poses specific challenges to the algorithmic analysis of artworks. Beyond acts of categorization, can computer vision challenge fundamental ideas about style and its role in art historical analysis? What kinds of questions do computer scientists need to ask in order to create meaningful data sets and to undertake stylistic analyses of the works contained in them? How should statistical anomalies be dealt with and do they have significance beyond their mere outlier status? In proposing answers to these questions, the chapter defends a contextual approach to the analysis of style. It is argued that close collaboration between computer scientists and art historians is required in order to embed the results of quantitative analyses in relevant socio-cultural frameworks and to move beyond the instrumentalization of existing formal or morphological models.