To See or Not to See: Understanding the Tensions of Algorithmic Curation for Visual Arts

Algorithmic recommendation is one of the most popular applications of machine learning (ML) systems. While the implication of algorithmic recommendation has been studied in the context of high-stakes domains such as finance and healthcare, there has been very little focus in understanding its impacts with respect to the arts domain. Given that ML is increasingly finding place in the arts domain such as in generative arts and content analysis, in this paper, we examine the tensions of algorithmic curation in the context of visual arts. Through case studies, we describe how curatorial algorithms that are oblivious of broader socio-cultural contexts could potentially result in ethical concerns such as over-representation and misattribution, to name a few. Towards addressing some of these concerns, the paper offers design guidelines. Specifically, the paper outlines repair strategies that suggest ways 1) to engage with cultural stakeholders in building visual art curatorial algorithms, 2) to unlearn biases embedded in digital artworks and their meta-data, and 3) emphasize the need to establish regulatory norms specific to the use of ML in visual art curation. Taking cue from the process employed by artwork curators, the paper also describes how authenticity can be prioritized by re-calibrating visual art curatorial algorithms. The paper also suggest ways through which the potential of state-of-the-art ML curatorial algorithms can be re-imagined towards empowering the audience of artworks. We hope the insights presented in the paper spark interdisciplinary discussions and pave way for fostering reformation in algorithmic curation of visual arts.