Computing aesthetic taste

Recent breakthroughs in computational and experimental techniques have produced major discoveries in the science of aesthetic taste. This paper reviews key findings from two complementary approaches: machine learning and large-scale human experiments. We argue that together these approaches represent a fundamental shift in computing aesthetic taste, from measuring preferences and choices to reconstructing the complex, high-dimensional representations that generate them. Using these representations, which we refer to as 'taste spaces', researchers can move beyond asking what people like to explaining why they like it, and how aesthetic taste is learned, structured, and transformed across individuals, cultures, and artificial systems. We conclude by discussing how these advances create new opportunities for addressing some of the defining scientific challenges of the coming decades, such as understanding the rich diversity of aesthetic taste across individuals and cultures, and the ways in which artificial intelligence increasingly distorts human taste.