Criticism, essays, etc.
This article considers how an AI might approach the problem of art appreciation. The article briefly covers (a) necessary background in aesthetic theory, (b) the development of modern and contemporary art, (c) identifies human agents, such as curators and writers, who influence collective opinion in the Artworld, and (d) discusses the recursive process by which these agents look and think about art, thereby arriving at value judgements. I then, as a thought experiment, suggest what key features an articial aesthetic agent might require, alongside what limitations, possibly conceptual in nature, it might face. This proposal is supported by a non-exhaustive review of relevant current research in deep learning, NLP, machine vision, and computational creativity. This is done from the perspective of a practicing artist, as opposed to a computer scientist. Moreover, this article tries to broaden the discussion of visual understanding in AI from a overly-historical, painting-focused approach to one that takes more account of the actual condition of contemporary art. The objective of this broadening is to illuminate the road to AGI, in a different and hopefully, additive, way to the mainstream conversation within the computer science community.