Data and Computer Vision

The discourse around artificial intelligence and computer vision often leads to misunderstandings due to the anthropomorphization of machine processes, which obscures the actual workings behind computer vision outputs. This includes the significance of datasets composed of digital surrogates of artworks that reflect human circumstances and material conditions. There exists a bias in available artwork images online, largely favoring institutions with resources for digitization, which is an ongoing problem in the digital humanities. By focusing on the underlying datasets and their limitations, art historians can demystify computer vision, recognizing it as a constructed product that requires critical engagement. Acknowledging the ethico-political issues related to datasets is vital, as is maintaining rigorous attention to source materials to support a deeper understanding and critique of images produced by computer vision programs.

This is an Accepted Manuscript of an article published by Taylor & Francis in The Art Bulletin on 2024-04-02, available online: https://www.tandfonline.com/10.1080/00043079.2024.2296273.