Visual artefacts through the Black Box: Analysing Deep Learning classifcation of Neoclassical furniture images

Te recent trend towards introducing digital instruments in the history of art is glaring at best. Like so many disruptive technologies new techniques in computer vision and natural language processing are about to fundamentally change the way we conduct and even envision our very disciplines. Image classifcation experiments for instance have become a trending topic in visual culture research and more traditional art history alike. Prominent among them are Artifcial Intelligence Deep Learning techniques such as Convolutional Neural Networks, (Krizhevsky, Sutskever, and Hinton 2012). New instruments however harbor new methodological and heuristic challenges since these techniques are a black box through which images are passed to create what is perceived as hermeneutic knowledge (Castelvecchi 2016).