Machine Learning and Notions of the Image

This thesis addresses how current notions of image production remain tied to historical ideas which often prove inadequate for the description of visual artefacts of machine learning (ML). ML refers to the simulation of in-formation acquisition using machines, and when applied to the generation of images, it enables visual content to be influenced based on the statistical analysis of data. The increasing use of ML in image production highlights several aspects which have been present in older forms of media, but which now take on new forms and relevance, especially within artistic contexts. This research seeks to clarify the mediating role played by visual technologies and to demonstrate how images produced using ML offer new ways of approaching theories of the image.

Images exist at the interstices between human perceptual experience and its technological mediation, which is especially relevant as the development and implementation of technologies offers new possibilities to produce visualisations from data. In so doing, technological mediation tangibly augments relationships between how images are produced, experienced and interpreted. The present incorporation of ML into various forms of visual media offers insight into this issue by enabling images to be produced as the result of the statistical analysis of datasets. Computational relations which are extracted and inferred between features within images help to construct learned representations which are in turn used to generate new images. This results in a form of computationally-determined representation which is informed by the interpretive processes performed by machines.

Artists have taken great interest in the potential of ML, in an aesthetic, but also a processual capacity, often considering its relation to human vision. Their productions offer insight into novel aspects of ML in the creation of images through experimental practice which is informed by theory and by art history. Using and reflecting on ML, often in novel or reactionary ways, artistic and humanistic perspectives provide vital counter-narratives to those of computer science (CS), and which facilitate cross-disciplinary understanding.

In spite of the hype which surrounds it currently, ML does not present an entirely novel approach to image production and rather builds upon existing modalities and narratives surrounding the technical production of images. Notions of technically produced images often lean heavily on historical narratives regarding the technical production of images, even perpetuating inaccuracies from them. These tend to misconstrue images either as inherently accurate reflections of reality or as the product of artificial perception and genius, by virtue of their engagement with technological processes. This research therefore adopts a media archaeological approach, in order to understand how processes that have been present in visual media much longer than the use of ML continue to colour discourse.