The production of machine vision’s aesthetic score 

How to approach the relations between aesthetics and infrastructure? The presentation will approach this question through the case of the AI image generation software Stable Diffusion. To function, image generators need large datasets or collections of images paired with descriptions and several other metrics such as an aesthetic score. This presentation will analyse the various actors, humans and machines, who intervene in the curation of these images and the production of this aesthetic score for which images are rated on a scale from one to ten, from the less to the most aesthetically pleasing. It will do so by unravelling a vast network of image production, exchange, description and exhibition that includes photosharing websites, annotation platforms, chat rooms, and university departments. Through the lens of a background aesthetics, it will show how this network functions as an exhibitionnary complex enabling feedback loops and modes of reflexivity in the design and deployment of image generators as well as the vagaries inherent to teaching the judgement of taste to machines. 

BIO
Nicolas Malevé is a visual artist and computer programmer and a post-doctoral researcher in the framework of the Knowledge Servers project at SHAPE and previouslyin the Artistic Practice under Contemporary Conditions project, Aarhus University. After a PhD on Algorithms of vision, he investigates the socio-technical networks of machine learning and their artistic and epistemic implications.

Æstetisk Seminar er tilrettelagt af AIIM – Centre for Aesthetics of AI Images, Lotte Philipsen og Mette-Marie Zacher Sørensen, Institut for Kommunikation og Kultur, Aarhus Universitet

• Poster
• Program