The increasing encroaching of ML into the creative fields has been spearheaded by generative models that couple an artificial perceptual system to a generative parametric model. This coupling that can happen at the training or inference phase is often presented as a black box that translates between two forms of signals. While the productive capacities of the generative part of such systems are often overrepresented in discourse and practical applications, there is substantial creative and theoretical interest in the “perceptual” side which arguably determines the intrinsic aesthetics of such technologies.
In this course, we will explore a few simplified ML models, dissect them and interrogate them to try and understand what kind of world structure their internal representations suggest and how we can use it in creative workflows. More specifically, we will focus on how the probabilistic nature of these models undermines ideas of fixed and clear categorical boundaries and identities imposed by symbolic language.
The course is intended for students with the aim of developing a critical approach to the creative applications of ML. Students will work on a group project that will take them through training, interrogating, and applying a model to a simple design problem.