Collaborative Artistic Productions with GANs


Abstract / About the project


The social and cultural significance of machine learning is often defined by polarised perspectives. Machine-learning might be seen as opening exciting new pathways for creativity, or as a fundamental threat to the role of the artist. The interdisciplinary focus of the Creative Media Practice/Research Cluster is exceptionally well-placed to engage with such questions. This project will use a practice-led approach to examine how machine learning might alter artistic approaches to form and composition. The development of Generative Adversarial Networks (GANS) as a mechanism for classification and production presents a new opportunity for artists and humanities scholars to re-examine fundamental questions of creativity. Recent publications in mesh-generating and cycle GANs suggest that such systems might have profound impacts for sculpture and design practice. These impacts might challenge basic principles of form, function and aesthetics, and the creative use of such technologies might result in new and novel forms of data encryption.

Given the rapid acceleration in the capacity and variety of these systems, it is critical that humanities scholars engage from both a practical and a theoretical level. Leveraging the unique structure of the Creative Media Practice/Research cluster, we propose a four-stage research project with a rolling PI system, where each stage is directed by one member relative to their specialisation, and each stage builds upon knowledge gained in the previous stage. Our research question of how GANs affect collaborative practice and artistic form is divided into four sub-questions relative to each stage. First, we examine how mesh GAN and cycle GAN systems can function as artistic tools, and how reversible encodings can be used to create new semiotic systems. Second, we examine how this mesh GAN system can be used to generate human hand tools outside of the evolution of form and function. Third, we repurpose this system for music and performance to see how performers and audiences react to the blurring of barriers between acoustic input and generated output. Fourth, we use a combination of cultural studies and ethnographic methods to see how our practice-led research can shed light on the impact of machine learning on creative practice. Our team will combine existing research from computer science to develop new creative tools, produce new datasets with both research and cultural heritage value, commission a new series of artworks, exhibitions and performances, and publish our findings in academic journals.