Letters from the In-Between – Series I & II
The work titled "Experiments in the GAN-Between", comprised of two series, is an exploration into the possibilities of creating graffiti pieces which are entirely generated by Artificial Intelligence (AI), i.e. using machine learning algorithms. In order for the AI to produce new graffiti imagery, it was trained on old images of NYC graffiti pieces from the 1970s/1980s.
Both series consist of three images which document the production process (left, center and right). In each series, the left image shows in a square grid the complete data set that was used to train the machine learning model, i.e. ~ 3000-4000 images of train graffiti pieces by pioneers Ajax (series I) and Pnut and Tracy (series II).
The center image shows the digital output of the machine learning model (StyleGAN) generated with a computer, which was then used as an inspiration and practical sketch for painting a new abstract graffiti piece with acrylic (spray) paint on a wall in the Californian desert in March 2020 (right image). This process of generating new graffiti pieces based on old source material in collaboration with a machine is what I call "GANfitti".
This work raises questions about the way in which machines "learn", based on highly subjectively selected data samples and investigates the role of such automated decision making processes in the creative practice.
How do machines produce images and predict a certain outcome? How can AI be used in applications without targeted monetary expectations? And what are the implications for niche subcultures such as graffiti?
At the end of Sept 2020, I gave a talk about my work process at Delphi Space in Freiburg. Below is a video recording of the talk (in German only unfortunately ;)