Parts Unknown

Parts Unknown is an exploration of the way machines can ‘see’ our world through machine learning, using the view of city from above as a metaphor for these networks required for a machine to learn. The city is flattened when looked down upon, forcing the roads, buildings, and other features into two-dimensional shapes. When satellite imagery of the city is introduced to a Generative Adversarial Network (GAN), it re-creates the patterns that are common throughout the thousands of images that it is fed, reducing it further to abstraction.

Whilst the GAN is capable of generating a diverse set of completely new images, the learning process creates an averaging of the city, where any unique features of cities are lost. It highlights the importance of what is in the data that learns from. These images, which are small sections of a machine-made urban environment, are then digitally collaged to create a larger map of an imagined city, one that comes from a collaborative process between the artist and the machine.