Still from exhibited work
© Andreas Refsgaard
Aagaards glasplader hacked, 2019
While the digital impacts our mnemonic capacities, historical archives are also digitized affecting how we collectively remember the past. Together with a StyleGAN model, Andreas Refsgaard re-writes our history in the video work Aagaards Glasplader. A generative adversarial network (GAN) is a machine-learning system that can produce an image from a preexisting visual database. From this database, it intercepts patterns and information, after which it can demonstrate its own reverie of the world. Refsgaard has run portrait photos from the mid-19th century through the model, generating new portraits of a non-existing past accompanied by likewise generated historical biographies. It is just plain math, and yet this maneuver can be described as a symbiotic creative process between machine and human. Refsgaard’s work imitates how machine-vision, such as intelligent surveillance cameras and image recognition on social media, detects and analyzes objects, faces, and actions in our everyday surroundings.
cand.mag in art history
Aagaards Glasplader is a collection of high-resolution portrait photos from 1857-1880.
I trained a StyleGAN model to generate new portraits of non-existing people. I used short historical biographies of the people from the collection to have a GPT-2 model produce new texts about the generated people, letting it dream about how their lives might have been.