For the press
In ‘Dreams of the Machine’ the artists turn to the concept of an algorithmic unconscious that implies the possible existence of the unconscious in artificial intelligence. A specially trained neural network reproduces dreams using texts and visual images, but there is no way to determine their origin — were they produced by a human being, or a machine?
The work uses the GPT-2 text generation model. This algorithm is able to predict the next word by focusing on the previous one, allowing to create pages of coherent text based on a single phrase. The neural network analysed a database of 34 thousand textual dream reports at the Sleep Neuroimaging Laboratory of Geneva University, then wrote new ‘dreams’ almost indistinguishable from the human ones. The results of this work became the input parameters for another algorithm — a customised combination of VQGAN and CLIP models, which created images corresponding to the presented texts.
Strategic partner
