pt9 - AF:CFFiIV
2019, Video art
Used media:
Video: grpahics was generated in StyleGAN neural network (duration 03'27").
Categorization: Betaface API.
Voice: Google WaveNet.
Sound on the background: 1 second of sound squeezed out of retrained Mozilla TTS neural network, processed in Ableton.
Project «AF:CFFiIV» explores imagination phenomena. It is based on an activity of three non-human agents - neural networks. The first one generates images of a human face in the ambience. The second neural network creates a description of these images according to category recognition algorithms. The third one reproduces these descriptions with a trained voice.

Neural networks (NN) is an acceptable model for cognitive process exploring inside a human brain. An imitation of neural assemblies, processes and connections allows comparing the NN with the base assembly of human consciousness. NNs help to illustrate the process of world perception and it's converting to an experience.

Imagination phenomena could be examined as the process of spontaneous occurring or premeditated images or ideas creation which cannot be perceived by senses. Consciousness reassembles the picture of the surrounding reality based on the images obtained from the outside and it manipulates them.

NN synthesizes new images having the humanlike experience based on the dataset. It models reproductive and productive imagination which is neither constrained with moral or aesthetic principles nor emotional bounds. According to David Hume, unimaginable does not exist to us. So, is it possible to say that neural networks can expand our own understanding of beauty, aesthetics, and boundaries?

Neural networks let us follow the process of constructing images that arise in the intermediate stages of the image generation where it is possible to investigate the ability to create a variety of values that allow generating images of realistic-looking people.

In this project, in turn, the focus was shifted to the limits of the normal distribution of numerical vectors, which are translated by the neural network into an image, examine the reverse process: generation of images from realistic portraits to a digital noise, in which the neural network "dreams" to find a person through its similar experiences images.