The work XSICHT was created in 2019 and is thus an AI of the penultimate generation that was trained over several weeks to match people’s voices to their faces. The basis for this was the interview series Ask a Cliché from the YouTube channel Hyperbole, in which people with different backgrounds are asked questions by the audience.
The videos are perfect for training an AI. However, in the learning process, two AIs are actually trained at the same time. One gets the audio signal and generates an image, the other takes the generated image and tries to decide if it is a generated or original image. This leads to a cat and mouse game, with the AI generating the image trying to create images that are closer and closer to the original, while the other AI gets better and better at distinguishing the generated images from the original – which in turn leads to the generating AI having to try harder, and so on.
It gets interesting when the purpose of use is alienated and abstracted. What happens if we feed the AI music instead of voices? Harmonic piano music, for example, more often leads to the recreation of stereotypically female faces, while bass-driven techno usually resembles speakers with deep voices.
For the exhibition „Painting with Machines,“ we want to involve visitors more in the creative process and have given the AI ears in the form of a microphone. We explicitly encourage humming with different frequencies or holding different vowels and experimenting with the AI. For example, a humming „mmmmmmm“ is clearly different from a buzzing „iiiiiiiii“.