Going for Broke: the Faces of Networked Intelligence.

Hakemuksen tiivistelmä

As humans we tolerate bad designs by going around them, avoiding streets with confusing parking signs, deleting autocorrected words. We treat our machines the same, assuming that if they behave badly, it’s by design, so we find them cute and silly. How bad do machines have to perform until we believe that they are broken? I’ll install 2 large machines that are flaky and shy, using the audience’s face and brain wave data to guide their actions, until they finally suspect that its reactions are flawed. One is a lighting system that follows faces in the crowd, lighting up areas in the room where faces are by moving between them. When faces come close to the machine, it moves away. Over time lighting of faces become random, lighting random areas of the room, forming suspect audiences. Finally it looks to the back of the room even when no audience is close, as if broken. The 2nd machine is on-screen-face display that measures attention level using NeuroSkyEEG headset. It imitates viewer’s actions by face-tracking using computer-vision. When attention levels are low (low EEG-beta-bandwidths), the face moves closer in style to previously GAN-machine-learned bored face. When attention’s high, the face moves closer to the learned alert face. Over time the mode moves away from EEG attention measures and morphs into other shapes spontaneously as in Deep Dream. Finally it crystallizes into a laughing face that stares perpetually at the viewer, looking like it’s broken.