Where machine perception breaks.

Computers are learning to watch us. They follow a hand, read a face, squash a person into a few numbers and draw them again at the other end of a video call. Nearly all of it is built for someone sitting still, in good light, doing something ordinary.

Performance is the opposite of that. On a stage an eyebrow can change what a sentence means, the lighting is actively working against the camera, and one wrong read ruins the moment. So I use performance to push these systems until they break, and then look at who gets hurt when they do.