Using AI recognition systems on body cameras could be a powerful system for teaching thugs to be less brutal. However, it could also be a powerful system for repression of everyone that comes into range of the camera.

To make body cameras serve and protect the people, rather than repress the people, we need to make sure that they capture video and audio when they should, and not when they shouldn't. I've made a technical proposal for part of that, but we also need proper laws about publishing the videos.

As for the idea that live monitoring of the video could enable a deescalation specialist to intervene and prevent a killing, that might have saved George Floyd. But many killings by thugs occur when thugs react without taking even a second to think. There would be no time to intervene, either. Meanwhile, the live transmission would imply transmitting the video all the time, which is wrong because most of the time the video should not be saved at all.