Yes, that's key that journalists don't cover adequately. The system has to be trained and told who is in the room and where (and presumably have them move around so it can learn what it looks like), THEN it can identify quickly. It would be like a security system using "AI" to show you a picture if your front door based on just the open/close sensor, after you give it photos of it being open and closed. > Devices on a wireless network regularly send feedback data known as beamforming feedback information (BFI) to the router. Because this information is transmitted without encryption, anyone within range can potentially read it. Researchers say these signal reflections can effectively create multiple "views" of a person, allowing AI systems to learn and recognize individual identities. > > **After the machine learning model has been trained**, identifying a person reportedly takes only a few seconds.