Re: False positives
"Even detection methods we actually consider working have high false positive rates when the thing they detect is rare."
While I agree that simply having a high false positive rate is not an indicator of a bad test, from what statistics I've seen from facial recognition from the Met it indicates that the system is generally not making ANY true positives. From 80 hours of surveillance they identified two persons of interest, one of whom was not a suspect, the other had already been dealt with, but he hadn't been taken off the naughty list.
So a grand total of ZERO actual criminals caught or crimes prevented.
Compared to other surveillance techniques, it's awful. Even CCTV is better, and that's saying a lot. ANPR from an elevated camera (spotter plane) has about 40% false positives, mainly as it only gets partial data. But since you've got other factors (vehicle colour etc) you can detect them pretty quickly.
Assume that it took at a minimum two sworn officers to man this. That's 80 hours that a pair of cops could have been on the beat, being visible etc. Which turns out actually reduces crime....
Biting the hand that feeds IT
