3 ms·
The problem (luckily!) is, that the thing they are trying to detect happens extremely rare. And when you try to detect rare events, even if your methods are 99.
by atoav 6y ago
The problem (luckily!) is, that the thing they are trying to detect happens extremely rare. And when you try to detect rare events, even if your methods are 99.9% accurate you will drown in false positives.
I once did the maths on facial recognition tech which was meant to recognize terrorists at Berlin main station. With the known number of commuters there and an accuracy of 99.9% (which is extremely optimistic) they would have over 600 false positives per terrorist if I remember correctly. The system they had installed there however wont even reach 80%...
Imagine you have an alarm that rings twice every day, and is real and serious only once a year. How likely are you to react in time?
I think this is the fundamental problem with the "we need more data and more rights"-strategy of most police and law enforcement.
I am convinced what they actually need is more personel with a good education.
Edit: what this kind of data is better at recognizing are broad movements in society like demonstrations, protests and such things.
- AndrewBissell 6y ago> Edit: what this kind of data is better at recognizing are broad movements in society like demonstrations, protests and such things. Yes, this is their real purpose. They're not rolling out these wide ranging systems at extraordinary expense when they're at all unsure they will actually solve the problem they're trying to solve.