Breaking The Creepy AI in Police Cameras

Funny you should mention this. Benn Jordan, on the heels of his Flock camera research, reverse engineered a non-flock ALRP unit and built his own system using some off the shelf parts, a tablet, and 3D printer. He did it mostly as an experiment but also due to the fact that Flock's algorithm for image detection is astoundingly bad and has a high incident of reading plates incorrectly. There are a handful of open-source models for license plate detection, I forget exactly which model outperformed the rest, but it was an excellent watch and help me really understand just how inefficient these commercial systems are and how easy they can be to defeat: [this video]

Ok Before I am accused of misrepresenting things and in the interest of clarity. 1. I am in San Francisco. 2. Upon further research Flock does not appear to be used by SFPD for ticketing. They have another camera system for that. That does not mean that Flock could not be used for speeding tickets or a host of other things like running red lights, littering etc. Microprocessors and CMOS cameras can do amazing things. Regardless of what company is doing the surveillance the debate remains the same. 1. Pervasive surveillance has the potential to be used inappropriately by authorities. Discussion is needed. 2. The data collected by a company such as Flock but not necessarily Flock could be used inappropriately or sold for other purposes. Benn Jordan has a good video on the subject [this video] 3. The claim that Flock is responsible for %10 of solved cases seems dubious but without data it is difficult to know or believe. Given that Flock is in YC's portfolio makes it even more suspect. Flock or a company like Flock may try and become pervasive and indispensable before a public discussion places any limits on their behavior.