> I'm given the option to track if that person shows up again in future video footage, you bet I would use that option.
Most people would use that option if they had the opportunity; tracking potential troublemakers has obvious benefits. The problem is you are not being given the option "to track [a] person". You're being given the option to record video and apply an automated tool to identify "likely faces" and as way to compare those images against a local database using a definition of similarity that is poorly specified[1], extremely complex, and opaque/unexplained. Jumping right thinking of this as "tracking a person" ignores the many layers of metaphor shear between the desired goal and what the actual implementation does.
Metaphor shear is what you experience when you realize that interface doesn't match what is actually happening. The metaphor was lost somewhere in the hundreds of layers of abstraction connecting input to output. The map is not the territory. Efficient and powerful tools are very useful, but they also let you efficiently shoot yourself in the foot in powerful new ways.
Since metaphor shear is about the error between an expected ideal and actual measured reality, it compounds across layers of abstraction similar to how experimental error loses precision with each calculation. Any particular layer of abstraction is probably fine in isolation, but the interface metaphor loses significant digits as more and more layers are added.
If intellectual labor is going to be surrendered to a device, we need to be very clear and specific in what we are surrendering and what the device is expected to do.
(Video essayist Kyle Kallgren does a much better job of explaining[2] the problem of metaphor shear, using examples from sci-fi [Black Mirror, HHGttG, R.U.R.])
[1] poor in the context of trying to track someone. I'm sure the face matching algorithm is very well specified with respect to matching images of faces. The poorly specified part is when "matching images of faces" becomes "tracking someone" (what does "tracking" mean? How are probabilities about matching faces mapped to identifying "that person"?)
[2] https://www.youtube.com/watch?v=hr9_DcO6G3A