Comment by eesmith

I took your list ("The issues are internet connectivity, data retention/mining/sale, and non-local processing") as being incomplete. The examples I gave were to give examples of additional issues. There are equivalents for my examples to private businesses, even putting recognition systems to the side.

I personally have noticed that "alert" and "suspicious" tends to mean "something unusual", and not "something illegal". Increasing alerts results in forced normality.

On the flip side, if the information was there and not used, then the security guards are blamed for not connecting the dots, so investigating alerts becomes a CYA task.

As an example, security guards have harassed people on public sidewalks who are legally taking pictures of the building they are guarding. They are incentivized to investigate the alert, face no consequences (so long as that harassment doesn't itself break the law) for a false alert, and risk losing their job if the photographs are used for nefarious purposes. Adding air-gapped AI may help the security guards, while increasing the amount of harassment.

Yes, I have had a security guard stand over me while I delete a photograph I took of a building while in a public park. I think I was not legally required to follow request. I wasn't going to risk escalating the confrontation over a picture of a neat-looking gargoyle. No, I don't want AI enabling more of that harassment.