I haven't read the transcript yet but are there no "human in the middle" loops here?

Surely we could just get a pair of eyeballs on everything flagged and if it's not 100% crystal clear what's going on, then it gets flagged for a senior pair of eyes to look at?

I'm against the Minority Report kind of stuff but a system (AI Flock cameras in this case) flags up <thing>, surely it should then go into a queue for humans to look at! Flock can add in a certainty score and the human must verify!

It could still increase efficiency this way.