Deep packet introspection is a completely different problem from the kind of facial recognition of people in video backgrounds that you envisage. Videos uploaded from cell phones are already MPEG-4 compressed and don't need re-encoding to be served. While it is true that social media companies don't allow unfettered access to their data, server farms for scraping web sites are a dime a dozen. I don't agree that it is safe to assume that systems there is no evidence for exists. It's one step removed from Chinese mind control viruses.