Hmm, "cameras reported a 97.3% match". I would assume that for a random person, the match level would be random. 1÷(1 −.973) ~ 37. IE, 1 in 37 people would be tagged by the cameras. If you're talk China, that means matching millions of people in millions of malls.
Possibly the actual match level was higher. But still, the way facial recognition seems to work even now is that it provides a consistent "hash value" for a face but with a limited number of digits/information (). This be useful if you know other things about the person (IE, if you know someone is a passenger on plane X, you can very likely guess which one) but still wouldn't scale unless you want a lot of false positives and are after specific people.
Authorities seem to like to say DNA and facial recognition caught people since it implies an omniscience to these authorities (I note above someone throwing out the either wrong or meaningless "97.3% value). Certainly, these technologies do catch people but they still limited and expensive.
> I would assume that for a random person, the match level would be random. 1÷(1 −.973) ~ 37.
Why would you assume that?
The only way a percentage match means anything here, is that the facial recognition software returns a probability distribution of representing the likelihood that the person identified is each member of the set. I'm sure that 97.3% is actually low for most matches, since she had extensive plastic surgery.
Another related thing to consider, if she had plastic surgery what are the odds that among a billion people there isn’t someone whose face looks more like her original face than her face looks like her original face.
The "97.3%" match is probably just the confidence value - I don't think a frequentist interpretation makes sense for this. I'm not an expert in face recognition, but these systems are very accurate, typically like >99.5% accuracy with most of the errors coming from recall rather than precision. They're also not _that_ expensive. Real-time detection on embedded devices has been possible for around a decade and costs for high quality detection have come down a lot in recent years.
Still, you're right that at those scales these systems will invariably slip once in a while and it's scary to think that this might enough to be considered a criminal, especially because people often treat these systems as infallible.