[nocturnial]

> If the technology is so spot on that 99% of the time you commit the crime you'll get caught

And in 1% of the time[1], you are an innocent who had nothing to do with the crime. Now you have to spend your time: getting arrested, spending time with lawyers, getting people to collaborate your alibi (<- In the best case! Otherwise you're f'ed).

If you want to make a case for facial recognition, criminal justice is one of worst possible cases you could make for it.

[1] I don't accept your claim of 99% accuracy when this is applied on a massive scale. Maybe I'm wrong, so let's go with this number

[sentinel]

The 1% there is that “the criminal won’t get caught”. Not that someone else, an innocent bystander, will get caught.

Also what you described can happen to you today as well. It’s the imperfect system we live in.

[smolder]

False positives always exist in these systems along with the false negatives. There's no avoiding every case of false conviction if you rely on imprecise methods. That's admittedly already a problem without facial recog, but is a problem when using it, too.

[nocturnial]

> The 1% there is that “the criminal won’t get caught”.

So the 99% is for the facial recognition system to identify criminals?

What error rate does the best facial identatification have? If it's not zero then my original point still stands. You are throwing a lot of innocent people into the judicial system. Or at best being pestered by police for no good reason.

> Also what you described can happen to you today as well. It’s the imperfect system we live in.

That's correct. So instead of improving it, you suggest to automate it? This sounds insane to me. Why would you automate something that you know is defective? Check the user support of all the top tech companies who use "AI" to automate things. Now check the worst of the worst customer support of the top tech companies. There's an overlap. And if you want to expand (*censored*) tech support to the criminal justice system, people like me are going to get upset.

[sentinel]

> So instead of improving it, you suggest to automate it?

I said no such thing. Read the whole thing again, take the upset level down a few notches and get back to us.

[nocturnial]

I'm sorry if I misread your message.

This is were I'm at:

You brought up the fact that FR could identify criminals. I brought up the fact it could also identify innocent people as criminals. When innocent people are accused of crimes they didn't do, they usually get upset.

I didn't say I personally was upset with your message. I said they would get upset. I'm not a "people" person, but this should be obvious. And I think I would be upset too if I got accused of something I didn't do.

You still didn't tell me how you would resolve an innocent person being misidentified by a facial recognition system. Can they sue the company or person who developed it?

[sentinel]

> how you would resolve an innocent person being misidentified by a facial recognition system

Due process, which they'll still be entitled to.

Not sure about suing. I'm sure people will try and sue.

[noodlesUK]

A claim of 99% sensitivity sounds good, and is often achievable. Any real system will also have false positives, so let’s say that we have a specificity (test true negative rate) of 99% as well. This is probably unrealistically good, most systems will false positive more. This sounds great.

However, Bayes’ theorem paints a very different picture.

If the prevalence of wanted criminals in the population is say 1/10000 (this is hard to guess), what are the odds that a person that is flagged is a wanted criminal?

The unintuitive answer is less than 1% of the time (~0.98%) will the suspect actually be a criminal.

By far the most important term is the prevalence of the thing you are testing for in the population, in this case criminality. Any dragnet facial recognition is invariably going to get more innocent people caught in its web than true criminals.