Police Feeding Celebrity Photos into Facial Recognition Software to Solve Crimes | Hacker News Reader

Police Feeding Celebrity Photos into Facial Recognition Software to Solve Crimes

vice.com

68 pointssaravana8528 comments

Feeding in doctored images, grainy social media pics of people that look "kinda similar" and other garbage data into these tools just to generate a list of possible suspects is a surefire way to generate garbage matches (i.e. non-matches).

I don't see how this is any better than just putting all the names of people who have the same demographic info into a hat and picking a set of them.

I guess incompetence might be the thing that saves us from a truly effective Orwellian state.

It won't save us from a truly pervasive mafia state though.

I write FR software, and this practice irritates the fuck out of me. A common phrase in the industry is you can't anticipate what stupid users won't do, and this is a clear case of end-users sticking their fingers into powered light sockets, at the expense of the general public's safety by the very agency enlisted to protect the public. The stupidity is amazing. If there is ever any regulations around FR's use, probing with doctored or "looks like" images needs to be illegal or at minimum no grounds for any warrant of any kind.

If the FR were to learn based on input I could see how feeding it doctored images would screw it up in the long run. But if its just the FR utility, minus any ongoing machine learning, how does this effect the public's safety exactly?

Not disagreeing just curious.

I had thought they were going to say that they use celebrity images to train the systems. Because celebrities are an incredible training set for facial recognition.

Consider: You want to train a one-shot matching system for taking photos from crime scene cameras and searching a database of drivers licence photos. You need a training set of different images of the same person, and then lots of those sets for different people. Boom: celebrities! They have thousands of photos from thousands of angles. Poorly taken shots by paparazzi. Stills from films at a zillion different angles. And some nice, face-on shots like a drivers licence photo. (Heck, if it's the police they have access to that celebrity's actual DL photo).

The one nice side about training that way is that most celebrities are really good looking people, so only the really pretty criminals will be caught thanks to training bias.

It’s also bad news for people that have stunt doubles or thought cosmetic surgery would keep them free.

dataset will be quite biased - few fat people, few people of color, few very young and very old.

While I've never been detained, I've been around police enough to know that they generally know the likely suspects immediately after many crimes (turning in a statement after an SF Mission smash-grab; talking with a friend about a -significant- tool theft from their barn). I suspect the reason this celebrity-photo thing is happening is that the police already know their primary suspect and they're looking for "modern", crappy supporting evidence. It's like the JS/PHP of police work...

I cannot fathom how they put obvious mistakes in the training material (e.g. hairline in the last example). How are they selling something when their own training material shows it's total bunk.

The time and money spent on that could be spent on actual police work...

Because it's not about accuracy. It's about generating a match, no matter how crappy. It's no different than back when they used to pick up random known criminals (usually non-white) and charge them with whatever crimes they had no leads for in order to clean up the number of open cases. Selecting the random person to pick up is now outsourced to the software. The cops can claim not to be responsible because they're just doing what the computer says. The company selling the crap can claim not to be responsible because the cops are using it wrong.

> It's about generating a match, no matter how crappy.

That's the worst thing about this kind of dragnet search is that the more comprehensive the surveillance and the more accurate the picture of the offender, the more likely it is that when you're picked up as the prime suspect in something you had nothing to do with, that you look identical to the culprit and you were in the area at that exact time and so on.

And so if you are the unlucky bugger that gets caught up in it, the better the system works, the harder it is for you to argue that you actually WERE innocent. Juries aren't good at the Law of Large Numbers.

Well we can look at this from the opposite side, it eliminates more people far faster giving the authorities fewer people to check. Now this is fine provided there are sufficient rules governing that second step.

That is where the process currently fails. There are no published checks and balances.

Look, we don't have to like this one bit but we are kidding ourselves to think it can be stopped. Even the actions of San Francisco can be overridden at the state or federal level when push comes to shove if not a shift in electorate. so the task ahead is to regulate strictly how it is used and what can happen from beginning to end of life of the data. the first rule being, it can never establish guilt on its own.

There will be some percentage of false positives. If you're feeding the system with a biased, handpicked population, you will get false positive matches from that population and not from other populations. If you are on the shit list, your odds of being accused when you're actually innocent are far, far higher than those of someone not on the shit list. If the shit list is predominately black, then innocent blacks will be sent to jail more often than innocent whites. It's profiling.

Fantastic, because I look like the literal twin of a very famous actor who appears in multiple current blockbusters. It's not that either of us is especially good looking, but we have that sort of face that is bland enough to take on many different looks.

It will be interesting to see how facial recognition is treated as evidence. Things like, will judges grant warrants based on a 80% match but not a 70% match? Will juries consider it more or less reliable than recognition by eye witnesses? Will defense experts be able to cast doubt on certain algorithms or techniques vs others?

I disagree that this is, per se, a violation of due process

As long as the matches are not being used as evidence on trials or as evidence to get warrants, the police can generate their list of leads by consulting a hamster, or by taking names out of a hat. Right? Or is there/should there be a right to not be considered a suspect on foolish motives?

Also, maybe these techniques are useful, in the sense that they allow for police to use a bad image to reduce a pool of suspects to a manageable pile that can be examined by hand.

Yudkowsky wrote about this.

https://www.lesswrong.com/posts/X2AD2LgtKgkRNPj2a/privilegin...

In short, it does seem like it should be illegal to pull names out of a hat.

All I am saying is there are 2 different questions:

* Is the method good? (FR with celebrity, maybe not. A hat, surely not)

* Is the method legal? (IANAL, but the relevant question seems to be if they are using it as evidence or just for lead generation)

So you would be comfortable with police essentially randomly selecting people to arrest for various unsolved crimes? And we'll just let the courts sort it out?

The reason the fourth amendment exists is exactly to prevent this sort of thing.

Considering a large percentage of inmates have yet to be charged with a crime this would probably be incredibly lucrative for the prison industry.

No, but it is my understanding that to even detain a person you need to meet some standard of evidence. I am saying: let the police use whatever lead generating method they can, as long as they are not using it as evidence to detain, search, or imprison.

That is, their system can be whatever they want to generate a list of names. Afterwards, they need evidence to search a persons house, or to detain them, or to arrest them.

(In legal lingo: I am okay with them using these methods, as long as they are not considered probable cause https://www.nolo.com/legal-encyclopedia/when-police-can-make...)

For looking for leads in an investigation it doesn't seem that different from having a person look through mugshot books. I agree it gets very sticky when figuring out where it appropriate to use it as evidence for a warrant or in a trial.

This reminds me of the common practice of using drug-sniffing dogs not to actually sniff out drugs but as an excuse to do a search.

I think there should be an examination and maybe some additional rules around establishing probable cause using what amounts to a divining rod.

Or the "you didn't use a turn signal" as an excuse for a traffic stop.

Considering that all the research points to image recognition having basically no robustness to noise, I imagine the path here is going to be to at least block it's use in court, if not to generate the lead to begin with.

Surely, having such a excellent track record- criminalistic techniques and sciences, will soon reach the level of a true science such psychology.

Can’t get past that awful ad play with no close button taking up half the screen

So this is what actual True Detectives do?

Reply on news.ycombinator.com