Eight months pregnant and arrested after false facial recognition match
nytimes.com
nytimes.com
Facial recognition is not factual. It is explicitly inferential.
In theory, every voter who claims to be pro-women's rights, or pro-civil rights, or liberal, or ... should be eager to make this fantasy come true.
In practice... Hahahaha!!! Yeah, right.
So - why the huge gulf? Could it be that the great majority of voters claiming such moral cred are just Performative Activists, who don't actually give a crap?
If all the politicians, wanna-be politicians, and their armies of consultants and pollsters, seem to believe that where it counts, in the polling booth, voters don't actually care about a certain issue...might all those amply-motivated experts possibly be correct?
If the position is appointed then you have yet another layer of indirection and competing factors for the voter to evaluate.
Even for presidential elections it's hard to feel great about your vote... Democracy kinda sucks.
True. But in a world where the elderly Secretary of the Methodist Ladies Gardening Society has no problem keeping track of donated funds and flower bulb locations and volunteer weeding in her 3-ring binder... it would be incredibly easy for a few people who actually cared to keep track of DA behavior, year after year, as a side hobby.
Bottom-end web hosting is ~free, to make that information available to the public.
Yet ~nobody seems to be doing that. Are the police and DA's quietly shutting them down? Perhaps...but 10000X the quantity of hot-headed political attack material is freely available on the web. And it's hard to assert that those police and DA's are every-so-quietly and -competently doing that by night, while regularly bungling similar work by day.
Or...is the problem that the modern version of "caring" about such such police and Prosecutor incompetence and misconduct is 100% performative activism, and 0% interest in taking any action which might lead to "positive" change? It's not like that behavior isn't seen on other issues. Lots of well-to-do liberal neighborhoods will poll as being very strongly in favor of (say) affordable housing for less-well-off people. But if there's any hint that such housing might appear near their precious neighborhood ...yeah. Don't expect them to behave the way that poll suggested they would.
If you made a list of things that "the other side" truly cares about, it's probably mostly what you care about.
When you strip away what people are currently angry about, people mostly care about things that come down to "am I making the world better for my (possibly putative) children."
It's a reductive view, but I think it's true in a useful way. That is to say, I think it's true for most people in a pairwise fashion, so it's probably true of you, and it's probably a useful way to think about other people you meet, most of the time.
In the context of social media, it is profitable for people to be performative because that drives engagement and engagement is profitable. For this reason, social media is a terrible proxy for other people because is not one to one (pairwise) as above, but one to many. Thus, social media is a massive distortion if you spend time a lot of time on it, or if you spend a lot of time reading the thoughts of people who spend a lot of time on it[1]. I thought of giving exaggerated examples of "the other side" here but I don't even have to draw the pictures, the caricatures literally draw themselves if you think for just a moment. For both sides. Which is kind of what I'm driving at.
Most people are essentially like you though, they just want things to be better. They aren't performance artists gunning for virtue signaling engagement, they just disagree with you about the implementation details. Sometimes these disagreements truly are utter incompatibilities, it's true. But there's an entire industry that now exists in convincing people with the same broad desires that the other side desires only to dash their hopes and dreams. My boring thought is that, broadly, there's a breathtaking overlap between what you want and what the (call them) deplorables on the other side want, and you're mostly just vigorously disagreeing on the cause célèbre you've been told is the keystone to everything this season.
[1]: https://www.niemanlab.org/2022/04/the-new-york-times-would-r...
We probably agree that an arrest warrant is not “appropriate based on the facts,” but that’s a different matter.
The article explicitly stated that the victim picked her photo out of a line-up. The facial recognition was NOT the basis for the arrest.
"Five days after the carjacking, the police report said, the detective assigned to the case asked the victim to look at the mug shots of six Black women, commonly called a “six-pack photo lineup.” Ms. Woodruff’s photo was among them. He identified Ms. Woodruff as the woman he had been with. That was the basis for her arrest, according to the police report. (The police did not say whether another woman has since been charged in the case.)"
Then what is the likelihood of the victim being misled by the doppelgänger the computer found?
It is ridiculous to pretend like this wasn’t a software pick.
It is ridiculous to pretend this WAS just a software pick when the article is clear that the victim picked the woman's photo from a line-up of MULTIPLE photos.
There ought to be a sensible procedure on how to approach people who have been flagged by facial recognition.
Now, I don't know what is the normal procedure in the US when police think they have recognised someone who's wanted. Perhaps they do exactly the same as they did to that woman in any case...
At least in countries where everyone carries an ID card police could ask for ID first or the person could show their ID to at least and hopefully keep the police in "not sure, let's go easy" mode.
I would suggest 'not'.
It's arguable than in most instances they don't overreact and throw pregnant women into jail for half a day. In a country of 330,000,000 people, you would expect to see stories like this all day, every day, if it were even a little bit routine. The fact that this is national news gives some idea how out of the ordinary it is.
We're reading (rightfully) about this woman's story because it has all of the "right" parts: her innocence is painfully obvious, she's pregnant, a professional, a working mother of two, etc. There are others just like her, and the evidence we have points to there being many others just like her; they deserve attention as well, and not to be erased as statistical anomalies.
I think I could make a plausibly convincing argument that the rate of abuse is low enough that the damage to police<->public relationship from media playing up these incidents has actually led to a more dangerous situation than the incidents themselves.
Not quite ready to go there, but I think it's worth at least considering that everyone can fall prey to media hype and blow little things entirely out of proportion, leading to a worse overall outcome. We do want to improve oversight of the police, but making them out as a public enemy seems like it might be counterproductive.
In this situation, the reporting seems pretty dispassionate to me (arguably more than it should be, given that a pregnant woman was arrested for a crime she manifestly could not have committed). I don't see a lack of proportion here.
The victim physically picked the woman out of a lineup of mug shot photos placed in front of him. The software placed the woman’s photo in the lineup but the victim of the crime literally ID’d her.
Wrongly, of course, but seem like a pretty crucial detail. This is more of an old fashioned police are idiots story than anything else.
Like how about investigating instead of just arresting her. Like was she near the crime, did she have an alibi? Etc.
More than one party can share responsibility in a situation, and in this situation both the police and the software should have blame identified with them.
The point is that there has to be MORE than "This person has a similar face". It's a confirmation step for suspects that already exist, or for whom you can find other strong circumstantial evidence for. You have an unlimited supply of similar faces and you don't get to just choose one of them.
The great problems with AI tools mostly arise from the fact that the supposed expert users were never taught what the tools do, or how causal inference works - how to discern reality from the evidence available, how to test hypotheses, how to change your reasoning to be valid in the face of wildly disparate scales. If you don't have at least the beginning of Bayesian probability, and you're feeding large datasets into policing, you have no fucking clue what you're doing and your ignorance is immediately dangerous.
> It's doing the same part of the task as a photo lineup does.
It's doing something subtly different, and arguably pernicious: a lineup is meant to contain a variety of individuals of varying similarities, so that the person selecting from the lineup are themselves tested in their recollection. We have no evidence that this system is being used in this way; instead, it seems to be finding the closest match in an existing dataset under the assumption that anybody who's ever been photographed by the police is permanently a criminal suspect.
So is a witness, for the most part, they're just logistically challenged. Which is why photo lineups are, in my alternate universe where everybody understands probability and causal inference (because we require 4-year degrees in the science of criminology), a specific tool to discover specific things rather than an instantaneous indictment. There are myriad things wrong with using them, and if the detective doesn't understand that or worse wants to exploit that, the detective should be regarded no differently than the beat cop shooting their gun into the crowd to scare it off.
https://lawcomic.net/guide/?page_id=5
Similarly: A detective's hunch about whether the person being interrogated committed a crime is a powerful tool. We just don't throw people in prison based on that alone. We're getting to the point that we could train an AI to develop such hunches, and that could be valuable. But we can't just throw people in prison based on that alone. We need to confirm such things with external evidence.
Just like we need to confirm a single-eyewitness recollection or a rough photo, rather than just measure that recollection & numerically characterize that photo without any indication (or more commonly, with a 100% fraudulent indication) of probability of matching.
The article gives a perfect description of how it happened. The software picked her because she looked like the criminal, and then they put her in the lineup.
They wouldn’t have had a photo lineup without the software.
That’s the whole point. They use the fig leaf of the victim ID, to cover the fact that it’s just facial recognition at the core. The article points out how likely somebody is to mis-ID somebody who the computer has already found is a likely match.
Sure, blame the police. But for what? For using facial recognition to ID and arrest people!
The proper way to use this tool seems to be
1. Find all the people who look similar to the low quality pic we have of the suspect.
2. Investigate those people to see if they could have done it.
3. Arrest someone who you hopefully now have additional evidence against besides just "they look like a grainy photo of the suspect"
Just throwing a bunch of people who look like the suspect at the victim and just arresting whoever the victim picks seems completely insane. But also, like it doesn't have too much to do with "Facial recognition" broadly speaking and more to do with completely idiotic investigative practices.
But I also don't think it's too cynical to believe that at this point, almost everybody up and down the chain knows that these tools don't work very well, but what they like is that they are allowed to defer to these tools. It's like how nobody is fired for buying IBM -- nobody is at fault for relying on the algorithm.
This applies to many domains, even consumer tech stuff, hiring practices, etc. Algorithms are just a way to have a color by numbers process where employees have no agency and deserve very little pay, and management has no responsibility and cannot be blamed for mistakes.
HN applies ample and eager skepticism to all kinds of scientific and science-adjacent fields; I eagerly await the day that we do the same for forensic "science."
[1]: https://www.theatlantic.com/magazine/archive/2016/06/a-reaso...
"Wait, the person who reported the crime didn't mention that, surely he would have noticed while they were drinking and fucking and reported it as an obvious distinguishing feature!?"
But nope, just follow through, and I'm willing to bet that for every step in the process the people involved thought it was silly too, but "just followed the system", either because they had no power to do anything else, or out of apathy, or maybe both.
Literally no one won with this: the police and criminal justice system wasted time on an obvious wild goose chase, the victim wasn't helped, and presumably more victims will be created and thus even more work for police.
“It is circular and dangerous,” Dr. Wells said. “You’ve got a very powerful tool that, if it searches enough faces, will always yield people who look like the person on the surveillance image.”
Dr. Wells said the technology compounds an existing problem with eyewitnesses. “They assume when you show them a six-pack, the real person is there,” he said.
A facial recognition match should just be the start of one possible line of inquiry, and nothing more.
I think the real issue is that the US justice system rewards and penalizes based on the number of convictions achieved. Low = bad, high = good. But high convictions does not equate to promoting justice or domestic tranquility.
Her Maxmind story is campfire-good https://fusion.net/story/287592/internet-mapping-glitch-kans...
https://gizmodo.com/facial-recognition-new-orleans-police-no...
Tool says you; arrest.
It’s obviously not the worst part of this story, but how on earth is anyone supposed to get actual, reliable facial identification from security cameras? Those things have hilariously low resolution on a _good_ day.
> Ms. Woodruff is the sixth person to report being falsely accused of a crime as a result of facial recognition technology used by police to match an unknown offender’s face to a photo in a database. All six people have been Black;
FacRec's history precedes it. The darker the skin the higher the false positive rate. https://duckduckgo.com/?hps=1&q=facial+recognition+skin+colo...
Surveillance abettors in general (and LEO specifically) haven't been overly put-off by that feature.
I see what you did there. It totally is a feature, not a bug.
I don't live in the US, but from abroad I can tell how messed up the police are over there. The number of stories of brutality, deaths and upright bullying of every day citizens...
It's really messed up, perhaps Americans are desensitised, but I can tell you from the other side of the world - it's nowhere like this in other countries, you guys have a BIG problem with how your cops are selected and trained.
Corruption is incredibly common overseas (yes, Europe too). Privacy laws regarding law enforcement are much worse.
There’s a really good video on YouTube called “Never Talk To Police.” It’s a law professor and police officer explaining to law students why they should instruct their clients not to talk to police.
The police officer compares American and foreign practices, and emphasizes US police use deception, but foreign police use force.
I’m sure bad and good police agencies in America and abroad. The best PD I’ve interacted with is in a mid-sized UK city. The worst I’ve seen is in a major American city.
Semi automatic deception that ranks the US 30 in term of death by police, in the middle of third world countries, far away from any developed countries
Translation I get all of my information about American policing through highly politicized and charged news storied that become stories only because they are exceptional and then feel justified passing judgment on an entire country that has 3 times the people and land as my country.
To say our policing systems are bad is a smear on the hard-working men and women in those systems. The US simply has challenges that European nations do not have. You can (and maybe should) debate whether or not those challenges should be present in the first place but blaming those challenges on the police is just mean.
Umm, no:
https://www.washingtonpost.com/news/worldviews/wp/2013/05/16...
I've read over and over again that intelligence is a disqualifying characteristic for being a LEO. They want people to follow orders and be a good partner (e.g., never go against what your partner says).
My experience in CA since the 80's is that LEOs are very corrupt and really only care about their own. There are exceptions. I live in an unusual city, LEO-wise, but it is totally not the norm.
This went to court and the man who sued for discrimination lost his case.
> -- A man whose bid to become a police officer was rejected after he scored too high on an intelligence test has lost an appeal in his federal lawsuit against the city.
> The 2nd U.S. Circuit Court of Appeals in New York upheld a lower court’s decision that the city did not discriminate against Robert Jordan because the same standards were applied to everyone who took the test.
Smart, well-educated police are how you end up with Frank Serpico[2]. Most police forces don't want men like Serpico on their forces.
[1]: https://abcnews.go.com/US/court-oks-barring-high-iqs-cops/st...
Police in the US have the least training of all Developed countries on Earth [1].
The results of less training are very clear.
Yeah, you might not be getting a complete or accurate picture.
The US is geographically huge and has a large, distributed, heterogeneous population. Be careful making broad generalizations about 330M people in 10s of thousands of cities, in 50 states across 3.5M square miles of land.
Awful stuff happens sometimes. That doesn't mean it's the norm.