General Discussion
Related: Editorials & Other Articles, Issue Forums, Alliance Forums, Region ForumsEight Months Pregnant and Arrested After False Facial Recognition Match
https://www.nytimes.com/2023/08/06/technology/facial-recognition-false-arrest.htmlhttps://archive.ph/LZDcD
Handcuffed in front of her home on a Thursday morning last February, leaving her crying children with her fiancé, Ms. Woodruff was taken to the Detroit Detention Center. She said she was held for 11 hours, questioned about a crime she said she had no knowledge of, and had her iPhone seized to be searched for evidence.
The ordeal started with an automated facial recognition search, according to an investigators report from the Detroit Police Department. 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 offenders face to a photo in a database. All six people have been Black; Ms. Woodruff is the first woman to report it happening to her.
It is the third case involving the Detroit Police Department, which runs, on average, 125 facial recognition searches a year, almost entirely on Black men, according to weekly reports about the technologys use provided by the police to Detroits Board of Police Commissioners, a civilian oversight group. Critics of the technology say the cases expose its weaknesses and the dangers posed to innocent people.
niyad
(113,995 posts)The software used a 2015 mugshot as comparison versus a 2021 DVM photo they had available.
Who gets to pick the comparison photo cuz no one would pick the one on the left if they wanted as accurate a list of suspects as possible. It's out of focus, it's older. WTF? The article says the final approval to arrest someone is given by a human so who picks the comparison photo?
Hermit-The-Prog
(33,596 posts)The splash of light on the forehead is somewhat similar in the two images, but the foreheads are not. Eyes (postion, shape, protrusion), cheekbones, chin, facial shape (including proportion of height to width of the skull) -- all different.
That's some buggy software and buggy wetware.
Ms. Toad
(34,137 posts)I do agree that they look very different - which suggests a different problem with using older mug shot photos for current identification (whether or not current ones are available).
Hermit-The-Prog
(33,596 posts)Some person chose to use a 6 year old, badly focused, badly lighted, badly posed photo of a young woman to feed some buggy software. Young folks change. Attitudes and health change photo appearances as much as lighting does. Unless the crime is from 6 years ago, the person choosing that photo on the left needs some chastising.
I'm ancient, so a 6 y.o. photo of me will just show less saggy facial skin, very slightly less gray in the hair, and wrinkles that are slightly less deep. For each decade you lop off, that 6 years becomes more significant.
Ms. Toad
(34,137 posts)was recognizable from my high school photos.
Mostly I've had long hair, so that's part of it, but some of the intermediate photos have short hair. But when I got cancer 2.5 years ago, the long hair went when no one in my family was willing to care for it while I healed from a skin graft. I have difficulty recognizing photos now - the appearance change was that dramatic. (And my earlier short hair photos resemble the long-hair photos more than they resemble my current short-hair photos.)
So until 2.5 years ago - as long as it was a decent photo, even a decades old photo would have been reasonably accurate.
Hermit-The-Prog
(33,596 posts)Photo portraits are 1-eyed, 2d projections of a 3d irregular object. Cameras can only record light. (Hollywood depends on those limitations as it uses lighting and angles to transform ordinary people into monsters and ghouls).
I hope you are doing well at this time, and are taking care of yourself.
Ms. Toad
(34,137 posts)to 6 month monitoring for metastasis, and the doctor and I agreed that based on everything we know so far, my risk is low enough that I can use X-rays (rather than the higher radiation, but more precise, CT scans). I'll probably have a CT scan at 5 years, just to confirm we aren't missing anything.
So far - 2.5 years NED (with a chance of metastasis in the future in the range of 3-4%). (But taking care of myself - and advocating for myself - is why an aggressive cancer posed less risk to me than most. I had been monitoring what my research told me was a suspicious area for several years and was able to convince doctors to remove it at 1.5 cm, rather than the standard golf-ball sized tumor. Even so, it was Grade 2 (out of 3) less than 2 months after I noticed it. Why they wait to even think about sarcoma until it gets to be golf-ball sized makes no sense to me.)
Hermit-The-Prog
(33,596 posts)It's not such good news to learn of the "wait and see" approach. I had assumed that the predominant attitude now is to treat cancer aggressively and early.
Ms. Toad
(34,137 posts)but spend a lot of time in a sarcoma support group urging people to advocate for themselves.
It is a complicated issue. Sarcomas are rare (about 2% of all cancers). Within that 2%, there are between 50 and 100 sub-types, all aggressive, but each with a different risk profile. Sarcomas are good mimics of some benign (much more common) conditions - so the assumption is (especially if you are not at a sarcoma center) that it is the benign thing. Sarcoma often doens't even cross the mind of routine care providers.
The advice is also to never let anyone remove a sarcoma unless they are a sarcoma specialist (it has to be treated differently than other cancers). But there's no clear way to determine if it is a sarcoma without removing it - AND - many/most sarcoma specialists won't see you until you are diagnosed. So if you get someone else to remove it or biopsy it, you are pretty much guaranteed a second surgery - and may have dramatically increased your risk of metastasis by making it harder for the doctor to get the 1 cm margins they need.
The standard criteria for evaluating for a sarcoma is golf-ball sized and growing. But - by the time it is golf-ball sized and growing, it is often too late. Mine was pea-sized when I noticed it. By a month later when I was seen by a nurse practitioner it was kidney-bean sized. Two weeks later when they removed as much as they could get it had grown 50% more. And, when I had the second surgery two weeks after that, it had grown back to kidney-bean or larger size.
Mine was what was called an "oops" surgery (removed by someone who was not a sarcoma specialist). They believed it was benign. I was pretty sure it was cancerous (but wasn't specifically thinking about sarcoma). This regrowth was very different from the previous ones - and there was enough question that my dermatologist agreed to remove it.
I had a dermatofibroma (DF) removed in ~2014. My research indicated that even though DFs are normally benign, there are a handful of characteristics that suggest it might be mimicking a DF and should be treated as pre-cancerous. My DF had all but one of them. So I had several intermediate biopsies of suspicious regrowths, and had been watching it very carefully.
There has to be a better way of finding sarcomas than an alert patient who read enough medical articles to know of the risk, and who badgered her doctors enough to get a biopsy. The people I chat with in the sarcoma support group have tumors between golf-ball sized and watermelon sized - and some in the lower ranges (say up to baseball size) still have to battle their doctors for an evaluation.
LisaL
(44,986 posts)Obviously mug shot database is something police is going to search with their software. The problem is. the software is obviously not perfect.
But both photos are of her.
So clearly there is a resemblance in both photos to her since they are both of her.
What we were not shown is the photo of the actual alleged perp.
Hermit-The-Prog
(33,596 posts)Angle of the head, lighting, age --
See my response (#17) to Ms. Toad, above.
There were buggy humans and buggy programs involved.
LisaL
(44,986 posts)So clearly the photos resemble each other and her, regardless of the angle. Just because she is older in one of the photos doesn't mean she is no longer recognizable from her earlier photo.
Photo of the alleged perp is not included in the article, but likely that one wasn't of best quality either, since it was taken by a surveillance camera.
Hermit-The-Prog
(33,596 posts)I argue the opposite of your statement, "clearly the photos resemble each other".
They barely resemble each other due to:
1. bad focus
2. bad lighting
3. different angles of the head in each shot
4. different ages
5. different health status at the time of the shoots
The focus error alone leaves all of the image too ambiguous for using software as a basis for anything other than getting a human to override it. Once a person saw that and got the name, that person should have used the better photo. Add in the bad lighting (yields hollow cheeks among other things), different angles (yields a receding chin), different ages and health status (differences in the neck, fullness of face), and it's a defense attorney's dream. Somebody should have stopped this thing early on.
I could not read the whole article -- paywall by NYT and a "no secure connection" error when attempting to access the archive link.
If the 'alleged perp' resembles that bad photo, it makes the arrest even worse. The two photos arguably look like two different people in the ways I outlined in my reply to CrispyQ above. The cops had both photos available to them before accusing this young woman.
LisaL
(44,986 posts)Are you arguing that unless the photos are at exact same angle, you can't recognize them being of the same person?
MagickMuffin
(15,985 posts)Thats how enforcers think.
Im curious after false arrest and mugshots and fingerprints are taken,
What becomes of them?
Do they get taken out of the system?
Or
Do they stay in the system to further any future crimes committed by another person?
Racist enforcers doing what they can to eliminate and humiliate others!
orthoclad
(2,910 posts)Jan = January
Feb = February
AI says Yep got it....
Marary
Aprary
.......
bullimiami
(13,115 posts)in that case fire them all. ai can do that without them.
LisaL
(44,986 posts)They used that as a basis for the arrest. Presumably she somewhat resembles an actual perp.
bullimiami
(13,115 posts)LisaL
(44,986 posts)NT
orthoclad
(2,910 posts)against the police and prosecutor, but mainly the tech-bros who wrote that POS software. Enough to bankrupt them for destroying people's lives.
"Entrepreneurs". Right.
KentuckyWoman
(6,701 posts)The software needs to be better if they are going to use it. Also, we need laws to stop police from making arrests solely on facial recognition.
orthoclad
(2,910 posts)I think bankruptcy court would be even more effective against tech parasites.
2naSalit
(87,005 posts)erronis
(15,481 posts)and individuals, personally, who permitted this.
We need to get rid of qualified immunity for public officials when they aren't doing their jobs.
orthoclad
(2,910 posts)the class action suits in order to pressure them to change policy, but taking down the profiteering tech bros would get the biggest bang. Make these fools afraid to release dangerous tech into the wild.
Maybe sue the administrative drones who approved the use of facial rec. They might not be police, and thus might not be qualified immune.
It's ironic that the citizenry who are victims of bad police behavior have to pay for bad police behavior. Now THAT is a racket.
DFW
(54,527 posts)And it should come equally out of the pension fund of the Detroit Police and from the AI company that provided the software.
NO explanation, rationalization or excuse is enough. I hope Governor Whitmer has something to say about this.
I get the need to be tough on crime, but if THIS incident isnt a crime, then I dont know what is.
LiberalFighter
(51,393 posts)WhiskeyGrinder
(22,527 posts)orthoclad
(2,910 posts)When police unions are weakened.
When qualified immunity is defeated.
Personally, I think instead of "training" police (heh, klansmen know how to game the system) for proper behavior, all police and cadets should have to pass psychological testing, where objectively measurable autonomic responses like blood pressure, skin conductivity, heart rate, arousal, whatever, are recorded in response to scenes of violence and abuse. Add in a racial bias component. I think there's plenty of research available on the subject.
The ones who get off on these triggers don't go on the street, make arrest decisions, train, or supervise. Preferably fired, but there might be legal obstacles. In that case, put them someplace as harmless as possible. Also, the test results should follow them if they apply to other departments or private security.
This alone will filter a lot of "bad apples" out -- the ones eager to beat and abuse citizens.