Slashdot Mirror
ACLU Examines Face-Recognition System
nate_drake and others wrote in about an ACLU report on face-recognition (PDF) (see also their press release and an MSNBC article). We've posted several previous stories about the Tampa police using face-recognition systems at the Super Bowl and on the streets of Ybor City.
So really, how well can this work? Half the porple I know look completely diferent from day to day anyway . . .
But take this for an example:
Soon after Sept 11th one of my colegues flew east on a buisness assignment. He was concerned because, as he said "Did I forget to mention, I look like one of those terrorists?"
He did, a little. Darker complextion, goatee, curlly hair, glasses. He looked more than a little like one of the suspects.
He's sinced shaved his goatee, so he no longer looks like that . . . but would this system have been abel to diferentiate?
I also seem to remeber that one of the guys they arested and incarcerated in Tampa turned out to be the wrong guy . . . .
\Drew National Data Director, John Edwards for President
After they take these digenerate pieces of sh*t down they need to go after the 'lie detectors'.
Soon we will have a camera on the market which is *AS* reliable as a polygraph test. 25% of the time they are wrong.
All of these things are a Bad Thing. These cameras will be used next to track you at the game [like the Bears? get calls at dinner about season ticket prices], at the bar [you like to go to bar!], at the strip club [the cops watch you now because they know you're a perv!]...
Just wait till this stuff goes all private!
Reminds me of Demolition Man
Get your Unix fortune now!
all this means is the companies developing this stuff will have to improve their face-matching algorithm and then we'll all be back at square one.
It's worse than that:
How long until these companies start lobbying the gov't for mandatory inclusion of, say, license photos in the pool of database data so that people can be picked up as soon as they do something?
- The system has never correctly identified a single face in its database of suspects, let alone resulted in any arrests.
- The system made many false positives, including such errors as confusing what were to a human easily identifiable male and female images.
- The photographic database contains a broader selection of the population than just criminals wanted by the police, including such people as those who might have "valuable intelligence" for the police, or who have criminal records.
I wonder why they didn't mention that man who was a demo face for the system, and was subsequently misidentified then questioned as a felon. Guess it didn't make the logs.I used to park my bike in the garage beneath my work. The bike rack was about 3 meters from the attendent, and covered by a video camera. I had a cheap lock, but Ithought I was safe.
One day, I came to the rack to find my lock cut and my bike stolen. The attendant refused to talk to the police. But that's OK, I had the perpertrator on camera.
After getting the tape from the building security people, I took it to a camera shop. We sped through it to find the point where, sure enough,m you could see a guy walk up, try on my bike helmut, and ride off with the bike. Due to position, you could not see him cut the lock.
I say him because I am pretty sure that it was a male. That was all I could tell from the poor quality of the tape. I could not tell skin color, clothes, hair color, or enough facial features to recognize.
I don't think the best AI added to this image would have been able to do anything as far a facial recognition.
I wouldn't want to be the attenandt working that booth. After they find his mangled corpse (ala Fargo) the police will tell his widow, "Sorry Ma'am, all we can say for certain is that they were in some sort of automiblie. We think a sedan, but we don't know for sure."
Facial recognition is going to be even harder than this. As a programmer, you have two choices , go with an algorithm or try to use a neural network.
Most of the weaknesses in the algorthim approach are what the ACLU document was complaining about.
A neural network may work if you are looking for a specific person. The problem is that to identify two things as being different, they need to be as orthoganl as possible. To separate the sea of faces into two groups those we are loking for , and those we aren't based on a series of images is going to be nigh unto imposible. Certainly not with the amout of computing power per camera that they would put into it.
Open Source Identity Management: FreeIPA.org
As I understood it, the issue with facial recognition was the possibility of false positives; ie, I'm just trying to watch the SuperBowl. the FR system tags me as a known terrorist (incorrectly :) and the next thing I know I'm being dragged off to the can for some serious interregation (and not only unjustly tramatized, but I miss the game too)
But from the ACLU's press release, there was always a human step in the process, where a real live human being would examine each purported match before anybody got dragged off anywhere.
As such, all the face recognition software is is a _filter_, cutting down on the number of people a human agent must examine. Where's the problem?
After all, law enforcement officers have placed themselves in public places, looking for people they knew, for probably as long as there have been law enforcement officers.
A friend of mine was a sergent in the British Army, and he did a few tours in Northern Ireland. Part of his training was memorizing the faces of a large number of known IRA "players" (and apparently the IRA did the same thing with British soldiers' faces)
How is this any different?
I guess I don't understand the ACLU's beef here.
DG
Want to learn about race cars? Read my Book
This facial recognition has a near-zero Hit rate and a high false-positive (Type III error???) rate. The false-positive rate is a killer because it may cause system operators to miss a Hit (true positive). So what do we end up with: an authoritarian tool that is completely worthless.
Meanwhile, the failure of this project can be a selling point for Larry Ellison's proposed National ID card system. Perhaps the streetlamp cameras in Ybor City will soon be replaced by turnstiles manned by undereducated, undermotivated, understimulated, minimum-wage-earning Security Engineers (read: displaced airport security screeners) checking each person's National ID card. These people probably won't be able to grasp the concept of Type II/III errors; thus the implementation of the National ID Card will suffer from the same problems as the facial recognition system.
In summary the two vendors will profit substantially from their products--which won't make the public any safer--and we will be eased into acceptance of the mercantilist authoritarian police state.
"What is the sound of one belly slapping?"
So I would say it is definately worth it to fight very hard against bad technology right from the start.
sPh
For more on McDonald's over-hyping of the emperor's new security blanket, see this article from the Fresno Bee .
Sample quote: "This breakthrough technology makes us the safest airport in America," Mayor Alan Autry said.
Quiquid latine dictum sit altum viditur
Your local convenience store can't detain you.
Well, technically they can. But the public, including that rent-a-cop in the convenience store, can only detain a person 1) if they witnessed a crime and 2) to turn them over to a sworn police officer at the earliest possible time. If either piece is missing, you can nail them for "false arrest." This is an important thing to remember if you're ever (wrongly) accused of shoplifting - demand a real cop, *now*, to either arrest you or release you. If they refuse to call the cops... life will soon get *very* interesting.
Even those bounty hunters have limited rights. They can detain someone who signed the bond papers, but there are some well-documented cases where the bounty hunters were prosecuted for kidnapping after detaining the wrong person and failing to exercise due diligence in verifying the identity of that person.
But sworn police officers can detain people even if the officer didn't witness a crime. They can detain people even if there's no witnesses at hand, e.g., if they reasonably believe that the person is the subject of an arrest warrant issued by another jurisdiction.
The cost of a false positive in a convenience store is minimal. They think you're a shoplifter because of their face recognition software? Fine, you walk away and shop at another store where they're more careful with their accusations.
But a false positive with a police officer may have you arrested, at gunpoint, and detained for hours or days until you can prove that you aren't the escaped mass murderer you resemble.
(IANAL, but this is stuff that should be required knowledge for a walking around on the street!)
For every complex problem there is an answer that is clear, simple, and wrong. -- H L Mencken
Remember the good ol' days of th'Internet, where no one could possibly track you and where anonymity was technologically "guaranteed"
My point is, that arguing the TECHNICAL weaknesses of this, or any other privacy-infringing item/product/software/etc. will only result in TECHNICAL innovations that make it more effective.
We must argue the LEGAL weaknesses - the 4th amendment. We need to argue that no person waives their constitutional rights simply by the virtue of entering a commercial, travel, or other legal relationship with any other entity. (unfortunately, I fear we lost this one a long time ago)
We need to argue against clickwrap agreements, and their cousins:
Our legal rights are important. The details of whatever technology the FBI, CIA, or any other no - such - agency uses in an attempt to violate those rights, are less so.
Don't Frustrate their efforts. Fight them head - on!
The Orwellian danger is having a camera on every curb, and the potential for John Q. Public to be under constant surveillance. I'd expect that most people would be upset or put off if the police followed them around all day watching what they do. The police argue that the contentious and widespread use of cameras is justifiable because facial recognition gives them a legitimate and bias-nuetral use.
The ACLU is saying that the system doesn't work, so there is no good reason for deploying lots of cameras. (Actually they are saying more than that, but that's the part that's easy to agree with.) In any case, it's clear that the survelliance capacity presents a danger even if the intended purpose for the system doesn't work.
How will this software be able to truly establish it's me if my photo in the database is 40 pounds lighter and with contact lenses if I walk around with my glasses on now.
I also wear false teeth so I can change my jawline at will.
Come and get me!
I've hit Karma 50 and gotten a Score:5, Troll... I win!
I put a visible non-working video camera pointed at the street and our local crack-house, and their business collapsed. The house in question was rented by a slum-lord to the lowest bidder and the drug traffic was driving me nuts. I made a good show of the camera and suddenly, the 'customers' were a bit leery. The druggies soon moved out and were replaced by a rather nice poor family.
I was set to get the camera working, but the it's presence was enough. Highly recomended. PS: I removed the camera once it became obvious that the new tenants were cool.
Moneyed corporations, non-working 'poor' and criminal prisoners are turning productive citizens into tax-slaves.
According to the report, it almost doesn't work at all. Other installations using face recognition have degenerated into checking out the girls. I understand; it's gotta be boring as can be after the first 50,000 false alarms.
It seems to me that this software isn't really an invasion of our privacy. The cameras themselves may be, but if we accept the cameras, we can't really quibble about the face recognition software.
It's time to reconsider our concepts of privacy, anyway. Read David Brin's The Transparent Society and see if you don't come up with a new view.
I think the cameras should be everywhere... especially the police station. And we should all be allowed to watch them. It would certainly make everyone think twice about their biases before taking drastic action.
For geek dads: Contraction Timer
If you read up on the history of the polygraph, this is clearly what is going on. Faced with a choice of admitting that they don't know who the heck committed a crime, or using a demonstrably useless machine, investigative agencies (particularly the FBI) have been using the useless machine for 70 years. And have now made such a culture of using it that they dare not admit to anyone, particularly themselves, that it just doesn't work.
sPh
Back In late September I read and/or heard about these face recognition devices. According to statistics, the results were exactly correct.
Assume a 90% accuracy with a database of 10 criminals. If there are 100,000 samples, 10% will be false positives. That's 10,000 alarms that mean nothing. Of the remaining 90,000 people, you'll get 9 true alarms (assume they walk by!) and 1 gets away. Therefore you have to fend off about a thousand errors to get a real suspect. At that point, human error could easily step in and assume it's another false alarm.
This is the same objective reason racial profiling is wrong. You spend all your time and resources harrassing innocent people.
Of course, a cheap solution to this problem is multiple camera angles.
Yeah, go ahead and keep thinking that the pictures just go away. And believe the FBI when they promise that they NEVER read the content part of the emails carnivore sniffs. I think you are grossly optimistic about what would cause a very very large public outcry (example: well documented evidence of ongoing wiretap abuse by the LAPD. No public outcry at all).
We should hold public servants to a (much) higher moral standard than we hold private individuals, and there should be signifigant oversight. I also don't agree that there IS a legitimate use for widescale monitoring of innocent, private people - one of the (theoretical) keystones of our legal system is that people are innocent until they are PROVED guilty, and that it's better to let 100 guilty people go than convict 1 innocent person. The modern corporate/government culture of assuming people will be criminals unless you stop them cuts at the foundations of our society.
Enter Facex. That's essentially what this face recognition stuff is doing - compiling a profile of a face that it sees, and then checking for other profiles that are similar. Given the multitude of conditions that can affect what it sees (lighting, angle, and facial expression to name a few), it's going to take some very ingenious programming to overcome these influences. Further, I'd love to see the ACLU (or other party) conduct some studies to see just how innaccurate this system is when dealing with other factors - like whether or not someone is wearing sunglasses, whether or not they have a beard and/or moustache, makeup (lipstick, blush, eye shadow), whether or not they're wearing a hat, etc.
FaceIt, it appears, is an overhyped technology that isn't any where near ready for prime time. And knowing what it's up against, I don't think it ever will be.