The tech will improve. Fingerprint data from cards use to be high tech.
Also the William & Will West case was an older example of misidentifying by photography. Modern tech has improved dramatically and will get more and more precise. Why apply early 2000’s tech issues to the advancements made in the past 15-20 years?
In this thread people forget that nothing is truly perfect or secure. But it does not make it worthless.
A threat model exists for a reason. A risk reduction model exists for a reason. Using a tool (computer/ML) to do 95% of the pattern matching is called one thing: not being a dumbass.
I sincerely wonder if any of these super-hostile anti-facial ID activists have truly had someone close to them impacted by a crime, and had said crime go unsolved. Or are they merely idealistic loafers, clutching their pearls, to signal how "woke" they are?
Lie detectors are woefully unreliable, but PDs use them.
Forensic “matching” of bullets to guns is basically pseudoscience, but continues to be used in court to convict people.
Roadside drug analysis kits are less useful than dowsing rods, but get people arrested on a daily basis. People that can’t afford bail, and plead guilty just to get back to their lives, despite the fact that an actual defense attorney would have had the case tossed in the trash.
Previous experience suggests that police tech does not just “get better.” They can’t even progress past “disproven pseudoscience.” And the innocent people that have their lives messed with while waiting for the tech to catch up might not be indifferent to the process.
I have heard of it being bunkum, but not quite of the same ~caliber~ as fiber matching and bite-mark matching.
It can, at best, match ammo from the same machine-manufactured batch to a factory-stock model of gun. It automatically fails on hand-load cartridges, or ammo that the analyst cannot readily source.
When there is a match, the prosecutor might say "this bullet was fired from this gun", but the science says "a bullet similar to the one in evidence was marked and deformed in a similar fashion, when fired by the weapon in evidence (just as it would with every other factory-stock copy of this model of gun)".
It's like matching a nail to the hammer that drove it. If the test nail is already different from the reference nail, you cannot match it. If you have two or more identical hammers, you can't tell which one drove the nail. The hammer factory made thousands of identical hammers.
In most cases, the tests are likely saying "the most commonly sold type of ammunition was fired from the most popular model of gun of that caliber". It's the sort of thing that should probably be used exclusively as exculpatory evidence. I.e. this model of gun always leaves distinctive markings, such as from an off-center diamond-shaped firing pin, or from an uncommon rifling twist, that were not found on the bullets in evidence, so the gun in evidence could not have fired them. Otherwise, a gun substantially similar, with all the same machine-made parts, could leave the same marks, so you could never know for sure which one fired the bullets.
The only time it would be useful to convict is if someone used a unique, hand-made, gunsmithed gun to fire factory-made ammunition. And the number of (violent) crimes committed with that type of display-piece gun approaches zero, because they tend to be both expensive and less fit for use as a weapon than cheaper factory-built guns.
and you don't come to the attention of the police by being innocent so it's safe to assume you're guilty and watch the evidence appear. And this is why the advice is to say nothing whatsoever to police. Don't help them get the falsehoods out of their case anywhere but in front of a judge in a court. And the police may not be bad people or involved in organised crime.
All this tech makes a turnkey police state right there waiting. Who is going to turn that key before we dismantle it? Think of the children.
In the past, if a license plate was noted, you’d have to manually go through records. Now through automation and scale it’s instant. Should that be prohibited too?
> Commission President Anthony Pacheco said Friday that he was highly concerned after learning police have arrested at least two innocent people because of faulty fingerprint analysis.
What happened to "It is better that ten guilty persons escape than that one innocent suffer." As expressed by the English jurist William Blackstone in his seminal work, Commentaries on the Laws of England, published in the 1760s.
Seems you're willing to punish an innocent man and a guilty man just to make sure a committed crime is followed up with a conviction.
Innocent people are killed by the same murderers in both scenarios. In yours, they are also sometimes killed or imprisoned by the state and are therefore at more total risk.
Considering fingerprint analysis is pseudoscience and almost entirely made up (along with most other "forensic science") [1] [2] we can assume LOTS AND LOTS of innocent people are caught with "fingerprint data"
>"There is some evidence that fingerprints are unique to each person, and it is plausible that careful analysis could accurately discern whether two prints have a common source, the report says. However, claims that these analyses have zero-error rates are not plausible; uniqueness does not guarantee that two individuals' prints are always sufficiently different that they could not be confused, for example. Studies should accumulate data on how much a person's fingerprints vary from impression to impression, as well as the degree to which fingerprints vary across a population. With this kind of research, examiners could begin to attach confidence limits to conclusions about whether a print is linked to a particular person."
>Considering fingerprint analysis is pseudoscience and almost entirely made up
Quite the stretch to go from "non-zero-error rates" to "made up pseudoscience" isn't it?
It's a problem if people are being convicted on fingerprint analysis alone. But as a piece of the puzzle, when combined with other evidence, what is the issue?
How about the fact that those fingerprints Haven long been stolen and likely abused by all sorts of undergeound criminals? Even federal workers fingerprints were stolen in the OPM hack.
That is essentially the problem to those of us who oppose it.
When you say "criminals" you are probably thinking of strictly people who harm others, but the state is thinking of anybody who breaks its laws. Currently many of those laws exist solely to keep certain groups in power, and there is no guarantee it won't get even worse in the future.
Human law enforcement ensures layers of decisionmakers who are at least theoretically capable of empathy, and limited manpower makes them prioritize the worst or most flagrant crimes. Automated law enforcement gives a smaller group of people horrifyingly granular levels of control over all of society.
I often criticize SF for being overly zealous around weird issues...but their ban on facial recognition is looking more prescient by the day. I hope other cities follow suit.
Also, non sequitur, have there been any proposals for how to detect when your country has become an authoritarian dystopia?
This is something I think is justifiable in time-sensitive cases like tracking down a kidnapper, but I fear there's no way to grant such power and expect its use to always remain narrow. In my opinion, most issues in our country come from the compounding of lots of well-intentioned systems built up over several centuries that no longer even remotely serve their original purposes.
That's how every infringement of our civil liberties has come into existence: It has one or two very limited justifiable cases, and everyone goes 100% all in on that and focuses on that, but not what we will lose. Everyone always dismisses it as a slippery slope fallacy but here we are, with our rights against illegal search and seizures almost completely chipped away. Asking an officer for a search warrant is almost a formality at this point. Cops routinely exercise civil asset forfeiture and make it impossible to regain the goods lost, the 3 letter agencies are downloading and analyzing anything we type online and building comprehensive profiles and behavior models and running our actions online against said models, etc, etc.
I see far more slippery slopes than slippery slope fallacies. It makes me want to reject these kinds of ideas from the get-go, because the pasts of our bureaucracies and representatives have given us really no good reason to believe every great new idea won't be expanded far beyond its scope sooner or later.
> every great new idea won't be expanded far beyond its scope sooner or later
And when the expansion is justified by shocking statements ("stop terrorists and pedophiles, think of the children" type stuff) that's when people should stop and think. Because anyone trying to convince you by shocking you probably wants to hide something in that shock.
This is why you want to set legal boundaries at the outset, rather than let the authorities push the envelope and allow bad practices to become the norm.
I think most people would be fine with using facial recognition technology in the instance of Amber Alerts or an escaped dangerous fugitive.
Likewise, I think most reasonable people would not feel comfortable in a situation where software is bulk scanning the faces of people entering a stadium and checking them for outstanding warrants for unpaid parking tickets etc.
A technologically sophisticated and ruthlessly efficient justice system in a progressive jurisdiction is far more oppressive than a backward, ineffective one under an authoritarian regime.
My local PD wants to do this. The need the City Commission to pass a resolution allowing them to sign this agreement.
I spoke with one of my commissioners who then had a meeting with the Police Department. The end result was the Police removed the item from the Commission meeting because they needed more time to prepare, justify and lay out policies in its use.
Had I not reached out this likely would have been rubber stamped.
Thus, I would at a minimum like to see a clear policy for when it can be used, image retention (if any), image sharing, etc.
This is what's important. There need to be clear legal expectations set about the security of this data. It has legitimate uses, but it should be a pain in the ass to access and share, like HIPAA. You have to really need it and make a case for needing it.
You must not own capital then. If you had they would have worked hard to help you. If you had been a big box store showing the footage to the police they would have helped. But as an individual, they are not there for you.
In a way, isn't this kind of what insurance against theft should be for? I.e. you can't eliminate all petty crime, as there's diminishing returns on efforts put into preventing it, but insurance can ensure the occasional victim can get compensation.
I've seen a mugging in Brazil. They are so frequent there that everyone I talked to has been mugged on the street. Friends of mine have been carjacked and kidnapped for 8 hours.
The police won't investigate a mugging. You're wasting your time to report it.
Let's not "accept" crimes that we think are not serious. All crime needs to be investigated and appropriate punishment given. Living in a society where some crime is OK is not a great place to live.
I don’t think that increasing police presence and surveillance is the answer to stopping petty crime, it just punishes people who need help and or creates a corrupt police dept, exasperating the problem further.
If we provide avenues for people to provide for themselves, crime will drop, and we avoid all of the chilling side effects of total surveillance.
I think that's a different statement than the one you said earlier.
Society should accept a little crime to be free != evaluate the true cost of crime.
So if there's a neighborhood that has the potential to sell Condos for $1.7M each but currently has houses that are $50k and we find out that the crime in the area is keeping that value so low - should we go after the thieves for stealing $1.7M from homeowners?
I'm just saying because that's one of the ways you add up the "true cost of crime".
People down voted you but completely ignore the fact that California basically tried just that and still has it on its books and the results were so predictable [1]
proposition 47 made anything below $950 keeps the crime a misdemeanor which likely means the thieves face no pursuit and no punishment. this led to a spike in car break ins as well. You can do this daily!
the simple matter he is that those voting on such changes to the law are not going to be troubled by the uptick as where they tread they are likely never to be victims of their policy
So let's add a face recognition database. Now, two possibilities.
One is they've got the DMV database or some other mass database with photos of millions of people. They run your video through and it comes back with 117 matches. They all look like your guy, because that's what facial recognition is, so which one is it? Still no way to tell. And that's assuming the perpetrator was actually in the database.
The other possibility is that they're only using something like mugshot photos. Now it's only thousands of people in the database, so they only get one match. Hurray, we've caught him! Except that it could still have been any of the other 116 people who weren't in the smaller database. Or any of thousands of people from out of state or other countries who aren't even in the bigger database. So now we're going to go convict this guy because he's the only suspect and we've got a video of somebody who looks like him, even though he probably wasn't even the perpetrator.
Notice that this is only a problem for justice systems that want to be justice systems. If all you want is a pound of flesh from some random schmoe to demonstrate that you're catching bad guys even though you're not, it works great. Same if you're an authoritarian dictatorship, because then you can murder all 117 matches and to heck with all this hard work of making sure we found the right person. Which is why mass surveillance does more harm than good.
Or, more properly, they use the 117 matches as a starting point in an investigation to eliminate most of them, i.e., they weren't in the area at the time, and identify the others who were in the area at the time. Probably not worth it for a petty crime, but for something more serious...
Or, they just store the info, and when the same petty criminal pops up again with a similar crime, as is often the case, also track the particulars of this one to see if they get a match and add that to the prosecution list.
Kind of like cell phone tracking. There's hundreds of phones in the vicinity of any one bank robbery, but when you get one that is in the vicinity of all three, probably a good idea to investigate the owner of that one...
>Or, more properly, they use the 117 matches as a starting point in an investigation to eliminate most of them, i
Maybe if you're a local politician, a cop or rich enough for them to know who you are you'll get that kind of service. For everyone else they'll just do nothing or nearly nothing and go back to playing with that APC they don't need.
If they find your car as part of some other crime they'll add another page to the book they're throwing but you don't need a database to do that.
Exactly! For most minor crimes like car break in, unless it's part of a string, they'll be unlikely to assign resources to the investigation (unless the victim is a VIP), but for a more serious crime, there's some good clues for the investigation.
> Or, more properly, they use the 117 matches as a starting point in an investigation to eliminate most of them, i.e., they weren't in the area at the time, and identify the others who were in the area at the time.
Which is how you get false convictions, because you're starting with a list of essentially random people, with no guarantee that the true perpetrator is even on it.
Then you ping cell phone records, and the actual perpetrator left their phone at home because they knew they'd be committing a crime, but one person out of a hundred on your list of random people happened to be in the area at that time, and that combined with what looks like their face on video is enough to convince a jury. Or convince the accused that a jury would be convinced so they'll plead guilty to a crime they didn't commit.
> Or, they just store the info, and when the same petty criminal pops up again with a similar crime, as is often the case, also track the particulars of this one to see if they get a match and add that to the prosecution list.
So you get another video with the same face matching the same hundred odd people, which increases the chances that one of the other random people on the list was in the vicinity of at least one of the crimes while the actual perpetrator always leaves their phone at home, and now the innocent party gets convicted of multiple counts.
> Kind of like cell phone tracking. There's hundreds of phones in the vicinity of any one bank robbery, but when you get one that is in the vicinity of all three, probably a good idea to investigate the owner of that one...
Then it turns out to be the alarm tech or insurance adjuster or regional bank manager responding to the robbery, or some company that always makes deposits on Friday around 3PM which is also when the robbery crew always robs the banks. Which makes it a lot easier to choose any of those parties as a scapegoat if you can't find who really did it.
Computers are really good at finding random coincidences in large data sets. Humans are really bad at not interpreting that kind of noise as signal.
How likely is it actually that out of the set of 116 people who look like you from your your state (which I see as basically a random sample of 116 people, unless you have a tendency to gravitate people who look like yourself toward yourself...), one of them was in the same area as you, at the same time that a crime was committed against you there?
The real question is how many of those random people could conclusively clear themselves with an alibi. If 20% of those people don't remember or can't conclusively prove where they were... You now have a ton of suspects that each are just as likely the criminal and obviously at most one of them could have committed the crime.
Yes if you can narrow it down to 100 people and then further confirm their identity with DNA or other real evidence that is potentially useful (and only potentially... It may waste your time). But for any sort of petty crime it seems like it will almost always do more harm than good.
> The real question is how many of those random people could conclusively clear themselves with an alibi
No, that's not the real question. The real question is how frequently such people would even exist in the first place, which is what I was asking. We're not trying to find solutions to non-problems here.
Now likely, well based on the number of false convictions we seem to get in the US legal system, I can tell you that the probability is non-zero.
Especially since there may be a genetic component of people looking the same. Chasing down all the Smiths in your neighborhood can increase the rates of false detection.
> How likely is it actually that out of the set of 116 people who look like you from your your state (which I see as basically a random sample of 116 people, unless you have a tendency to gravitate people who look like yourself toward yourself...), one of them was in the same area as you, at the same time that a crime was committed against you there?
That's going to depend a lot on how stringently you define "in the area" and "the time of the crime" as well as things like population density and demographics. For example, if the perpetrator is a particular race and so consequently are all the facial recognition matches, and the state's population has particular races concentrated in particular areas, expect that to affect the probability in a way that leads to more false positives all around, because it's then effectively the case that people with similar faces tend to live near each other.
But the really perfidious thing about all this is that it's not just facial recognition and cell phone matches. Suppose the chance of a random person out of a hundred with a similar face being in the same area is "only" 5%. That would be the case if there were around 500 people in the vicinity (not unusual in an urban area) in a state with a million people, and then around 5% of your convicted criminals would be innocent.
But we're not just checking faces and cell phones, we also check all kinds of things like internet history, family and work connections, financial and purchasing history, etc.
Then we come to find out that one of the facial recognition matches did a web search for Halloween masks in October of last year and the co-conspirators whose faces we don't have were wearing masks, and the same person has a low paying job, a fair amount of debt and a second cousin with a twenty year old conviction for marijuana possession, and the cousin's cell phone was in the vicinity of the crime.
It's a pile of circumstantial nonsense. It's all selection bias and dependent variables. Everybody searches for Halloween masks around Halloween. Each of your hundred odd facial recognition matches has a hundred odd family members, there were a few hundred people in the vicinity, you start multiplying numbers like that and you get coincidence matches. The random cousin who hangs out in the same areas as crimes are committed is not unlikely to have a checkered past.
But all the jury sees is a huge pile of evidence. Which gets bigger the more mass surveillance there is, independent of guilt or innocence, because more surveillance can uncover more coincidences. (And then they pick up your cousin and find drugs on him again, and suddenly he's testifying against you for a lesser sentence even though you haven't even seen him in five years.)
To me going around with some photos of the guy and looking for him, maybe asking folks around the neighborhood if they know him, searching through mugshots of known criminals in the area for a possible match, maybe posting his picture on their website or on posters seems like exactly the sort of thing police should be doing. Good old fashioned honest police work. I know they have quotas to meet and they could bring in more revenue by ignoring it and just ticketing people for minor shit like traffic offenses but it just seems lazy to say that because a computer can't do all the work for them they have no way to ID someone. Frankly, just by having video you've already done part of their job for them.
I have a hard time getting too upset about this as described in the article. Sharing police fingerprints and mug shots doesn’t sound too bad.
There are of course “but”s:
- should mug shots be retained for people later freed/not convicted? Certainly. It he case today for photos, prints, and (in California) DNA.
- expanding into the DL database, as mentioned in the article seems like a dangerous scope creep — though what if they are investigating a driving offense like a hit and run?
Anyway, a useful article. But not necessarily a bad thing, for a change.
Facial recognition technology has all sorts of situations where one can imagine where it would be useful. Unfortunately all of the ways you could use it to suppress, discriminate, silence, and intimidate people outweigh any upsides by miles. Facial recognition tech should be outlawed, unequivocally, plain and simple. I’m not holding my breath that will happen.
Good. This means police will be more efficient in identifying and catching suspects. Facial recognition is always used as a first-pass filter, so the false positive rate is not relevant - it simply narrows the search space so that limited police resources are used efficiently.
Why does it need to be proven before it is implemented? Did you ask for proof that use of Microsoft Office or electricity by police departments would make them more efficient?
It seems inherently obvious and frankly, common sense, that a network of cameras and facial recognition software are going to be more effective at identifying perps in the wild than a handful of officers patrolling the streets, needing to remember the faces of everyone they're on the lookout for, and scanning their surroundings constantly for matches.
> I don't understand. Are you saying that using a filter negates a false positive? I don't think that's accurate at all.
The concern often raised is that trusting an imprecise algorithm is risky. But in reality matches from facial recognition software are reviewed by humans as a second-pass filter prior to taking action (like dispatching officers). We already trust humans to perform policing inherently by the act of having police forces exist, so there is no elevated risk introduced by use of facial recognition that isn't proportionally offset by a higher degree of true positive matches, since any elevated false positive match raised by the algorithm is contained by the humans reviewing potential matches first.
> What about the rights of someone who's been falsely implicated?
They still have rights, and can fight those claims. Why would you expect this to be any different due to the use of facial recognition? If the suspect looks like whoever police are seeking, then they would be exposed to the risk of implication orthogonal to the existence of facial recognition. And there are ways (even if not perfect) to fight those claims. We retain policing as a necessary part of society despite its imperfections because it is necessary and hugely better than the alternative of not having police forces. And all this holds true even with police departments adopting technology - whether that is electricity, Microsoft Office, or facial recognition.
> Why does it need to be proven before it is implemented?
We're talking about peoples' safety and justice. It's not just some product that people can return if they don't like it.
> Did you ask for proof that use of Microsoft Office or electricity by police departments would make them more efficient?
I didn't have that opportunity. But yes, I would have.
> It seems inherently obvious and frankly, common sense, that a network of cameras and facial recognition software are going to be more effective at identifying perps in the wild
I don't think it's obvious. Indeed, I think the potential for abuse is very high. It would be easy for someone to say "the algorithm said so" to someone who's been accused. How would someone without specific knowledge of the algorithm be able to dispute that?
> But in reality matches from facial recognition software are reviewed by humans as a second-pass filter prior to taking action (like dispatching officers). We already trust humans to perform policing inherently by the act of having police forces exist, so there is no elevated risk introduced by use of facial recognition
You are wrong.
Matches from facial recognition software might be reviewed by humans. But then again, payroll is reviewed by humans too and that has tons of problems. If we can't fix our own money tracking systems, then how exactly do you think we can keep our facial recognition software correctly working?
To claim that "there is no _elevated_ risk introduced by use of facial recognition" makes me wonder whether you've thought about how it could even unintentionally cause problems. That's a very dangerous line of thought to ignore, by the way.
> They still have rights, and can fight those claims.
Bullshit. Not just bullshit, but fucking bullshit. People who are accused today don't have the knowledge and money necessary to fight claims that don't even involve magic hidden behind the guise of an algorithm. When [1] officers [2] plant [3] evidence [4] and prosecutors [5] go [6] for [7] plea deals, the public suffers. It's especially perverse when the people who are supposed to be a part of the justice system argue that "they're guilty of something". Innocent peoples lives can be ruined when they don't have the ability to demonstrate their innocence [8]. I'm going to stop adding links because these ones were found with literally just thirty seconds of searching. It's that common of a problem.
I could see you then arguing that the goalpost is moving: solve the officer problem or solve the prosecutorial problem. Of course you're right. But you'd also miss the point: adding more abilities for "law enforcement" to put people behind bars is not what the world needs today.
You support adding a new, ever more complex, way for people to be caught up by the law for reasons completely foreign to them. Your love for computers and algorithms and AI, and desire to see them in action, blinds you to the real world. I challenge you to meet people, innocent and not, who've been affected by law enforcement. I challenge you to really think hard about what you want from facial recognition.
The cat's out of the bag with this technology - it exists, citizens will demand that the police use it to help solve crimes. The way forward is to ensure that it gets used as an investigative tool, rather than part of mass surveillance.
The way to do that is to make sure that querying a facial recognition database is too expensive for ubiquitous use. A court order per face you want to identify is likely enough, though I'd like an additional $100-$1000 fee to discourage rubber-stamping.
IIRC, the police have hired people who are just really good at recognizing faces and given them a giant stack of photographs of criminals. Driving down the difficulty of that sort of process - identifying suspects with known faces - seems actually reasonable. Driving down the cost of identifying faces to the point where you can have a historical database of identity-annotated sidewalk footage of an entire city is the sort of thing we need to fight.
>IIRC, the police have hired people who are just really good at recognizing faces and given them a giant stack of photographs of criminals. Driving down the difficulty of that sort of process - identifying suspects with known faces
It's believed in some circles this at least sometimes just BS and is a cover for parallel construction.
> Today, you are not you, you are your data, a persona. And you are somehow responsible for it or anything that casts a similar shadow.
I think it is a similar case here. The main problem won't be that my face is out there. The problem will be that the police will stop you or stop by at your house to "just have a chat" because some algorithm matched your face to some input somewhere with a 90% confidence.
By the way, who owns the photographs and how do the departments have authority to share them with some company? If they can share with with that company, why not make the data public and let everyone have access to it? I'd like to play with the data as well.
Do you think there's more chance of this happening, versus Barbara down the shops being shown a photograph and saying "oh yes, that looks just like mcny from two doors down"?
I'm not mcny, but yes, of course. It'll be massively cheaper to just yank the confidence slider down until results pop out, vs. sending meatbags walking around an area, showing photographs to neighbors.
SFPD don't even enforce homeless laws properly and let people live in same street in tents for years. However facial recognition wont fix any of this, and just make things worse.
Question for people that are anti facial recognition. What methods of identifying people are you ok with? Are you ok with the police/news asking other people to help identify someone? To me that sounds like facial recognition with extra steps?
The arguments against facial recognition like that there can be false positives, or that can affect some groups more than others, doesn't that also apply when people are identifying people? If so isn't the real solution to require more evidence than just a facial match, not to ban an effective way of narrowing a suspect pool. That way police can spend less time manually identifying people and more time getting other evidence.
I’m concerned because probability is not intuitive.
Suppose a store is robbed, and there’s a video.
The police identity some suspects - the guy who just got out of jail for robbing the same store, and another person the store owner had a dispute with. Neither of them look like the robber in the video. Then the police take a still from the video and knock on some doors around the block. Somebody recognizes the person in the video, and the police investigate that person. This scenario seems pretty fair to me.
Now suppose the police run it through the facial recognition system. It identifies one person as a 99% match, and the police go investigate this person. This scenario does not seem so fair to me.
Here’s how I see the math:
P(A) = P(robber has a doppelgänger living on the same block) = .01
P(B) = P(robber had a doppelgänger somewhere in the database) = .9
P(X) = P(police screw up investigation, and will convict the suspect whether or not they are guilty) = .2
P(AX) = .002
P(BX) = .18
The exact numbers are made up, but as long as P(A) << P(B), you can see you this tech will result in a huge increase in false convictions. Even if P(X) is low, the number of false convictions increases by P(B)/P(A).
I'd argue the %s are not intuitive either way. In fact, if P(A) does happen, and there really is an unlucky doppelganger, that person is very likely to be charged. That could have been avoided with technology produced gave 5 other suspects that don't live on that block. Should it therefor be allowed for criminal defense if not prosecution?
The issue with the P(A) and P(B) argument is that police already use databases heavily, and most people don't have any problem with it. But why when it comes to facial recognition, is it too dangerous to use technology to drive efficiency.
If they're looking for somebody named Jane Doe, anybody with that name shows up on a list and police investigate. Of course if there are Jane Does in a 2 mile radius, they start with those. So why not just say if the system delivers a match within x accuracy and the person is within y residents (plus a variety of other variables), and x/y is below a threshold, then the match can be presented to police for further investigation.
Searching databases for matches is fine for names, or fingerprints, shoe prints, tire track, fiber analysis - but not faces? I personally wonder if it's really any different, or if its just better tailored for the media outrage machine because "China does it", or because "facial recognition targets minorities".
I think police searching databases for low-quality evidence like tire track, shoe prints, and fiber analysis is a very dodgy practice, for these exact reasons. The key is the specificity of the match, and I don’t think facial recognition is good enough. Fingerprints and DNA can be, but there are still known cases of people being falsely charged based on databases searches with a partial match.
This tech can be good if applied to a narrow range of people like you suggest (eg. only searching people who live in neighboring blocks) but nobody is actually doing that. We should pass laws requiring a rigorous analysis of these probabilities for such databases to be used, including a conversation about what rate of false positives we are willing to tolerate. Guardrails should be put in place to enforce those limits. If this is too hard, we don’t have a strong enough handle on this technology to be using it.
Here’s the scenario that scares me the most:
Police identify a suspect using facial recognition. Then puts that person in a lineup for a witness. Of course the witness is going to say “that’s the one!” because the suspect actually looks like the perpetrator. The witness will be sure, the cops will be sure, and a jury will convict. And this scenario is completely determined by the use of the facial recognition database. This will happen unless we pass laws to prevent it.
There's some accountability in the human-based approach.
A person understands the context in which they are naming another person based on their face and can weigh internally whether they are sure enough to tell the police. The algorithm just spits out a confidence for the Police/DA to do whatever they want with.
A person can be cross-examined in court. Is that little old lady a blind racist who is accusing a person of color? Her credibility can be attacked in court. The algorithm was trained by racists? That's a LOT harder to attack.
I don't think the police should be barred from using the tech, but I certainly don't think it should be admissible in court.
The danger I see in it, is that everybody in the database is automatically considered a suspect for any crime in which the database is used.
If the police have to ask people to perform identification, then they have to follow some sort of investigative procedure to arrive at who they ask to perform identification, and who they present to be identified. Of course this has its own problems, but facial recognition databases really ramp those problems up.
Then if you get falsely identified, you’re gonna have to prove it. How much time do you spend every day, not committing crimes, but without an alibi to prove it?
I came to a realization on this subject while visiting Las Vegas: Pragmatically, a maximally high trust society and a maximally low trust society permit the same behavior patterns. In Las Vegas casinos, which are low trust and have high surveillance, one can comfortably leave tens of thousands of dollars or more on a gaming table while one goes to use the bathroom or whatever. In a hypothetical high trust low surveillance environment one could do the same.
The difference is, we know how to engineer for low trust and high surveillance. We don't really know how to engineer for high trust and low surveillance. So our practical choices are to either live in a low trust low surveillance society and having to constrain ourselves as that requires, or to live in a low trust high surveillance society that simulates a much more pleasant high trust society.
> live in a low trust high surveillance society that simulates a much more pleasant high trust society
And in a high surveillance society, whatever the initial conditions, the cost/benefit ratio of cooperation vs. defection is biased towards trust. In other words, the ersatz high trust society would become the real thing, of course only provided the surveillance is trustworthy. For example, China's social credit system might work if it is administered sans corruption. It has to be self-reflexive.
As incredible as it seems, Mrs. Grundy will save us.
These are essentially 'super-technologies', much like government-requested back doors and master keys. It's not an overstatement to say that the potential they hold is on par with human genetic engineering - something the civilized world has thus far agreed is a road best avoided. Sure, the ethical dilemma of them is enough to give pause - but are the following not much greater risks?
* These systems will crowdsource "evidence", and by proxy, accountability - what happens when a mistake is made?
* Increased centralization will eventually enable 'single point of failure' scenarios in systems of a national/international scale.
* What happens when someone misuses or gains unauthorized access to such systems? Will the damage be reversible?
Such technologies will be deployed before accountability for them is defined, leading to a temporary absence thereof. Systems of the past (digital or otherwise) were inherently tolerant to misuse, malfunctions & misfires because only so much damage could be done before the errant behavior was discovered & handled. Today, how many seconds would it take to plunge a country into chaos?
113 comments
[ 3.0 ms ] story [ 156 ms ] threadHow many innocent people are caught with fingerprint data?
Also the William & Will West case was an older example of misidentifying by photography. Modern tech has improved dramatically and will get more and more precise. Why apply early 2000’s tech issues to the advancements made in the past 15-20 years?
https://news.sky.com/story/met-polices-facial-recognition-te...
A threat model exists for a reason. A risk reduction model exists for a reason. Using a tool (computer/ML) to do 95% of the pattern matching is called one thing: not being a dumbass.
I sincerely wonder if any of these super-hostile anti-facial ID activists have truly had someone close to them impacted by a crime, and had said crime go unsolved. Or are they merely idealistic loafers, clutching their pearls, to signal how "woke" they are?
Forensic “matching” of bullets to guns is basically pseudoscience, but continues to be used in court to convict people.
Roadside drug analysis kits are less useful than dowsing rods, but get people arrested on a daily basis. People that can’t afford bail, and plead guilty just to get back to their lives, despite the fact that an actual defense attorney would have had the case tossed in the trash.
Previous experience suggests that police tech does not just “get better.” They can’t even progress past “disproven pseudoscience.” And the innocent people that have their lives messed with while waiting for the tech to catch up might not be indifferent to the process.
It can, at best, match ammo from the same machine-manufactured batch to a factory-stock model of gun. It automatically fails on hand-load cartridges, or ammo that the analyst cannot readily source.
When there is a match, the prosecutor might say "this bullet was fired from this gun", but the science says "a bullet similar to the one in evidence was marked and deformed in a similar fashion, when fired by the weapon in evidence (just as it would with every other factory-stock copy of this model of gun)".
It's like matching a nail to the hammer that drove it. If the test nail is already different from the reference nail, you cannot match it. If you have two or more identical hammers, you can't tell which one drove the nail. The hammer factory made thousands of identical hammers.
In most cases, the tests are likely saying "the most commonly sold type of ammunition was fired from the most popular model of gun of that caliber". It's the sort of thing that should probably be used exclusively as exculpatory evidence. I.e. this model of gun always leaves distinctive markings, such as from an off-center diamond-shaped firing pin, or from an uncommon rifling twist, that were not found on the bullets in evidence, so the gun in evidence could not have fired them. Otherwise, a gun substantially similar, with all the same machine-made parts, could leave the same marks, so you could never know for sure which one fired the bullets.
The only time it would be useful to convict is if someone used a unique, hand-made, gunsmithed gun to fire factory-made ammunition. And the number of (violent) crimes committed with that type of display-piece gun approaches zero, because they tend to be both expensive and less fit for use as a weapon than cheaper factory-built guns.
All this tech makes a turnkey police state right there waiting. Who is going to turn that key before we dismantle it? Think of the children.
In the past, if a license plate was noted, you’d have to manually go through records. Now through automation and scale it’s instant. Should that be prohibited too?
http://www.nbcnews.com/id/27233798/ns/us_news-crime_and_cour...
> Commission President Anthony Pacheco said Friday that he was highly concerned after learning police have arrested at least two innocent people because of faulty fingerprint analysis.
How many guilty people were convicted based on true positives?
Seems you're willing to punish an innocent man and a guilty man just to make sure a committed crime is followed up with a conviction.
Why have such a low bar in 2019?
[1] http://www8.nationalacademies.org/onpinews/newsitem.aspx?rec...
[2] https://www.nap.edu/catalog/12589/strengthening-forensic-sci...
>"There is some evidence that fingerprints are unique to each person, and it is plausible that careful analysis could accurately discern whether two prints have a common source, the report says. However, claims that these analyses have zero-error rates are not plausible; uniqueness does not guarantee that two individuals' prints are always sufficiently different that they could not be confused, for example. Studies should accumulate data on how much a person's fingerprints vary from impression to impression, as well as the degree to which fingerprints vary across a population. With this kind of research, examiners could begin to attach confidence limits to conclusions about whether a print is linked to a particular person."
Quite the stretch to go from "non-zero-error rates" to "made up pseudoscience" isn't it?
It's a problem if people are being convicted on fingerprint analysis alone. But as a piece of the puzzle, when combined with other evidence, what is the issue?
That's an odd criticism since nothing has a zero error rate.
DNA analysis has a non-zero error rate and we routinely rely on it.
When you say "criminals" you are probably thinking of strictly people who harm others, but the state is thinking of anybody who breaks its laws. Currently many of those laws exist solely to keep certain groups in power, and there is no guarantee it won't get even worse in the future.
Human law enforcement ensures layers of decisionmakers who are at least theoretically capable of empathy, and limited manpower makes them prioritize the worst or most flagrant crimes. Automated law enforcement gives a smaller group of people horrifyingly granular levels of control over all of society.
Also, non sequitur, have there been any proposals for how to detect when your country has become an authoritarian dystopia?
Cats out of the bag with facial recog. Might as well have strong legal documentation and judges on the line with this.
And when the expansion is justified by shocking statements ("stop terrorists and pedophiles, think of the children" type stuff) that's when people should stop and think. Because anyone trying to convince you by shocking you probably wants to hide something in that shock.
I think most people would be fine with using facial recognition technology in the instance of Amber Alerts or an escaped dangerous fugitive.
Likewise, I think most reasonable people would not feel comfortable in a situation where software is bulk scanning the faces of people entering a stadium and checking them for outstanding warrants for unpaid parking tickets etc.
A technologically sophisticated and ruthlessly efficient justice system in a progressive jurisdiction is far more oppressive than a backward, ineffective one under an authoritarian regime.
I spoke with one of my commissioners who then had a meeting with the Police Department. The end result was the Police removed the item from the Commission meeting because they needed more time to prepare, justify and lay out policies in its use.
Had I not reached out this likely would have been rubber stamped.
Thus, I would at a minimum like to see a clear policy for when it can be used, image retention (if any), image sharing, etc.
This is what's important. There need to be clear legal expectations set about the security of this data. It has legitimate uses, but it should be a pain in the ass to access and share, like HIPAA. You have to really need it and make a case for needing it.
Me to cop: That's him you can see him breaking into the car here at 10:15, then at 12:05 we have a perfect clear view of his face.
Cop: We can't really do anything with this, I mean we can take the copy of the video but we don't have any way to identify him.
Me: So we can't really do anything?
Cop: Yep sorry
I would not want to be in your situation, but the needs of the many outweighs the needs of the few.
I've seen a mugging in Brazil. They are so frequent there that everyone I talked to has been mugged on the street. Friends of mine have been carjacked and kidnapped for 8 hours.
The police won't investigate a mugging. You're wasting your time to report it.
Let's not "accept" crimes that we think are not serious. All crime needs to be investigated and appropriate punishment given. Living in a society where some crime is OK is not a great place to live.
If we provide avenues for people to provide for themselves, crime will drop, and we avoid all of the chilling side effects of total surveillance.
And I say over and over, it is time to evaluate the true cost of crime.
Society should accept a little crime to be free != evaluate the true cost of crime.
So if there's a neighborhood that has the potential to sell Condos for $1.7M each but currently has houses that are $50k and we find out that the crime in the area is keeping that value so low - should we go after the thieves for stealing $1.7M from homeowners?
I'm just saying because that's one of the ways you add up the "true cost of crime".
I would rather someone speed through my neighborhood than have 3 police cars make a spectacle and spend thousands of dollars.
Speeding is a problem fixed by a proper education (in theory) It is not fixed by chasing someone down and giving them a speeding ticket.
proposition 47 made anything below $950 keeps the crime a misdemeanor which likely means the thieves face no pursuit and no punishment. this led to a spike in car break ins as well. You can do this daily!
the simple matter he is that those voting on such changes to the law are not going to be troubled by the uptick as where they tread they are likely never to be victims of their policy
[1] https://californiaglobe.com/section-2/while-cities-homeless-...
>As crime dries up, Japan’s police hunt for things to do
>https://news.ycombinator.com/item?id=20570763
> Cop: Yep sorry
So let's add a face recognition database. Now, two possibilities.
One is they've got the DMV database or some other mass database with photos of millions of people. They run your video through and it comes back with 117 matches. They all look like your guy, because that's what facial recognition is, so which one is it? Still no way to tell. And that's assuming the perpetrator was actually in the database.
The other possibility is that they're only using something like mugshot photos. Now it's only thousands of people in the database, so they only get one match. Hurray, we've caught him! Except that it could still have been any of the other 116 people who weren't in the smaller database. Or any of thousands of people from out of state or other countries who aren't even in the bigger database. So now we're going to go convict this guy because he's the only suspect and we've got a video of somebody who looks like him, even though he probably wasn't even the perpetrator.
Notice that this is only a problem for justice systems that want to be justice systems. If all you want is a pound of flesh from some random schmoe to demonstrate that you're catching bad guys even though you're not, it works great. Same if you're an authoritarian dictatorship, because then you can murder all 117 matches and to heck with all this hard work of making sure we found the right person. Which is why mass surveillance does more harm than good.
Or, they just store the info, and when the same petty criminal pops up again with a similar crime, as is often the case, also track the particulars of this one to see if they get a match and add that to the prosecution list.
Kind of like cell phone tracking. There's hundreds of phones in the vicinity of any one bank robbery, but when you get one that is in the vicinity of all three, probably a good idea to investigate the owner of that one...
Maybe if you're a local politician, a cop or rich enough for them to know who you are you'll get that kind of service. For everyone else they'll just do nothing or nearly nothing and go back to playing with that APC they don't need.
If they find your car as part of some other crime they'll add another page to the book they're throwing but you don't need a database to do that.
Which is how you get false convictions, because you're starting with a list of essentially random people, with no guarantee that the true perpetrator is even on it.
Then you ping cell phone records, and the actual perpetrator left their phone at home because they knew they'd be committing a crime, but one person out of a hundred on your list of random people happened to be in the area at that time, and that combined with what looks like their face on video is enough to convince a jury. Or convince the accused that a jury would be convinced so they'll plead guilty to a crime they didn't commit.
> Or, they just store the info, and when the same petty criminal pops up again with a similar crime, as is often the case, also track the particulars of this one to see if they get a match and add that to the prosecution list.
So you get another video with the same face matching the same hundred odd people, which increases the chances that one of the other random people on the list was in the vicinity of at least one of the crimes while the actual perpetrator always leaves their phone at home, and now the innocent party gets convicted of multiple counts.
> Kind of like cell phone tracking. There's hundreds of phones in the vicinity of any one bank robbery, but when you get one that is in the vicinity of all three, probably a good idea to investigate the owner of that one...
Then it turns out to be the alarm tech or insurance adjuster or regional bank manager responding to the robbery, or some company that always makes deposits on Friday around 3PM which is also when the robbery crew always robs the banks. Which makes it a lot easier to choose any of those parties as a scapegoat if you can't find who really did it.
Computers are really good at finding random coincidences in large data sets. Humans are really bad at not interpreting that kind of noise as signal.
Yes if you can narrow it down to 100 people and then further confirm their identity with DNA or other real evidence that is potentially useful (and only potentially... It may waste your time). But for any sort of petty crime it seems like it will almost always do more harm than good.
No, that's not the real question. The real question is how frequently such people would even exist in the first place, which is what I was asking. We're not trying to find solutions to non-problems here.
Especially since there may be a genetic component of people looking the same. Chasing down all the Smiths in your neighborhood can increase the rates of false detection.
That's going to depend a lot on how stringently you define "in the area" and "the time of the crime" as well as things like population density and demographics. For example, if the perpetrator is a particular race and so consequently are all the facial recognition matches, and the state's population has particular races concentrated in particular areas, expect that to affect the probability in a way that leads to more false positives all around, because it's then effectively the case that people with similar faces tend to live near each other.
But the really perfidious thing about all this is that it's not just facial recognition and cell phone matches. Suppose the chance of a random person out of a hundred with a similar face being in the same area is "only" 5%. That would be the case if there were around 500 people in the vicinity (not unusual in an urban area) in a state with a million people, and then around 5% of your convicted criminals would be innocent.
But we're not just checking faces and cell phones, we also check all kinds of things like internet history, family and work connections, financial and purchasing history, etc.
Then we come to find out that one of the facial recognition matches did a web search for Halloween masks in October of last year and the co-conspirators whose faces we don't have were wearing masks, and the same person has a low paying job, a fair amount of debt and a second cousin with a twenty year old conviction for marijuana possession, and the cousin's cell phone was in the vicinity of the crime.
It's a pile of circumstantial nonsense. It's all selection bias and dependent variables. Everybody searches for Halloween masks around Halloween. Each of your hundred odd facial recognition matches has a hundred odd family members, there were a few hundred people in the vicinity, you start multiplying numbers like that and you get coincidence matches. The random cousin who hangs out in the same areas as crimes are committed is not unlikely to have a checkered past.
But all the jury sees is a huge pile of evidence. Which gets bigger the more mass surveillance there is, independent of guilt or innocence, because more surveillance can uncover more coincidences. (And then they pick up your cousin and find drugs on him again, and suddenly he's testifying against you for a lesser sentence even though you haven't even seen him in five years.)
There are of course “but”s: - should mug shots be retained for people later freed/not convicted? Certainly. It he case today for photos, prints, and (in California) DNA.
- expanding into the DL database, as mentioned in the article seems like a dangerous scope creep — though what if they are investigating a driving offense like a hit and run?
Anyway, a useful article. But not necessarily a bad thing, for a change.
That's quite a claim. Prove it.
> Facial recognition is always used as a first-pass filter, so the false positive rate is not relevant
I don't understand. Are you saying that using a filter negates a false positive? I don't think that's accurate at all.
> it simply narrows the search space so that limited police resources are used efficiently.
What about the rights of someone who's been falsely implicated?
Why does it need to be proven before it is implemented? Did you ask for proof that use of Microsoft Office or electricity by police departments would make them more efficient?
It seems inherently obvious and frankly, common sense, that a network of cameras and facial recognition software are going to be more effective at identifying perps in the wild than a handful of officers patrolling the streets, needing to remember the faces of everyone they're on the lookout for, and scanning their surroundings constantly for matches.
> I don't understand. Are you saying that using a filter negates a false positive? I don't think that's accurate at all.
The concern often raised is that trusting an imprecise algorithm is risky. But in reality matches from facial recognition software are reviewed by humans as a second-pass filter prior to taking action (like dispatching officers). We already trust humans to perform policing inherently by the act of having police forces exist, so there is no elevated risk introduced by use of facial recognition that isn't proportionally offset by a higher degree of true positive matches, since any elevated false positive match raised by the algorithm is contained by the humans reviewing potential matches first.
> What about the rights of someone who's been falsely implicated?
They still have rights, and can fight those claims. Why would you expect this to be any different due to the use of facial recognition? If the suspect looks like whoever police are seeking, then they would be exposed to the risk of implication orthogonal to the existence of facial recognition. And there are ways (even if not perfect) to fight those claims. We retain policing as a necessary part of society despite its imperfections because it is necessary and hugely better than the alternative of not having police forces. And all this holds true even with police departments adopting technology - whether that is electricity, Microsoft Office, or facial recognition.
We're talking about peoples' safety and justice. It's not just some product that people can return if they don't like it.
> Did you ask for proof that use of Microsoft Office or electricity by police departments would make them more efficient?
I didn't have that opportunity. But yes, I would have.
> It seems inherently obvious and frankly, common sense, that a network of cameras and facial recognition software are going to be more effective at identifying perps in the wild
I don't think it's obvious. Indeed, I think the potential for abuse is very high. It would be easy for someone to say "the algorithm said so" to someone who's been accused. How would someone without specific knowledge of the algorithm be able to dispute that?
> But in reality matches from facial recognition software are reviewed by humans as a second-pass filter prior to taking action (like dispatching officers). We already trust humans to perform policing inherently by the act of having police forces exist, so there is no elevated risk introduced by use of facial recognition
You are wrong.
Matches from facial recognition software might be reviewed by humans. But then again, payroll is reviewed by humans too and that has tons of problems. If we can't fix our own money tracking systems, then how exactly do you think we can keep our facial recognition software correctly working?
To claim that "there is no _elevated_ risk introduced by use of facial recognition" makes me wonder whether you've thought about how it could even unintentionally cause problems. That's a very dangerous line of thought to ignore, by the way.
> They still have rights, and can fight those claims.
Bullshit. Not just bullshit, but fucking bullshit. People who are accused today don't have the knowledge and money necessary to fight claims that don't even involve magic hidden behind the guise of an algorithm. When [1] officers [2] plant [3] evidence [4] and prosecutors [5] go [6] for [7] plea deals, the public suffers. It's especially perverse when the people who are supposed to be a part of the justice system argue that "they're guilty of something". Innocent peoples lives can be ruined when they don't have the ability to demonstrate their innocence [8]. I'm going to stop adding links because these ones were found with literally just thirty seconds of searching. It's that common of a problem.
I could see you then arguing that the goalpost is moving: solve the officer problem or solve the prosecutorial problem. Of course you're right. But you'd also miss the point: adding more abilities for "law enforcement" to put people behind bars is not what the world needs today.
You support adding a new, ever more complex, way for people to be caught up by the law for reasons completely foreign to them. Your love for computers and algorithms and AI, and desire to see them in action, blinds you to the real world. I challenge you to meet people, innocent and not, who've been affected by law enforcement. I challenge you to really think hard about what you want from facial recognition.
[1]: https://www.nbcnews.com/news/us-news/st-louis-officer-execut...
[2]: https://www.cbsnews.com/news/body-cam-video-baltimore-police...
[3]: https://news.yahoo.com/former-florida-offi...
The way to do that is to make sure that querying a facial recognition database is too expensive for ubiquitous use. A court order per face you want to identify is likely enough, though I'd like an additional $100-$1000 fee to discourage rubber-stamping.
IIRC, the police have hired people who are just really good at recognizing faces and given them a giant stack of photographs of criminals. Driving down the difficulty of that sort of process - identifying suspects with known faces - seems actually reasonable. Driving down the cost of identifying faces to the point where you can have a historical database of identity-annotated sidewalk footage of an entire city is the sort of thing we need to fight.
It's believed in some circles this at least sometimes just BS and is a cover for parallel construction.
https://news.ycombinator.com/item?id=20563430
> Today, you are not you, you are your data, a persona. And you are somehow responsible for it or anything that casts a similar shadow.
I think it is a similar case here. The main problem won't be that my face is out there. The problem will be that the police will stop you or stop by at your house to "just have a chat" because some algorithm matched your face to some input somewhere with a 90% confidence.
By the way, who owns the photographs and how do the departments have authority to share them with some company? If they can share with with that company, why not make the data public and let everyone have access to it? I'd like to play with the data as well.
The arguments against facial recognition like that there can be false positives, or that can affect some groups more than others, doesn't that also apply when people are identifying people? If so isn't the real solution to require more evidence than just a facial match, not to ban an effective way of narrowing a suspect pool. That way police can spend less time manually identifying people and more time getting other evidence.
Suppose a store is robbed, and there’s a video.
The police identity some suspects - the guy who just got out of jail for robbing the same store, and another person the store owner had a dispute with. Neither of them look like the robber in the video. Then the police take a still from the video and knock on some doors around the block. Somebody recognizes the person in the video, and the police investigate that person. This scenario seems pretty fair to me.
Now suppose the police run it through the facial recognition system. It identifies one person as a 99% match, and the police go investigate this person. This scenario does not seem so fair to me.
Here’s how I see the math:
P(A) = P(robber has a doppelgänger living on the same block) = .01
P(B) = P(robber had a doppelgänger somewhere in the database) = .9
P(X) = P(police screw up investigation, and will convict the suspect whether or not they are guilty) = .2
P(AX) = .002
P(BX) = .18
The exact numbers are made up, but as long as P(A) << P(B), you can see you this tech will result in a huge increase in false convictions. Even if P(X) is low, the number of false convictions increases by P(B)/P(A).
The issue with the P(A) and P(B) argument is that police already use databases heavily, and most people don't have any problem with it. But why when it comes to facial recognition, is it too dangerous to use technology to drive efficiency.
If they're looking for somebody named Jane Doe, anybody with that name shows up on a list and police investigate. Of course if there are Jane Does in a 2 mile radius, they start with those. So why not just say if the system delivers a match within x accuracy and the person is within y residents (plus a variety of other variables), and x/y is below a threshold, then the match can be presented to police for further investigation.
Searching databases for matches is fine for names, or fingerprints, shoe prints, tire track, fiber analysis - but not faces? I personally wonder if it's really any different, or if its just better tailored for the media outrage machine because "China does it", or because "facial recognition targets minorities".
This tech can be good if applied to a narrow range of people like you suggest (eg. only searching people who live in neighboring blocks) but nobody is actually doing that. We should pass laws requiring a rigorous analysis of these probabilities for such databases to be used, including a conversation about what rate of false positives we are willing to tolerate. Guardrails should be put in place to enforce those limits. If this is too hard, we don’t have a strong enough handle on this technology to be using it.
Here’s the scenario that scares me the most:
Police identify a suspect using facial recognition. Then puts that person in a lineup for a witness. Of course the witness is going to say “that’s the one!” because the suspect actually looks like the perpetrator. The witness will be sure, the cops will be sure, and a jury will convict. And this scenario is completely determined by the use of the facial recognition database. This will happen unless we pass laws to prevent it.
A person understands the context in which they are naming another person based on their face and can weigh internally whether they are sure enough to tell the police. The algorithm just spits out a confidence for the Police/DA to do whatever they want with.
A person can be cross-examined in court. Is that little old lady a blind racist who is accusing a person of color? Her credibility can be attacked in court. The algorithm was trained by racists? That's a LOT harder to attack.
I don't think the police should be barred from using the tech, but I certainly don't think it should be admissible in court.
If the police have to ask people to perform identification, then they have to follow some sort of investigative procedure to arrive at who they ask to perform identification, and who they present to be identified. Of course this has its own problems, but facial recognition databases really ramp those problems up.
Then if you get falsely identified, you’re gonna have to prove it. How much time do you spend every day, not committing crimes, but without an alibi to prove it?
The difference is, we know how to engineer for low trust and high surveillance. We don't really know how to engineer for high trust and low surveillance. So our practical choices are to either live in a low trust low surveillance society and having to constrain ourselves as that requires, or to live in a low trust high surveillance society that simulates a much more pleasant high trust society.
And in a high surveillance society, whatever the initial conditions, the cost/benefit ratio of cooperation vs. defection is biased towards trust. In other words, the ersatz high trust society would become the real thing, of course only provided the surveillance is trustworthy. For example, China's social credit system might work if it is administered sans corruption. It has to be self-reflexive.
As incredible as it seems, Mrs. Grundy will save us.
* These systems will crowdsource "evidence", and by proxy, accountability - what happens when a mistake is made?
* Increased centralization will eventually enable 'single point of failure' scenarios in systems of a national/international scale.
* What happens when someone misuses or gains unauthorized access to such systems? Will the damage be reversible?
Such technologies will be deployed before accountability for them is defined, leading to a temporary absence thereof. Systems of the past (digital or otherwise) were inherently tolerant to misuse, malfunctions & misfires because only so much damage could be done before the errant behavior was discovered & handled. Today, how many seconds would it take to plunge a country into chaos?