Hacker News Reader

Comment by cactus2093 | Hacker News Reader

cactus2093

Genuine question - why not ban cameras altogether? Or ban the use of computers in police stations, make them write up all their reports by hand. I truly don't understand why there would be a line at facial recognition, it's just a law against making a process more efficient.

It's clear to me that there should be a line to prevent fully convicting someone of a crime without any humans in the loop at all. Don't replace the jury or public defenders with robots. But facial recognition could just be a way to look through a large amount of footage to find relevant clips that would then be reviewed in more detail by humans. Without facial recognition, you just pay cops overtime to look through footage themselves. It doesn't necessarily affect the outcome of the process at all.

I think facial recognition just needs some help with its image, it's just a tool that would save tax dollars. Or a means to Defund The Police, if that's your thing.

View on HN

>it's just a tool that would save tax dollars

No, nothing is "just a tool". Surveillance technology enables all sorts of new attacks. Automation makes things feasible. Cheap storage that allows weeks or months of security footage to be preserved changes what's possible. License plate scanners are "just" a technology that could be done with a bunch of cops on corners with notepads, but in aggregate they're a tracking of where everyone is and has been that could never be done without the technology. Facial recognition is very similar.

I still don't see why this is the place we should draw the line. If we want to ban license plate scanners because they can be used to track location, should we also make it illegal for police to access cell phone records or cell tower ping data with a warrant? Should we ban cell phones altogether so that the location data can never be collected in the first place to be misused?

Lots of new technology has some downsides, the laws should more specifically target the downsides instead of require that everyone pretend the technology doesn't exist. If nothing else, it's just far too impractical for a technology that is easily available and in everyone's pocket to be illegal to use. It will just incentivize obscuring and lying about using it, rather than truly prevent it from ever being used. It reminds me of the history of banning encryption, you're basically just making math illegal.

The trick is that police need a warrant to invade your privacy via cell phones or whatever, but they won’t need a warrant for the flawed surveillance stuff that can end up severely fucking over someone’s life/their family if the police ever decide they don’t like them for whatever reason, such as being an advocacy against police abuse ;)

> Surveillance technology enables all sorts of new attacks

Sure, but like any other tools it also enable all sorts of new benefit.

The preferable action would be to take advantage of the benefit while also try to fix whatever it is that cause problem with the tools instead of simply banning it.

The problem is that constant surveillance of people makes exerting power over them trivial.

So then that is the actual problem, try to fix that problem instead of banning it. Allowing people to own gun also make it trivially easy to kill people. So ban gun ?

If there’s a way to make it impossible for the government to use facial recognition to monitor people other than legally banning them doing so, it’s hard to imagine it.

why would i want to make it impossible for government to monitor people ?

government monitoring people by itself is not a problem, government monitoring people to harm people is the problem.

What need to be fixed is the usage of tools to harm people, not the tool itself.

The only thing that keeps them from harming you is your power relative to them. Handing them more power only increases the probability that they will abuse what powers they do have.

I thought the election of Trump would disabuse people of this idea that it’s smart to hand more and more power to central authorities. But at this point I think it’s hopeless. People just want the state to be Daddy, and assume or hope that its power will always be wielded by people they like against people they don’t.

For your analogy: yes, guns should be banned. That bottle has been opened in most places in the world, and some countries have actually put the genie back in, and it has resulted in dramatically lower gun death rates. The UK for example, as it has one of the lowest gun homicide death rates in the world: https://en.wikipedia.org/wiki/Firearms_regulation_in_the_Uni...

Moreover, yes, other deeply awful weapons are also banned. Biological weapons and Chemical weapons are both widely banned by all decent countries in the world. Nuclear weapons should join this list. They almost did, during the Reagan era: https://www.theatlantic.com/politics/archive/2016/01/ronald-...

That's what they're doing.

The problem is automated identification of faces. The solution is banning that.

I don't think that's the problem. The problem is that the justice system believes that it's enough to convict someone of a crime.

Even just arresting them or bringing them in for assessment is enough to put someone in a dangerous position, because cops can effectively legally execute people. So the problem is any police interaction is potentially deadly and this is giving cops another arbitrary, potentially racist tool to interact with people who are just living their lives.

Let me ask you a question: How many unarmed people were killed by police in 2020?

https://www.washingtonpost.com/graphics/investigations/polic...

According to this, 405, which is 405 too many.

So we have 50 total out of thousands, if not millions of potentially negative police interactions.

You can't build a system of robots without some fault-tolerance. How are humans supposed to keep up to that?

Based on this data - how can you make the assumption that a police interaction is an inherent risk to life that is always unjustified?

We mostly pay police to show up and be generally aggravating to people who might be doing bad stuff - in some sense "being sketchy in the vicinity" is a valid reason for police action, which doesn't even always turn into an arrest, let alone a life-or-death situation. Those cases are the vast minority of police interactions total.

I only see 50 there. Are you sure you're looking at unarmed 2020 and not for all years?

I also just looked at a few unarmed shootings at random to get an idea of what they were like. Of the three I looked up I saw one guy who was shot while fighting with a police officer after being discovered at the scene of a burglary. One guy who was shot after a car chase when, after they had caught him, he got behind the wheel of a police car and the police shot him to prevent another chase. One guy, after stabbing someone to death and being approached by multiple police officers, faked a draw to "suicide by cop". My point is, I don't think it's even clear that all (or even most) of the 50 unarmed shootings are unjustified.

People have a right to be armed; why would you consider using your rights to be worth a death sentence?

If it's illegal to be armed, you should change the constitution, rather than asking police to shoot evry armed person they see

This ignores the coerced plea bargains, which i don't think involve any conviction

>The problem is automated identification of faces

Whats the problem with this ?

I had a dream a few months ago where I was standing in front of a computer. I typed in my name and got back a stream of video which showed every single moment of my life from multiple angles starting from the moment I was born. The footage came from a variety of sources, CCTV, smartphone cameras, dashcams. The system had videos of everything anyone had ever done. I didn't like it. Found it kind of disturbing.

Then basically you says its boil down to matter of subjective preference. So you don't like it, fine fair enough then don't be surprised that other people may like it.

I don't say that? You just have to think about how this stuff is going to be used in the real world where we actually live. The thing which is objectively disturbing is how easy it is to lose control of one's life once the government is able to record, observe and judge every single thing a person does. I believe such a level of transparency is only beneficial when it's applied to people who hold power over others; to subject everyone in a society to that kind of scrutiny would just be laying the foundations for abuse.

The government simply record, observe and judge every single thing a person does is not problem, in fact it can be used for my benefit.

The problem is when they use it to harm me.

What need to be fixed is The government record, observe and judge every single thing a person does to harm people.

>to subject everyone in a society to that kind of scrutiny would just be laying the foundations for abuse

It can be used for abuse but not always, It can also be used for good, what need to be prevented is the use for abuse.

Fortunately, we can stop that from being a reality.

Everything else that you described – computers, digital records – has simple algorithms understandable by the average police officer or citizen. You type a document, it gets saved. You can do a full text search on it. You can pull up an old case and look at photos. You can quickly cross reference. All these tasks could be done step by step by a person and it would just take more time.

When it comes to facial recolonization or AI in general, could anyone really tell you why the computer decided to flag your face and why another similar one was rejected? Would you accept a search warrant when the judgement wasn't based on any human's perception but something which came out a black box? Who makes sure that the data sets used to train the models was 100% free of bias?

Since when do search warrants get automatically approved by a black box? It is on the judge to approve, not a machine.

If that's not enough, make a human sign off that they have confirmed or at least agree with the black box's conclusion, and put the department on the hook if it was obviously wrong.

> Who makes sure that the data sets used to train the models was 100% free of bias?

The box is comparing a source image to an image with a name attached to it. Like you said, no different than a human would do, with all of their own biases in play. We aren't talking minority report here, so there's no reason to think this is a hurdle that would be difficult to overcome.

I mean, FISA warrants had (have?) a 99+% approval rate despite many applications having serious deficiencies.

Utah’s e-warrant program has a 98% approval rate. Some warrants are approved in less than 30 seconds.

Warrants across the US are frequently approved on evidence as shaky as a single anonymous tip.

The problem (at least in my opinion) isn’t that facial recognition itself is inherently evil, it’s that:

a) There are no standards around the algorithms themselves in terms of reliability. Worse still, this can inadvertently lead to serious racial bias.

b) It can launder lousy evidence (this person looks vaguely like this grainy photo) into solid scientific evidence (the algorithm says it’s a 99.5% match!) providing justification for all sorts of warrants and raids on even flimsier ground.

c) It creates lousy incentives in favour of creating a surveillance state. The police already have out-of-control hardware budgets, and many don’t like the idea of that budget being used to monitor and record people at all times they’re in public.

If we had better police and court systems and better privacy law this might not have been necessary - but we live in the world we live in so I can see why this county would do this.

> Warrants across the US are frequently approved on evidence as shaky as a single anonymous tip

Here is the crux of the problem. If facial recognition were good, then it would actually make things better. If it were bad, then it isn't making things substantially worse- if anything, it creates a track record which makes things like racial bias more obvious and auditable.

The real issue is that there isn't sufficient judicial oversight, nor recourse for when it fails.

> then it isn't making things substantially worse

At least in the United States, any encounter with law enforcement can potentially end in violence, so anything that makes it easier to get a questionable warrant will make things worse.

As the parent post pointed out, questionable warrants are already supposedly trivial to get. This system, being more auditable as a single source rather than individual anonymous statements, means that it is far less likely to stick around if it consistently punts out bad matches.

Why aren't you afraid of good matches? Good matches open up far more horrifying possibilities.

> then it would actually make things better.

If you are stuck living somewhere where music is illegal, and you happen to listen to some music on the sly and get caught by a perfect system, and punished, how is that better?

Better would be if people can live freely without having to worry about a perfect system tracking their every action.

There is no person alive on earth today who does not do things that would get them punished under the law of some government somewhere.

And you want the law enforcement systems to be perfect?

No, I want them to be held accountable for when they are wrong, and the hypothetical example of a good facial recognition system removing human bias was an example of that, and could protect people from the other example the post I replied to, which was search warrants granted under a single unreliable source's statement.

Facial recognition isn't going to care that you procured illegally distributed copies of protected works (your ISP already does that anyway), so lets stop moving the goal posts. It is a reactive, not proactive system, and while I am not fully convinced that it is actually ready, I am also not buying the slippery slope nonsense.

>isn't going to care that you procured illegally distributed copies of protected works

Sorry, I wasn't clear maybe, but somehow you misunderstood me.

I wasn't talking about pirated music.

I was pointing out that some regimes have laws that forbid listening to music, period.

Fixing bias doesn't help when all it does is help accurately catch innocent people who in fact were accurately detected to be doing innocent things, and punishes them for doing those innocent things because those innocent things are illegal in that country for no good reason.

I don’t discount the usefulness of ai or facial recognition, however,

>Since when do search warrants get automatically approved by a black box? It is on the judge to approve, not a machine.

The USA PATRIOT ACT removed judicial oversight and allows for warrants to be issued that notify the judiciary and simultaneously issue gag orders on those notified. For federal cases warrants no longer require judicial approval in an undisclosed number of cases. Try and find a list of current Guantanamo Bay detainees or the judges who approved the warrants used to detain them. Until that’s dealt with I see why King County would ban this use as a defense against automated enforcement.

and people who take the oath as a court witness never lie

There are many tools to help make feature activations in Computer Vision interpretable by humans.

Class Activation Mapping [0] and Salience Maps are some of the most used and intuitive approaches. Lime and Shap [1] are tools to visually represent feature activation in an understandable way.

The idea that we have no idea what is going on with machine learning models is just flat out wrong.

[0]http://cnnlocalization.csail.mit.edu/

[1]https://github.com/slundberg/shap

> Would you accept a search warrant when the judgement wasn't based on any human's perception but something which came out a black box?

Isn't every PD that uses facial recognition keeping a human in the loop to confirm algorithmic matches first? And if so why would the overall process be any less accurate than a human match?

I think it's a matter of accuracy and bias. If the tech keeps making false accusations, especially against the same kind of people over and over, then a ban might be in place.

You could argue that humans make similar mistakes and bias, but the scale is always smaller when humans are involved, just because of how slow they are.

Say a model is wrong 37% of the time, and so are humans, but the model runs 1000 times a minute, where as humans perform the task 10 times a minute. That means the model makes 370 false accusations a minute, where a human makes only 3 a minute.

In effect, the number of people you bothered falsely accusing is still less when done manually, simply because it can't scale.

Lastly, there's an argument that the police shouldn't be too efficient, because some crimes might be better off existing in a grey area.

People do fear police supremacy and police states, the idea that they can find you for j-walking or jumping a fence as a dare, etc. Or that the state could become authoritarian, but you'd have no way to fight back as rebels due to advance tech of surveillance the police would have, etc. I'm not saying those are justified, but I think it plays into it, people are fine saying that only the highest risk cases are worth pursuing, but if a tech comes along that lets police pursue all cases, it might start to feel abusive to people.

P.S.: I'm neither against nor in support of a ban here, I'm still making up my mind, just answering some of the reasons I know off for being for the ban.

Is facial recognition more biased than manually comparing against mugshots? A lot of people seem to be writing under the impression that if facial recognition software is banned, police and government won't use facial recognition. The reality is that they will still use facial recognition, just in the old fashioned way by looking at suspects and looking at mugshots. And humans are often biased, too.

The problem is that jurors place higher faith in technology than they do in people, and that’s not necessarily warranted.

It’s one of the things swept under the umbrella of the “CSI effect.”

> Say a model is wrong 37% of the time, and so are humans, but the model runs 1000 times a minute, where as humans perform the task 10 times a minute. That means the model makes 370 false accusations a minute, where a human makes only 3 a minute.

Or they spend 100x as many man-hours and achieve the same number of false accusations, but at a much higher cost to the taxpayers.

> I think it's a matter of accuracy and bias. If the tech keeps making false accusations, especially against the same kind of people over and over, then a ban might be in place.

It seems like we should then ban eyewitness testimony. That has tons of accuracy issues and bias.

Please read the full post. The issue is inaccuracy and scale.

The scale is a complete strawman. They aren’t falsely accusing 300 people for a crime.

A great example is the Capitol Riots. They are narrowing the list down to manageable numbers. They aren’t suddenly arresting thousands of people for a crime committed by three.

I think there are very good reasons to keep the government out of the FR business, at least for now.

Facial recognition is the enabling technology for automated real-world repression. Maybe it is true that we can find a way to tame it to gain efficiency gains without destroying open society. But right now it looks like a dangerous one-way ratchet.

I think the tough part is you can't really ban failing to think in the presence of an algorithm. People see an algorithm produce a result and often assign far too much confidence to it.

"The satnav directions say to turn onto this pier."

"The model says this house is worth 200k less than the ones in the whiter neighborhood a quarter mile away."

"The system recommends the patient have 40% allocated home health care hours per week going forward."

"The algorithm says this grainy footage of someone far away from the camera in the dark and the rain is X."

If you put humans in the loop informed by an algorithm making judgements, the human will often defer to the algorithm. Does the algorithm give an output that indicates its uncertainty? Based on what model? How equipped is the human to consider that?

Depending on how you read the specific text of the statute, I think they banned the usage of electronic mug shot databases.

>"Facial recognition" means an automated or semiautomated process that assists in identifying, or verifying the identity of, an individual based on the physical characteristics of an individual's face.

So I have a still from security camera footage. And I have 50,000 mug shots stored in the mug shot data base. I can see that it's a man in the image, he's white, between the ages of 15 - 50. So I go to my computer and put in the filter for the mug shot database to include those criteria. Sounds to me like I just used a semi-automated method to attempt to identify the person in the still image.

This ordinance is dumb. The use of a computer to pattern match, to automate what a person can do, is a lever that makes everyone more efficient. What's the logic in making someone thumb, or click through, every photo in the state's mug shot collection instead of allowing a system to reduce the possible candidates, if any.

If when you take people's mug shots, you also record their age, gender and race (should you?), then you can filter by those characteristics without the automated process being "based on the physical characteristics of an individual's face" -- you'd just be filtering on a DB column like any other.

But if you didn't record or have access to those fields, and you tried to infer age, gender and race from the photo, and then filter -- well, what if your system is much more accurate guessing ages for some racial groups than others, or more accurate in distinguishing between white and latinx than between latinx and asian? Suddenly your "filtering" process can also have a bias problem, where some people are more likely than others to be included in the results for a query which shouldn't actually include them.

> then you can filter by those characteristics without the automated process being "based on the physical characteristics of an individual's face" -- you'd just be filtering on a DB column like any other.

A case agent looks at the still photo from security footage, concludes the person is male and white. They put that information into the mug shot database and it returns the subset of all mugshots that are male and white. They have just used a semi automated system to assist to ascertain the identity of the person in the security footage based on the physical characteristics of their face. This is forbidden by the ordinance.

>what if your system is much more accurate guessing ages for some racial groups than others, or more accurate in distinguishing between white and latinx than between latinx and asian? Suddenly your "filtering" process can also have a bias problem, where some people are more likely than others to be included in the results for a query which shouldn't actually include them.

What if it isn't? Doesn't matter. It's a tool to attempt to match two photos. If a tool is not as effective due to the way light and different skin tones interact, it just limits the leverage factor of the tool. It doesn't make it wrong and it isn't an infringement on people's rights. We use DNA matching even though it can't distinguish twins and sometimes it comes back and says it's a close match for Bob which leads them to suspect Bob's brothers family, e.g. Golden State Killer. I fail to see why tools have to be perfect or not be useful.

Laws can always be changed, seems sensible to me to prevent this stuff being rolled out when there are so many unresolved problems and abuses with it.

I generally don’t want my government to “move fast and break things”

>I think facial recognition just needs some PR help

They are making their own bed in many cases. Here's one of several interesting moves by Clear View AI, for example: https://twitter.com/RMac18/status/1220876741515702272

I agree. It's hard to understand how this is different than using a computer to detect, for example, credit card fraud. Both pour through tons of private data (typically using algorithms that don't make sense to a layperson -- or maybe even to experts since they likely use some DNNs) to determine is something problematic may have occurred. Both technologies can be used for nefarious purposes, but both can have safeguards that minimize the likelihood and impact of these purposes.

I'm way more scared of guns than I am of facial recognition and as a country we will never ban guns.

You won't have to choose. We'll see guns attached to facial recognition systems soon enough, both "authorized" and not.

Watch Netflix' Coded Bias if you want a deeper understanding of the political motivations here - and they really are almost purely political.

I'm not for an instant saying that there aren't any implications in terms of how the technology can and should be deployed - but some people just want to be angry and offended despite a lack of understanding of what they're actually on about.

Facial recognition struggles to accurately identify black people and women.

The difficulties around both are probably around edge and contour detection:

Black people have darker skin and less angular features - the shadows on their faces are not only harder to see (due to blending with skin-tone), but softer to begin with.

Women routinely leave the home wearing enough makeup to not only look less like themselves, but often look more like other women.

There are ways to mitigate these issues to some extent: cameras that work better in low light and codecs that do a better job of encoding relative rather than absolute light levels will do a better job of preserving the information in the faces of people with darker complexions.

As far as makeup is concerned, it can misrepresent the shape of a face - if you only have one angle.

Treating the face shape as invariant as Δt -> 0 and running on video (even better with multiple synchronised cameras) will allow for topography extraction despite dealing with someone deliberately misrepresenting what they look like.

That's all very true but any person using their own eyesight would have the same issues. This should be used as a narrowing of a search, same as Google's algorithm. Google doesn't give you the answers to your questions, they hopefully point you in the right direction, but sometimes don't give you a clear answer at all, but in clear and obvious cases it does match. Google also has programmed biases, but we all accept Google is one of the best tools.

Not to the same extent - we've got good enough binocular vision and pretty good handling of dynamic lighting.

Compare that with the usable colours and bad blacks of standard absolute colour encoded video.

Your (poorly informed view, IMO) of the world does not consider immoral and illegal activities carried out by law enforcement and those with authority. Stalking, and harassment and abuse based on stalking, is a problem. Making it easier and simpler with facial recognition causes bigger problems. And this is just one example.

> Or a means to Defund The Police, if that's your thing.

Are there many sincere researchers studying flaws in facial recognition advocating its unequivocal ban forever? Joy Boulamwini:

> At a minimum, Congress should require all Federal agencies and organizations using Federal funding to disclose current use of face-based technologies. We cannot afford to operate in the dark. https://www.govinfo.gov/content/pkg/CHRG-116hhrg36663/pdf/CH...

Timnit Gebru in the NYTimes:

> Do you see a way to use facial recognition for law enforcement and security responsibly?

> It should be banned at the moment. I don’t know about the future. https://archive.is/JqiqP

Are the flaws Algorithmic Justice League finding real? Someone has definitely been wrongly accused by a facial recognition error (https://www.nytimes.com/2020/06/24/technology/facial-recogni...).

Is there certainly an impression that activists want a forever ban? Yes, Joy Buolomwini and Timnit Gebru are frequently represented as advocating against even "perfect" facial recognition.

It's true, lawyers tend to advocate for a forever ban. I don't think these researchers advocate for a forever ban. If you read their raw words, they are much more intellectually honest than the press allows.

Is your line of thinking co-opted by people acting in bad faith? Certainly. You would feel very differently about the deployment of the system if you were so much more likely to be falsely accused by it due to the color of your skin.

Every intellectually honest person's opinion of the justice system eventually matures. You'll come around to delegating authority to researchers over lawyers and police.

There are many things the law permits at a small/manual scale but forbids at a larger/automated scale. This makes sense because in the latter case, the cost to the person doing the act is smaller.

For example, there's no law against making one phone call to a random number in the phone book. But if you use an autodialer to call everyone in the city, you're breaking the law.

I was thinking much the same thing. Facial recognition is just a tool that makes it easier. If there are problems with false detections, you should be able to insert a human check in the process. But banning something makes it look like they are doing something. Meanwhile, it makes the job harder for police.

it's about security and convenience.....it's a tradeoff so it's impossible to achieve both perfectly.

typically you find the balance....right now you are suggesting that the improvement is a convenient way of increasing security but the reality is, that security in one sense can lead to loss of it in another.

example. Assuming all things are Good and no possibility for corruption then it would be purely beneficial....but since those aren't controls, and not realistic either, we cannot sacrifice security for convenience in this case.....if that makes sense.

Would you make a similar argument that allowing the police to track location of all citizens 24/7 is just making the process of having officers follow people around more efficient?

At least ban speed cameras! I think we can all agree on that. Also, all those cameras that detect when you make illegal turns and send you nasty fines through the post.

Sure...just give the public the same access that police have. Then if dirty cops break the law or mislead the public, everyone will know.

There is a subtle misdirection when some jurisdiction "bans government use of facial recognition" - that ends up, or is by design, how private agencies are created that law enforcement then contracts with less oversight than before.

The lack of understanding of the downfalls and conflicts with facial recognition with justice system and this post is every reason why it shouldn’t be near the justice system.

I agree with you - this is just forcing policing to be more expensive and inefficient. It would be absurd to ban a police department from using Microsoft Office for example, and instead forcing them to track data in physical spreadsheets. Banning facial recognition is equivalent to forcing a police department to hire enough staff to monitor county-wide feeds and match feeds against a long list of criminals they're looking out for. With humans in the loop and requirements for when facial recognition can be used, I feel like there isn't a "real problem" here. But when we look at quotes from this article, for example the person quoted from Latino Advocacy who is concerned about ICE enforcing the law against illegal immigrants, it's clear that the motivation to ban facial recognition is really more political in nature - it's about letting some people get away with breaking the law more so than fundamental concerns with facial recognition.

Reply on news.ycombinator.com