Hacker News Reader

Comment by ankit219 | Hacker News Reader

When i read these implementations, one thing I think about is the British post office scandal, also known as Horizon IT scandal[1].

These technologies are probabilistic with some nonzero chances of error which can't always be easily detected. These are implemented by people with some knowledge of the tech used (and hopefully the understanding of error), but are to be used by people who expect the output to be 100% true. I do not know how they would deal with tech errors which do not look like errors, but feel like there should be more cognizance about this and expectations should be set accordingly. Same goes for any facial recognition tech deployed by authorities.

[1]: https://en.wikipedia.org/wiki/British_Post_Office_scandal

View on HN

Which is why a human being has to review the camera footage and decide to issue the citation. If the system makes an error, and the human checking it makes an error, then the human receiving the citation can contest it. Seems a reasonable number of safeguards to me.

Certainly a lot better than the big tech "we banned you but won't tell you why or allow you to talk to a human about it".

My only problem with more camera-based law enforcement for traffic violations is the fact that they can't tell who was driving. SO the fine goes to the registered owner of the vehicle, so folks won't want to lend vehicles. Of course, one argument is that you shouldn't loan your vehicle to some one who is going to be breaking the law... but is that really the world we want to live in, where we can't loan anything to anyone for fear of them misusing it and the owner being held responsible? I don't have a good answer there.

> one argument is that you shouldn't loan your vehicle to some one who is going to be breaking the law... but is that really the world we want to live in

If you are talking about a car, then yes, that's the world I want to live in. Would they even be insured anyway?

You can get insurance that will cover you when driving some one elses car, and you can get insurance that covers some one else driving your car. Wether either of you have such insurance, is a good question to ask before lending your car to some one.

By default most insurance policies (in the US, at least) cover incidental occasional use of drivers you permit to use your car that don’t actually live with you. They have to be licensed to drive, but that’s essentially the only restriction.

In Ireland/UK, I don't think they cover that at all - They are even a bit sketchy about named drivers. Insurance covering any vehicle tends to be rare b/c they are way more expensive, and usually vocational.

Interesting. Just a quirk of different regions then. In the US it’s essentially assumed that incidental use from friends and such is such a small usage that the insurance risk is minimal.

To be clear, insurance for US vehicles typically only covers a specific named vehicle, although most policies have a few carve outs for specific purposes (eg, 30 days of grace period when you buy a car before adding it to a policy, as well as applying to any car you rent). If the owner of the vehicle doesn’t have insurance and you borrow it, then there presumably wouldn’t be any insurance coverage at all.

On the issue with named drivers: I had two cars, and tried to add a named driver (my partner) to one, but they wouldn't let me. I assume the issue is they thought it likely the ND was the real primary driver, and that I was simply trying to lower their premiums since theirs was higher.

I suspect there's a similar issue with F&F. The cost of insurance is based on you personally, and if they allow NDs, or friends to drive, it's an unknown risk as they can't really tell how much you will actually use the car.

Also, I don't know how much insurance differs between US and EU. IMHO, Irish car insurance is expensive. perhaps due to lenient courts?

I mean, if you switch the word car to any other dangerous object then yes that is the world I want to live in. "Sorry Bob borrowed my handgun and shot up some stuff" isn't a thing that sounds like it would make the world a better place, and yet we treat cars like they are not death machines if misused.

> SO the fine goes to the registered owner of the vehicle, so folks won't want to lend vehicles

You just declare who was driving when you receive the fine, if it's not you. It already works this way in many countries.

>human receiving the citation can contest it

Only if it is that easy to contest it.

It’s not 1603 where you have to walk 60 miles uphill both ways through the snow and packs of wolves to beg the local lord for mercy. In most places, it’s a form on a web page (I first did that in the 90s!) or dropping a letter in the mail. It’s also very easy to avoid simply by not choosing to stop in the bus lane, which anyone capable of passing a drivers test can trivially do.

If your friend borrows your car and gets a fine in that time, you can pass the fine on to them. If they don't pay it, you can reconsider your friendship with them and possibly take legal action if it's a large fine.

Not a problem, if person contests, double-check the picture. Like with a speed camera.

For most of the cases, won’t need a human in the loop.

One might argue that a citizen shouldn't need to prove they didn't break a law, but rather the government has to (reliably) prove that they did.

I wonder if there's a phrase for that.

I've contested the red light camera fines in my municipality a few times now. In every case I request the footage and each time it's been overturned because the footage clearly shows me stopping before turning right on read. Ive never had to provide proof that is the states job.

The state should have reviewed the footage and not charged you in the first place.

That should be codified into law because the state is doubly incentivized to do nothing about false positives. If you pay it, they get free money. If you don’t, they get free labor.

In this case, the article clearly states that a human will review the footage... not sure what happened in the above cases, if there was no review, or if the review was simply rushed? I do think it would be fair if the state had to pay the same fine back to the person they incorrectly fined as compensation for their time.

You did have to provide proof! you were assumed guilty until you proved yourself innocent by using the evidence that the state themselves already had, but didn’t present as proof of the infringement.

In this case the assumption is that the camera evidence is accurate and conclusive. If it was a blurry picture, I don’t think they would be asked to provide additional evidence to prove they weren’t guilty, the ticket would simply be thrown out.

In other words, the burden of proof is still on the government to provide sufficient evidence. If they won’t or can’t, there is no case.

The person I replied to was required to challenge the assumption that they were guilty, and the camera evidence exonerated them every time, so by your standards should have been thrown out. The “burden” was on them to “prove” themselves innocent, its literally in the name. Just because the state possessed the evidence to prove his case doesn’t mean the burden was on them

Exactly. Forcing someone to challenge a fine that they've already been issued is not presumed innocence, it's presumed guilt. If someone fails to act affirmatively to challenge the fine for any reason (work commitments, disability, death in the family, the list goes on) then they're on the hook for the charge in spite of the fact that the state knowingly used an inaccurate system to issue the fine.

Meanwhile, I've never had to contest a false positive because my state doesn't allow red light cameras. That's a lot of time that I haven't had to spend defending myself against spurious charges.

It doesn't especially matter where the burden of proof lies if the process for contesting requires any substantial amount of effort on the part of the victim of the false positive.

None of our processes have a 0% false positive rate. Not tickets written by humans, not criminal charges filed by humans, not court cases tried in front of a jury. So maybe a phrase would be "unrealistic expectations".

But what we should expect (and demand) is that the process have no higher an error rate than human reporting, and that the errors should be no harder to correct/appeal.

> the process have no higher an error rate than human reporting, and that the errors should be no harder to correct/appeal.

I would actually ask that the errors be easier to correct, not just no harder.

Even if the error rate is lower, automated systems will detect a larger number of errors and therefore are likely to create a larger number of false positives. If we're going to automate the creation of false positives we need to equivalently automate their correction.

Even if we accept automated evidence gathering we don’t have to accept automating charges based on that evidence.

That’s what the picture is for

You might be looking for one of those?

https://en.m.wikipedia.org/wiki/Reverse_onus

https://en.m.wikipedia.org/wiki/Burden_of_proof_(law)

I was being sarcastic, apologies :) The phrase I was referring to is "innocent until proven guilty."

False positives can be human validated

false positives can also be human generated

These tickets get sent in the mail with proof, the photos.

The only thing you can argue is that it shouldnt require a court case to dispute unless it fails a human review.

lol. No they don't.

It depends on the location, my local automated speed camera process involves getting a human-approved ticket in the mail that includes the photos.

I can get behind this idea if and only if the government makes it absolutely trivial to request a human to review the ticket, with no additional fee and no attempt to dissuade people from exercising that right.

There should be a phone number, a QR code, and a printed link on the ticket that all provide a one-step process to contest a ticket. As soon as a ticket is contested it should be put on hold with no payment due until a follow-up ticket is issued by a human.

If we don't do this then we've just automated the creation of false positives without giving the victims of those false positives any equivalent advantage.

Better yet, the website could allow you to login an instantly view the evidence against you, in addition to filing the appeal. Also, if your appeal is found valid based on the video evidence, you should get the amount of the fine back from the government as a fee for them wasting your time.

They will do this. Later when some place like Booz Allen Hamilton picks it up, they will observe that revenue goes up when dark patterns and obfuscation goes up. So the initial implementation will go away.

Los Angeles doesn’t have a meaningful process to contest parking tickets: it’s an automatic denial, regardless of merit. The only way to get actual review is to go to civil court.

Do you have any statistics on that or is it personal experience?

You know, that's a good question. I was basing that on personal (and social circle) experience. But from a Google search, it looks like 34% of contests are successful: https://www.nbclosangeles.com/news/local/is-las-system-to-fi...

That sounds a lot better than I typically hear but one thing I do keep in mind is that some people, especially men, take comments about their driving very personally and react only slightly less strongly than if someone said they were bad in bed. I’ve known a few guys who would rant about tickets as a tax without noticing that nobody who’d seen them driving believed it.

Sounds like the civil court is the meaningful contest?

My point is that no one should expect responsible or ethical behavior from LA in this area.

This doesn't work if the failures are correlated and very often happen to a certain group of people.

Imagine if the system was flawed and assigned all tickets with an unrecognized license plate to NULL or 000000. THe owner of that specific plate would have to spend their entire life contesting tickets.

A less extreme case of this would be a badly-trained model that always detects certain kinds of cars as occupying bus lanes, even if they're parked legally. If you're in the tiny minority that has one of those, you can't park anywhere without getting a ticket which you must contest.

If those cars end up statistically more likely to be owned by the poor / black / homeless, now the local government has a scandal on their hands.

Neither Kafka, Joseph Heller, nor Huxley could conceive of technology-enabled ubiquitous, concealed brutality and extortion. The Western world is rapidly flirting with the specters of pre-crime and automated enforcement that puts the burden on the accused to prove their innocence with Hobson choices while extorting fees from average citizens as a profit-center.

And in Western countries with weak privacy laws, this class of technology also raises the specter of private "social credit" behavioral tracking to discriminates against and quantize people by grading their online and offline behaviors in totally opaque manners.

how does it compare versus human error?

What the exact error rate is isn't relevant to OP's concern, which is that operators of the tooling will treat it as having a 0% error rate in a way that they don't with human judgment. We've already seen this play out with algorithms designed to predict recidivism or what have you—law enforcement and judicial officials place undue weight on the outputs of algorithms and trust them implicitly where they might ask questions of a human expert.

> which is that operators of the tooling will treat it as having a 0% error rate in a way that they don't with human judgment

This is possible but we have decades of camera enforcement experience suggesting that courts are aware that these systems are not as reliable as the manufacturers claim. People have been challenging photo tickets since the previous century and there’s a well-established process for challenging them.

I am not comparing it to a human error. There is a conventional wisdom that humans make mistakes, and we have redundancy built into various systems to counter the same. With machines and tech, the expectation is that they are accurate (which is not true, but mostly true in the important ways). The reason I brought it up is to hope we have that redundancy in this case and future implementations of AI as well. (Horizon had a bug in calculation system - which seems easy and straightforward and hence unexpected and undetected.)

A typical example would be breath analysers. If humans were to judge who is drunk, there is a high error rate. With tech, the error rate is low, but if there is a bug in one of the equipments, many wrong and unnecessary arrests might be made.

> With machines and tech, the expectation is that they are accurate (which is not true, but mostly true in the important ways).

I find this an interesting comment with your choice of examples as there is an extensive legal history about false positives (medical conditions, mouthwash, etc.) and the need to show maintenance and calibration records.

I was on a jury back in the 2000s where the San Diego police had to have someone from the crime lab testify about their process for managing their breathalyzers, and we ended up unanimously finding the guy not guilty on that charge because they’d messed up the process and had been a year late on calibration testing despite everyone agreeing he was drunk (based on his own testimony, receipts, and the bar staff, he’d had 3 double vodkas in the hour before getting behind the wheel, so we found him guilty on the other prong of California’s DUI statute - which is exactly what his attorney told us afterwards had been his prediction).

Your experience is very helpful. I took a theoretical example since where I live breath analysers are how you get fined for drunk driving. I didn't think it was possible to contest that.

I would rather that these things happen preemptively so that there are no such challenges to fight or tickets to contest just because an AI flagged something and humans went ahead with it. I mean that other tech is deterministic, while AI is probabilistic, but people still havent updated their models for such tech. And we need to keep those examples in mind.

Techbros always love citing the supposedly devastating human fallibility yet experience shows poorly configured/developed automated systems are much more of a concern.

Nice straw-man, but getting back to reality, nobody in this thread has put forth the belief that automated systems are infallible, or is arguing that point. Also the fact that humans are fallible, hasn't stopped us from setting up judicial systems based on human judgement - which work pretty well in several scenarios.

sTrAwmAn

Reply on news.ycombinator.com