AI watches millions of cars and tells cops if you’re driving like a criminal | Hacker News Reader

AI watches millions of cars and tells cops if you’re driving like a criminal

forbes.com

83 pointsthehoff95 comments

There's no way searching someone's car, based off their legal driving behavior, could be misused!

The police have no repercussions for sending warrants to Google for phone information. Is searching someone for their legal driving behavior any better?

https://www.nbcnews.com/news/us-news/google-geofence-warrant...

"Searching" a car is generally understood to mean looking inside of it.

Observing the same car coming and going on public roads seems like it requires a lower bar. IANAL.

They did search his car (based on his driving patterns) and found cocaine.

The question isn't whether it was legal to observe the driving, but whether it was legal to search the car based on the observed patterns.

They pulled him over based on his driving patterns. They searched his car based on a K-9 unit alerting on the vehicle.

https://www.leagle.com/decision/infdco20221109635

Why do we allow K-9s to count as probable cause when they can be roughly as accurate as a coin flip? Eg. U.S. v. Bentley, where the particular dog in question had a documented 40.5% false positive rate in the field.

True but irrelevant. According to the Supreme Court K-9 alerts are probable cause. He was observed based on AI, pulled over based on multiple traffic violations including changing lanes without signaling, illegally passing on the right, and driving 15 mph over the speed limit, then it was searched based on a K-9 alert. Unless the police are simply lying about the traffic violations and have no evidence to back them up, this guy is going back to prison.

> True but irrelevant. According to the Supreme Court K-9 alerts are probable cause.

Then we have a Supreme Court problem.

Good thing I only drive 14 over!

> documented 40.5% false positive rate

Doesn't that mean that more than half of the people that the dog signaled on were actually found to have drugs? That actually sounds pretty good. I assume it would meet the legal standards of "reasonable suspicion" or "probable cause".

If I have my documents folded/crumbled up, stuffed into pockets... such that someone standing near me can see them sticking out, do they get to ignore warrant requirements to my person and papers because of that?

I'm not a lawyer either, but I'm pretty sure the answer is no.

Paper's not magical. If my behavior is sticking out of my pockets for everyone to see, they're not allowed to search through those unless they've got a fucking warrant. The amendment wasn't written with the express (or even covert) purpose of giving them ludicrous excuses to bypass its requirements. Reading it as if it were is wrong and deceitful.

The local politicians (city, county) provide the funds. Can you take this up with them at a city council meeting?

> The police have no repercussions for sending warrants to Google

Yes, because judges, who are also human beings, love when cops flood them with stacks of bullshit requests.

What judge? Like ~third paragraph states system operates with no judicial oversight.

> The police have no repercussions for sending warrants to Google

Those warrants.

The police cannot issue warrants. They can ask. Only a court can issue a warrant.

A warrant requires a judge, otherwise it's not a warrant.

If Google wants to obey non-warrants, that's on them but not legally required.

Most of these new data surveillance tools don't have a judge in the loop.

New like sniffer dogs?

They do if they want a conviction.

This system seems to work by using the locations a car has been to compare see if other cars with a similar location history are more likely to be used to transport drugs if I understand correctly.

Assuming that this is how that system works I can't see how this is legal. Let's say police is only searching cars between place A and place B, but not between place A and place C. Therefore they find drugs in cars that drive between A and B. This system therefore is now flagging all cars that drive between A and B as suspicious due to the bias of the searches the algorithm was trained on. It could be that there are more drugs being smuggled between A and C, but the system wouldn't know due to data containing no searches on cars between A and C.

If we now say A=downtown, B=lower income area and C=high income area I think it becomes very obvious how such a system that allows police control cars based on past searches is discriminatory and everything but fair. It enables searches without any reasonable grounds and I have honestly no idea how any of this could hold up in court, but what do I know.

How are you making the leap from discrimination to the search being without "reasonable grounds"?

Let's say I'm driving from A to B and the police stops me with a trunk full of heroin.

I'm supposed to argue that if they'd only searched drivers travelling from A to C they'd find trucks full of cocaine instead?

Maybe that's true, but I'm still breaking the law.

You could argue that police attention should be more distributed. But surely "reasonable grounds" for a search should be measured primarily on the basis of the "hit rate".

I am pretty confident that having a higher hit rate doesn't have any affect on whether it's reasonable grounds.

This argument has been made time and again with respect to racial profiling. Even if it works that doesn't make it right. You either have an actual suspicion or you're just engaging in profiling of some sort and that means that everybody is suddenly a suspect. This is why dragnet surveillance is such a bad idea.

Why is it a bad idea?

I live in a dense European city, and the other day I was driving down a busy road at a certain time of day, and was made to undergo a breathalyzer, along with anyone else going the same route.

If I'd failed the test, would I have been the subject of injustice?

After all it would probably be true that had I been driving drunk I'd likely have gotten away with it if I was driving in a more rural area.

Now, in the case of the article an AI system accurately predicted a crime being committed.

So, to extend that to my example: If the police hadn't tested me at all, because they could have used some AI system to spot that I wasn't driving like a drunk person, we'd all be ... worse off?

Surely such systems could benefit everyone. They'd be better at ignoring the law-abiding, and be more effective at catching criminals.

You can't compare the use of a breathalyzer which takes manpower to operate and analyze the results of to something that you can run 24x7 year after year. Such an increase in capability turns the whole principle of policing on its head.

Note that there is no brake on this at all, not even in principle. You could be followed from the cradle to the grave by technology and if it gets it wrong the onus will be on you to prove that. That's what you get when you start with the assumption that everybody is guilty. Your breathalyzer results are the result of a fairly simple bit of chemistry, one that is reliable enough in practice to have a low enough number of false positives that it is deemed acceptable, and in case of a positive test there is a backup test to ensure that a false positive doesn't lead to detrimental effects.

AI based mass surveillance starts out from the premise that if you make everybody a suspect and then filter out the 'good people' all you are left with is the bad apples. But that is not how law enforcement is supposed to work. It's possible that you would like to change the whole presumption of innocence concept. But until that is the case any kind of surveillance should be specific and limited in scope as much as possible so that ordinary citizens can live their lives unobserved by the authorities.

I recognize the power of utilitarian arguments, but they are a very bad idea when you start basing your law enforcement rules of engagement on them.

There are people that can make these arguments far more eloquently and from a more informed perspective, so I'd like to close with a link to an article about this that may help shed some more light on this.

https://harvardlawreview.org/print/vol-126/the-dangers-of-su...

There are many more like it but this is a pretty good starting point.

    > Such an increase in capability turns the whole principle
    > of policing on its head.

Another way of looking at this is that we're getting a computer-assisted regression back to the historical norm. If I lived in a small village I'd probably be waived through a breathalyzer checkpoint, as it would be well known that I don't drink.

Similarly the local cops would know who's acting suspiciously in a manner similar to the subject of the article, and would have pulled him over via traditional policing methods.

    > [...]That's what you get when you start with the assumption
    > that everybody is guilty.

I think there's a clear line to be drawn between systems being discussed here, which are inferring things about your behavior in public, v.s. the more slippery slope of Minority Report-style thoughtcrime.

    > [...]so I'd like to close with a link to an article about[...]

Thanks, yeah I'm quite familiar with them. I'll just note that there's a bit of a prevailing difference in philosophy between the US & European perspective on this. I happen to like the American one better in principle, but whenever I visit policing seems to be more intrusive in everyday life as a practical matter.

So I'm not interested in continuing the argument.

I'll only note that at the start of this thread I raised a question that's entirely orthogonal to this. I.e. whether you agree with (to use your words) the "assumption of guilt" or not, it seems like an odd hill to die on to claim that such a system (or lack of one) is injust if it's not equally rolled out to the entire population, and all socioeconomic groups.

> it seems like an odd hill to die on to claim that such a system (or lack of one) is injust if it's not equally rolled out to the entire population, and all socioeconomic groups.

It isn't equally rolled out to to all socioeconomic groups because 'criminal behavior' automatically disenfranchises the poor and the people of color (or both) because they are more likely to engage in such behavior. This is exactly why the 'war on drugs' really is the 'war on blacks'. A war on crime would be no different.

But since you are not interested in continuing the argument we'll let it rest, but it does make me wonder why you started it in the first place.

This has been done for a decade. If you regularly drive between San Diego and Seattle, say, you will eventually get pulled over for some made up infraction(i.e. parallel construction) because the DEA has informed the local/state police that you're likely a drug mule.

https://www.csmonitor.com/USA/Justice/2015/0127/US-car-spyin...

I’m not in favor of predictive policing. It’s not illegal to be suspicious. So I can’t get on board with this program.

But I am a huge fan of speed cameras from a public safety perspective. Excessive speed is a huge contributing factor in accidents yet I rarely see it enforced. There’s no point in having a law if it’s only going to be enforced against the poor or minorities. I don’t see it as a privacy violation since any person can plainly see you’re speeding.

I’m also in favor of more cameras to catch criminals. The first order effect of this is immediate: fewer criminals are around to crime. The second order effect is that fewer crimes will be attempted as the chance of being caught and punished approaches 1. This actually decreases the amount of police work that has to be done.

Having cameras is effectively the same as having a police officer. Except cameras don’t shoot people, and footage is a lot more reliable than sworn testimony.

>Except cameras don’t shoot people

Obvious photography jokes aside, this seems like a case of bad police training, or excessive guns in the US causing police to need to 'shoot before being shot'.

If someone is looking a bit dodgy standing outside a kindergarten, I'd absolutely want "predictive policing" to go have a chat to the guy. May just be a case of someone excessively caffeinated waiting for their kid, in which case no foul.

However the appearance of the police keeping on top of things deters others from committing actual crimes. I'd rather police prevent abductions than solve them.

I live in the UK with a high use of cameras for everything. For speed it clearly works where there is an actual camera (it works less well between cameras).

For crime there’s some evidence it might have a deterrent effect on property crime and drug crime.

There is very little evidence to show that it has had any effect on violent crime, public disorder offences, and the like.

There’s also an argument that over-reliance on cctv has reduced the kind of policing people most favour – community policing.

The government has recently banned Hikvision cameras (one of the most ubiquitous brands) from its buildings over concerns about the Chinese company’s use of data and their role in the oppression of Uyghur Muslims in Chinese state run “re-education camps”.

Cameras can be useful in certain circumstances, but their efficacy is often overstated, and there are many downsides.

Most new cars come with telematics, some have always-on cellular connectivity for data collection (and supposedly for automatic crash notifications).

It's only a matter of time until that becomes required (like the backup cameras became required in 2018 for example), and then it's only a matter of time before accurate / GPS positioning becomes part of the "required telematics" that get automatically shared with authorities (in addition to data brokers, advertisers and so on).

> (like the backup cameras became required in 2018 for example)

But there's no law saying you can't disable the backup camera.

Ugh. Their model will guaranteed to be borked: PR(is criminal | visited locations) != PR(visited locations | is criminal).

In other words, drug traffickers probably follow patterns typical for a larger subset of the population, which in of itself is not an indicator of guilt.

It seems like this goes directly against the Constitution.

I'm no constitutional scholar, or lawyer of any kind. But, I don't see how.

The closest thing seems like "searches and seizures". But none of that seems to be happening. That would probably be applicable if it was deployed on private roads or something.

Or maybe there's some other part of the constitution you had in mind.

Searching through a database of license plates to determine cause for a stop is plainly a warrantless search, SCOTUS is a dumpster who bend over backwards for LE but lets not pretend there's any merit to most of these decisions.

But it's not a search of the citizen. Presumably the owner of the database consents.

Pulling over the vehicle to search it on that basis is a search of the individual.

Yeah I hate what's happening, but I don't think it's unconstitutional (though I would strongly bet that the founders never could have imagined the capabilities that modern tech provides). It's more the equivalent of hiring billions of cops and having them watch every part of the road all the time and flagging when they see something "suspicious." I don't think the constitution would prohibit hiring billions of cops, if some jurisdiction were inclined to do such a thing. The mass surveillance is an extension of that principle.

> It's more the equivalent of hiring billions of cops and having them w

That itself would be unconstitutional. There is also the "spirit of the law" which is "you're not allowed to go digging around to find ways to violate these rights by inventing outlandish schemes that no one could imagine in the late 1700s".

When technology has no other effect than to cause the same result that might be accomplished by hiring billions of cops to watch everyone, then that effect itself is, without a doubt, unconstitutional.

> I don't think the constitution would prohibit hiring billions of cops, if

Why, though? That's the real question. Do you think the point of it was to just create a few hoops for the authorities to jump through, maybe for shits and giggles?

Why do you think this?

> That itself would be unconstitutional. > When technology has no other effect than to cause the same result that might be accomplished by hiring billions of cops to watch everyone, then that effect itself is, without a doubt, unconstitutional.

Which part of the Constitution would this violate?

> Why, though? That's the real question. Do you think the point of it was to just create a few hoops for the authorities to jump through, maybe for shits and giggles? Why do you think this?

I'm going to assume good faith on this question. The US Constitution mainly applies to the Federal government. The vast majority of these police units are at the state and local levels. The US Constitution enumerates the powers of the federal government, and then delegates all the remaining to the States.

In this case, they searched a person's car after the software system highlighted their behavior as suspicious. The constitutional issue would be whether the resulting search was reasonable or not.

We can agree that if the system was very poorly made and flagged 90% of drivers, then the flagging wouldn't constitute "probable cause" in a legal sense... so how good does the system have to be before we're willing to accept searches based solely on the system's recommendation?

Well no, they stopped him to investigate after the pattern recognition. They used a K9 who alerted. Turns out the system was right on the money because there was criminal activity.

The US constitution did not address the matter of privacy, but US legal tradition - over the centuries - has interpreted it to _imply_ such a right, or some kinds of privacy protection.

See: https://en.wikipedia.org/wiki/Right_to_privacy#United_States

Also, The search part, and perhaps even the inspection of cars using the cameras, may be considered "unreasonable search" of the kind forbidden in the 4th amendment to the US constitution.

Still, remember that both law and policy are made by the state and applied by the state, in the service of state interests, which overlap the interests of subjects only to a limited extent. The US government already basically wiretaps most people's private communications (via Google, Apple, Yahoo, Microsoft etc. - Snowden revelations) - so why not spy on people on the roads?

> But none of that seems to be happening.

Google civil asset forfeiture.

I'm aware of it. I'm opposed to the practice, and it seems unconstitutional to me.

But I wasn't talking about that.

The use case for this product is manufacturing suspicion. Perfect for increasing revenue from civil forfeiture.

Just imagine what John Madison would think of the British red-coats doing this.

>The right of the people to be secure in their persons, houses, papers, and effects, against unreasonable searches and seizures, shall not be violated, and no Warrants shall issue, but upon probable cause, supported by Oath or affirmation, and particularly describing the place to be searched, and the persons or things to be seized.

>The enumeration in the Constitution, of certain rights, shall not be construed to deny or disparage others retained by the people.

Which part?

Do you refer to the Fourth Amendment to the US Constitution?

That prohibits unreasonable searches, but can the AI define what counts as "reasonable"?

I, like other intelligent criminals, break only 1 law at a time, so this should be fun

It seems like our world is getting more dystopian every week that goes by. I can't wait until this pendulum starts swinging the other way.

What makes you convinced that it still can swing the other way?

Mostly because history indicates that this is a pattern that repeats over time. But also because if I don't have at least that small hope, the despair would be hard to handle.

Ok. I'm a bit concerned that the ways the pendulum has historically swung back are more or less closed off because of all the ratchets in place now. You'd have to upset a very large number of applecarts to get to the point where you can reverse the swing because with these tools it is now possible to identify the agents that would cause the swing to reverse. Breaking that is very tough, essentially once power is entrenched to that level it is unassailable. The DDR comes to mind, they had an iron grip on the populace and if they had access to today's tools I have no idea how any kind of dissident could even begin to get a movement off the ground before being identified and hauled off. The same for the various underground movements during WWII, they wouldn't have stood a chance.

China may well get there first, (if they aren't there yet) and we'll get to see how doable this is, I really fear the worst.

It looks to me like what's happening now isn't really anything new. In terms of "ratcheting" and the like, we've seen the same thing happen repeatedly throughout history. And, eventually, things do swing back. The only question is how long the cycle takes (it may very well take much longer than a single generation) and how much work is required to get there.

The famous paraphrase does (so far) appear to be correct: the arc of history bends towards justice.

I don't know if that's true. When you get to the point that you can pinpoint your prospective dissidents in their teens it becomes possible to contain any kind of potential source for a future revolution long before it is even visible to the participants. And given enough crimes on the books you could make them go away practically forever when using this sort of pre-crime detector, after all there is always something that will serve to put someone behind bars for a couple of decades in a properly organized dictatorship.

The next bad government to come to power in the developed world may well be the last.

It’ll swing, I worry about how violently it’ll swing.

Yes, this is the question. Historically, it can be very violent.

You can still buy battery-powered angle grinders for $130 at your local hardware store with cash.

All of these fancy AI systems rely on the sensors on the ground in your community (aka those solar panel + webcam pole combos randomly scattered about). It would take some teenagers on bikes one weekend to cripple a municipal minority report network. Sensors on major roadways are more difficult to tamper with, but not a whole lot of our land is a major roadway so that's not a huge deal.

Some even argue all of our modern infrastructure will fail on its own despite our best efforts. Complexity is a hell of a thing. A tiny bit of brute force applied might be all it takes in the end.

What I really want, is a small electronic device that can fool the gunshot detectors. I realize that the easiest way would be some kind of hopped up firecracker, but the use (even minute amounts) of explosives would probably be bad.

This device might only get 1 or 2 uses without recharge, that'd be ok. But I'd like for it to be light enough to be carried by drone. And if somehow they could be programmable to sound like different calibers, even better.

They could just go out on night missions, running the cops ragged.

There is a Gaston La Gaffe episode along those lines, where parking meters are equipped with fireworks when they expire they launch a rocket. This drives the local policemen bonkers because they are running from one end of the city to the other to try to keep up.

> Sensors on major roadways are more difficult to tamper with

The second amendment says no, not really.

If you’re suggesting people fire at small targets on the freeway with guns, then it has absolutely nothing to do with the 2nd amendment.

I'm pretty sure, the second amendment has something to do with the availability of guns in the United States.

Yes, but it doesn’t have anything to do with using them on public roads or for other illegal uses. Come on.

LOL... they're boiling the frog while we wait for things to get better.

Spoiler: we're the frog.

But you know that in the real world, that frog thing doesn't work, right? The frog will, in fact, get out when the water gets too hot no matter how slowly the heat increases.

Sure, unless it's lobotomized. Feel free to write your own joke about the current cellphone-addicted state of the world compared to a lobotomized population.

Every generation has been accused of being lobotomized by one thing or another.

Time to re-watch minority report

We are getting closer to Pre-crime Units, with each passing day. Except that instead of precogs (mutant humans), we will replace them with the almighty infallible AI.

"Sorry, Joe, but our AI says you will commit a crime sometime in the near future. So we have decided to charge, convict, and sentence you to 10 years."

To me this smacks of guilty until proven innocent. AI or any automated enforcement assumes checks everyone all the time. Put simply data like this should never be collected in the first place as that's the only way it will never be abused.

Clam down. No one is making an accusation of guilt, they’re just keeping an eye on an individual the way you might keep an eye on someone pacing around erratically. In most cases, they know it could mean nothing.

No, I disagree. It's the fact that they do check everyone. The same logic can be applied to automated speed traps and red light cameras. (Which do assume guilt)

The only difference here is a ticket isn't sent in the mail or an officer showing up at the door.

Could you please stop breaking the site guidelines? We eventually have to ban accounts that keep doing so, and we've had to warn you multiple times already (https://news.ycombinator.com/item?id=35558518).

Your comment would be fine without that first bit.

https://news.ycombinator.com/newsguidelines.html

IMO there should be strict controls on who can access this data and for what, and how long data can be retained, but given those limits, this is exactly the type of thing we should be doing to catch criminals.

Criminals like mothers who buy abortion pills for their daughters?

Minority report was a great prediction

It is getting clearer by the day that humans will use AI for laws written for humans. And this will cause a lot of grief for the people on whom these laws are imposed.

The title is ominous but we have had speed cameras for years and I can guarantee that if I do 15 over the limit I’ll still get passed by someone doing 25 more.

Almost every article about this has glossed over the fact that probable cause for the vehicle search was a K-9 unit alerting on the car, not the traffic AI.

You’re glossing over the fact that the dogs are trained to respond to the officers commands, and the officers can make a trained K-9 alert on any car.

Essentially parallel construction. Know their perp from other means, make dog alert on it, continue to abuse your power.

Stopping people due to AI reports and having dogs smell them is kind of the problem everyone is concerned about.

He wasn't stopped due to AI. He was observed due to AI. He was stopped for speeding 15-20mph over the limit, illegally passing on the right, and changing lanes without signaling.

Right and the reason I wasn’t able to date Mary Sue in the 11th grade is because she had an urgent need to wash her hair.

Dog alerts are about as objective as UFO witnesses. Having trained multiple working dogs of multiple breeds, nonverbal communication between dog and human is very easy to establish.

Unfair take, UFO witnesses actually have a chance of being credible.

That’s even worse, since K9’s have failed every independent test ever done on them.

They alert when their owners want them to, and make a mockery of the 4th.

We are officially no better than china when it comes to mass surveillance now. We’re just a few years behind on deployment compared to Xi

https://archive.is/2tfCC

So, stay for longer at your dealer's house, occasionally bring over a cake, got it.

How many false positives?

In Soviet Russia, police chase your car and has to stop traffic to get you.

In Capitalist America, traffic stop AI chase your car and has police get you.

... hmm, not quite there :-\

Reply on news.ycombinator.com