Hacker News Reader

Comment by jqpabc123 | Hacker News Reader

Promoters claim self driving cars are safer than human drivers.

This is based on a rosy interpretation that deflects the blame for most of the accidents they are involved in.

Self driving cars are in accidents much more frequently than human drivers. However, most of these accidents are rear end collisions --- naturally blamed on the human driver.

But the reason human drivers rear end self driving cars is the fact that these cars are programmed to hit the brakes any time they can't comprehend the current situation --- which apparently happens quite a bit.

The reality is, human drivers can't comprehend or anticipate a driverless car hitting the brakes for no apparent reason.

View on HN

Youre citing old data. Autonomous cars are currently safer than human drivers when measured by accidents per miles driven:

> Accidents per million miles driven for Waymo self-driving cars are 0.59, compared to the general U.S. rate of 2.98.

https://blog.gitnux.com/self-driving-cars-safety-statistics/

Furthermore, we should expect a constant stream of news headlines about self-driving cars being involved in accidents, even when self-driving cars are safer than humans because they are newsworthy and interesting. No one writes "dog bites man" articles, they just cover "man bites dog" stories.

This is a bad faith comparison not using apples to apples. Humans drive in all sorts of conditions that self driving cars flat out don't, and this data doesn't control for that at all.

Exactly. Wake me up when the ice road trucks (half joking) are self-driving.

Given the right away privileges ice road trucks are given (vs. normal cars...since they can't stop easily!), it seems like this would be a better first application of self driving than more chaotic environments.

It's fine to point out the flaws in this data, but do you have any better data to point to?

The humans five times worse in that data. That is a very large gap. Even after adjusting for driving conditions, with a gap that large, self-driving cars are likely safer.

Also, don't assume a comparison is made in bad faith, just because it has flaws.

A comparison like this from any trained statistician would be treated as borderline fraud, and is akin to the "we proved cigarettes are safe to smoke" studies done in the 50-60's.

So to be clear, this is not a "flaw". This is a catastrophic, wrong from the get go, not mentioned and no effort made to control for, failure.

Deliberately misrepresenting data, or in the most charitable case, by sheer mass incompetence, when it involves the safety of people is never ok.

The fact that automated vehicles operate with much broader safety margins with regard to driving conditions definitely contributes to them being safer overall - humans choosing to drive in worse conditions objectively makes them more dangerous drivers. I also suspect that what others are getting at - that under the same driving conditions safety margins as robocars, humans have a lower accident rate - isn't true, considering the bulk of motor vehicle accidents occur during normal commutes during daylight hours in nominal weather at low speeds. I suspect that even if you control for the elderly, the drunk, and inclement weather, humans will still have worse statistics. I think that humans have a tolerance for risk and ambiguity that makes for statistically worse, albeit more predictable, driving behaviour. But much like you, I don't have any better data to back this up.

> Humans drive in all sorts of conditions that self driving cars flat out don't

that's exactly why they are not allowed in all sorts of conditions! Where did this all-or-nothing demand come from? If a self-driving car shows less accidents than humans in Phoenix, why would you apply numbers from e.g. Montana, how is that apples-to-apples? and if it's shown safe in Phoenix conditions, why not let it run in Phoenix, just like the current approach?

Then you don't get to argue by them being safer overall. That argument then amounts to lie when the actual topic is general driving ability.

Second, then use only data about human drivers in perfect situations to compare. Their accident rates also go down.

Waymo had 20 accidents by February of this year but categorized 18 of them as "minor collision events".

Unless the authors similarly recategorized the driver data that is suspect.

For instance 1.24 is the number of injuries per million miles for drivers so that doesn't sound like 2.98 excludes minor ones.

The reality is the jury is still out on self driving car safety. Driving a few million miles and extrapolating isn't terribly useful especially when you only drive at low speeds.

There is a hypothetical potential to reduce fatalities but it is still very hypothetical.

I actually looked at the website you linked and frankly I don't think you did. Surprise: The website linked doesn't cite any data.

Here's what it does do:

- Links to: "ZipDo, cited June 2023: Self Driving Cars Safety Statistics". which is NOT a study, it's just a carbon copy of this website

- Links to website landing pages, like "https://www..deloitte.com"

- States "Statistics researched: 18", but never tells us what those were.

- Finally it has the most inane clearly ChatGPT written commentary:

  > This statistic is a testament to the safety of Waymo self-driving cars

  > This statistic is a powerful testament to the safety benefits of Tesla Autopilot technology

  > This statistic is a powerful reminder of the potential life-saving impact that self-driving cars could have on our roads.

  > This statistic is a stark reminder of the public’s apprehension

  > This statistic is a powerful indicator of the potential safety

  > This statistic is a powerful indicator of the public’s confidence

  > This statistic is a testament to the progress being made

  > This statistic is a testament to the safety of Waymo’s autonomous vehicles

And so on.

Why wouldn't you even read the website you linked?

Oh well, at least you can take a free "Personality Leadership Test" to see if you're a stable diffusion version of Mark Zuckerberg, Bill Gates, or Elon Musk.

Waymo is not Cruise.

It’s only going to get worse as the average human age skyrockets.

“ A pedestrian crossing a San Francisco street on Monday night was hospitalized in critical condition after she was hit by two cars — first a regular vehicle which hurled her into the path of a driverless taxi that then ran her over, stopping on top of her as she screamed in pain, according to witnesses, investigators and the autonomous taxi company.”

It’s a little difficult from the description to know what a human would have done in the second car. Can’t tell how much time there was to stop.

Also, when driving behind someone you should always be prepared to stop. It doesn’t matter why the car in front stops.

It's fair to argue that a human driver may well have hit her. It's less obvious that they would have parked on top of her until someone came to lever the car off of her.

Not that also running her over with the rear wheels looks like a good choice ...

You'd keep going and roll over her a second time?

No, you'd get out, realize what was happening, and get back in the car and put it in neutral or otherwise gently move it 6" to stop crushing the person.

Aren’t you not meant to do that until paramedics arrive in case the removal of pressure from the injuries makes the situation worse?

I think in a scenario where this just happened you'd want to remove the pressure immediately. The pressure is going to lead to lower or no blood flow to the limb, which will greatly increase the chance of losing it.

The article mentions that police asked them to leave the car in place until the FD arrived.

They had to drive the car remotely, right? That wouldn't have 10% of the finesse of a human in the front seat with a hydraulic brake pedal. But I wasn't there so just speculation.

What surprises me is that people didn't lift the car off her. 3-4 adults should be able to lift even an EV a few inches.

Just removing the object crushing a person could result in their death. You need medical professionals on hand to ensure that removing the object doesn't lead to untreated issues.

The article is pretty clear that a human driver did hit the victim.

> Self driving cars are in accidents much more frequently than human drivers

Source? (I’ll believe it for Cruise.)

> human drivers can't comprehend or anticipate a driverless car hitting the brakes for no apparent reason

Human drivers tailgate. If you have a safe cushion in front of you, an erratic stop won’t cause a collision.

Yeah, the tailgating makes no sense unless preparing to pass. When I catch myself doing it it's because I'm emotional and feeling rushed, and I take a breath and remind myself to back off until it's safer to pass.

During a freeway slowdown I try to dampen the brake-waves by going slow and not braking so much, but if I leave too much space in front of me (for coasting), another driver moves in and we're back to playing accordion.

If all cars could sync up and move together like train cars on the freeway there'd be less brake dust and more efficient travel, but then I'd rather take a train anyway.

Heroic. I try to do this too but everyone I know thinks I'm being neurotic when I explain it to them. I still see so many people accelerating into stoplights/brake scenarios.

I think it should be seen the other way around. Why should I have to stop moving, meaning using my brakes and subsequently the gas, just because the other drivers wanted to rush up to the light and wait? It's much easier for everyone if we do this smoothly

All those people you caused to miss a left turn light because you didn't feel the need to clear the turn-lane entry, because you wanted to keep slow-rolling, are cursing you. I have had that exaqct situation more times that I can count and every time I get angry at the other drive who thought "Why should I have to stop moving".

> I still see so many people accelerating into stoplights/brake scenarios.

Every time I have to use the brake - rather than downshifting - for anything other than holding the car in place, makes me feel like I've committed a failure in planning and/or situational awareness.

My driver’s ed instructor explicitly framed it like this in class, and it’s made me a much better freeway driver.

> but if I leave too much space in front of me (for coasting), another driver moves in and we're back to playing accordion.

stop caring about this and just slowly back off a little more.

someone else will eventually leave your lane in front of you and make up for it.

>>> Source? <<<

National Law Review (look it up or see associated link below).

>>> Human drivers tailgate. If you have a safe cushion in front of you, an erratic stop won’t cause a rear-end collision. <<<

That's the theory --- and the misinterpretation.

It's true that almost all human driver's tailgate at some point --- but they do so while being involved in half the accidents that self driving cars are --- 4.1 accidents per million miles versus 9.1 for driverless.

https://carsurance.net/insights/self-driving-car-statistics/

> National Law Review (look it up or see associated link below)

Your Carsurance source cites the National Law Review, which cites the Carsurance source for that statistic [1]. It’s important to consider from when they polled data and how they’re segregating by manufacturer.

EDIT: Found the study [2]. It’s for 2012 to 2015. Given these are decade-old data, I’m actually surprised the gap is only 2x while fatality and injury rates were already lower.

[1] https://www.natlawreview.com/article/dangers-driverless-cars

[2] http://websites.umich.edu/~umtriswt/PDF/UMTRI-2015-34.pdf

> Human drivers tailgate. If you have a safe cushion in front of you, an erratic stop won’t cause a collision.

This is why in my state, if you rear-end someone then you are automatically presumed to be at fault. "They slammed on the brakes unexpectedly" is not a defense. You should have left adequate braking distance so that wouldn't cause an accident.

Same in Europe

Downvoters need to stop tailgating.

I’m genuinely surprised “human drivers can't comprehend or anticipate a driverless car hitting the brakes for no apparent reason” is an argument people can relate to.

You absolutely should be expecting the driver in front to slam on the brakes at full force without warning. There’s an infinite number of things that could cause any driver - human or autonomous - to do this. If you drive without this expectation then you’re a very dangerous driver and I wouldn’t want to be anywhere near you on the road.

There seems to be a cabal of downvoters who, I assume, truly believe they are entitled to drive like complete jackasses.

A lot of drivers tend to drive like assholes so this isn't too surprising.

It's fault of the person who didn't have enough room to stop. Full stop.

It doesn't change the fact though that human drivers don't slam the brakes out of the blue because their "AI" got confused about nothing in particular, and that it is dangerous.

Slamming brakes for not reason IS being complete jackass. And if you do it frequently enough, you have no business being on road.

You are conflating “unexpectedly” with “no reason”. The parent comments said “unexpectedly.”

There is a world of difference in those word choices. Saying “no reason” certainly gives you an excuse to write a sanctimonious comment like this, but it’s misrepresenting what was said.

A common insurance fraud trick is to change lanes close in front of someone and then slam on your brakes to cause a rear-end collison, then file an injury claim for back or neck pain against the other driver. They don't have time to react but are presumed at fault. One of many reasons to have a dash camera in your car.

That doesn’t really have anything to do with what we’re talking about, but many of those types of accidents can also be avoided.

There’s a significant number of people who feel personally attacked by a car merging in front of them, despite the fact that a merging vehicle has right of way (in all countries I’ve ever driven in). There’s a video that keeps popping up on my Instagram where a car changes lanes, and the car behind has plenty of time to react and slow down slightly, but they instead crash into the merging vehicle intentionally without touching the brakes at all. Most of the comments side with the driver who caused the incident, not the merging vehicle who did nothing wrong.

Anyway, my point is, there are a lot of people who refuse to give space to merging vehicles, which increases the chance of these accidents happening (fraud or not).

If you put your ego aside and immediately back off when a vehicle in front merges, you’ll be far more likely to avoid this type of accident. Alternatively, if you tailgate a merging vehicle out of “principle” you’re just inviting an accident.

> However, most of these accidents are rear end collisions --- naturally blamed on the human driver.

Rear end accidents are practically always the fault of the person who didn't stop in time. If someone rear ends a driverless car, even if it stopped because of some glitch, that driver is at fault. They shouldn't be driving around city streets with such speeds and limited following distance that they couldn't stop in time.

Exactly. The car in front might be reacting to some road hazard that just appeared that is not visible from behind it or it may have some mechanical failure. It doesn't matter why it comes to a stop, it is always and all times the responsibility of the trailing car to maintain enough distance to be able to come to a complete stop given the speed and road conditions.

It’s easy to say that but you can’t just ignore the reality that this is how people drive, and self driving cars in that environment do lead to more crashes.

I'd argue we should bar people from driving who collide with stopped vehicles without any other extenuating circumstances. Maybe adding self-driving cars to the mix will increase these collisions for a few years, but if we bar people from driving for doing these kinds of accidents we'll quickly weed out these massively unsafe drivers from the driving population and be much safer overall. Maybe we'll finally start decreasing the pedestrian fatality rate instead of increasing it.

If you refuse to drive safely you have no place to be on the road. We shouldn't just accept people driving in massively unsafe fashion.

I wouldn't mind this but it's completely politically unfeasible. There is absolutely no chance of political support for banning people from driving for anything other than a DUI. Even "accidentally" killing people or being so old that you can't see enjoys wide support of still being allowed to drive. As long as that is true self driving cars need to be able to handle bad drivers.

> Promoters claim self driving cars are safer than human drivers.

And opposers always ignore or brush off the fact that humans have proved themselves to be terrible drivers. A human driver ran over this woman first yet you didn’t even recognise this anywhere in your argument against autonomous vehicles.

Everyone is saying "is AI better" not saying humans are good.

As terrible as humans are we do have a very important trait. Being able to handle unexpected information efficiently.

All the current AI models do have ways of slightly generalizing but only slightly. You still need to effectively train "when there are cones you need to switch lanes" and "when there are barriers you can drive on the line" kinds of things.

And while yes everyone mostly talks about low stakes examples that is mostly because the high stakes ones haven't been handled by AI yet.

E.g. a Tesla driving under a semi because it thought it was a billboard.

I look forward to any replacement to human drivers (although eliminating cars from Metro areas would be much more efficient I digress..) but autonomous vehicles need to be better before they can be that.

>As terrible as humans are we do have a very important trait. Being able to handle unexpected information efficiently.

You’re joking right?

The human driver didn’t handle the “unexpected information” of a human being on the road when the light was green.

The human driver didn’t handle the “unexpected information” of running over a human being and drove off instead.

Humans are awful at unexpected information. This is exactly the moment that emotions take over and emotions are what makes humans really bad at tasks like driving.

Computers are worse.

The only way we have gotten any semblance of safety is "slam on breaks if unrecognized" which happens quite frequently.

> E.g. a Tesla driving under a semi because it thought it was a billboard.

People don’t like what they can’t understand. Most of the accidents caused by autonomous vehicles are weird and probably would have been avoided by human drivers. However, this isn’t the whole picture, because those same vehicles also avoided a bunch of accidents that humans would have caused.

My interpretation of the data is that autonomous vehicles are already statistically safer than human drivers.

Problem is, humans are emotional beings who aren’t always capable of pure rationale. If you tell someone they’re statistically safer in a Tesla, but it might accelerate into a wall at 60mph, many people won’t take that risk.

It's rational to trust yourself over corporations that want to make money, governments that sit in their pockets and their naive enthusiasts.

Are they actually safer though?

There hasn't been a ton of data especially in areas where the risk is important.

> As terrible as humans are we do have a very important trait. Being able to handle unexpected information efficiently.

At least half the people out there are terrible at this. We have expression like "deer in the headlights" to describe how they behave.

A Tesla drove under a semi because "that flat thing isn't moving it can't be a car it must be a billboard".

A jay walker in Arizona got ran over because "people are only in crosswalks" (I don't remember the company offhand)

Autonomous driving isn't automatically safer. Companies keep focusing on the potential for improved safety without technically claiming they are that much safer.

The potential is great but that means nothing. What matters is what we make.

Wrong way driver in a truck at a high rate of speed killed two 18 year-olds earlier this year:

https://westseattleblog.com/2023/03/crash-blocks-westbound-w...

Friend of mine just witnessed a crash the other day where a driver at a high rate of speed lost it on a wet corner and the passenger got ejected.

AI has driven 44 million miles so far.

Americans drove 3.2 million million miles (aka trillion).

Comparing things that have managed in total barely 1/100,000th of a years usage means anecdotes don't matter.

Because there is at minimum one unavoidable aspect here: we don't know what accidents an AI could prevent and way way more importantly we don't have enough data on what accidents an AI would cause at highway speeds.

Insufficient data is available for how effective AI is at preventing accidents.

The Arizona one I gave for instance, the company was found not liable since a person might have done the same but could they have slammed on the breaks turning a fatality into an injury? We will never know and can never no.

The best we can do is measure real world performance and pay close attention.

> anecdotes don't matter.

exactly. you posted anecdotes.

Humans are overall actually good drivers, better then AI drivers currently.

Folks reading this comment should scroll down in the thread to see that the data source for this claim is not useful, being from roughly a decade ago.

Every driver should be prepared to hit the breaks at any given moment with enough room not to hit whatever is in front of them. Period.

The fact that so many cannot is a great example of how driverless vehicles will easily surpass human drivers in the not-too-distant-future.

Swiss Re's opinion is that self-driving cars are safer -> https://arxiv.org/abs/2309.01206

Summary:https://jdsemrau.substack.com/p/paper-review-comparative-saf...

It's specifically Waymo.

Google generally seems to be much more safety oriented than Cruise. The Cruise experiment should just be stopped at this point.

In the US, it’s always considered the fault of the person who read ends someone.

Not always - if you have dashcam footage that proves brake checking or other poor behavior, it can be decided in the rear car’s favor.

Don't brake check someone and rely on this as you will end up partially at fault for creating a dangerous situation.

But braking for safety is always allowed and the person behind you is supposed to leave room to allow that.

> In the US, it’s always considered the fault of the person who read ends someone.

This is simply false. Brake checking is explicitly a traffic offense in some states, and can get you cited under something like reckless driving in some other states. Your insurance company is not obligated to find the other driver at fault if you're cited for brake checking.

Yes I’m sure that driverless cars are purposefully trying to cause rear end accidents by brake checking.

Your reply is totally relevant in the context of someone blaming driverless cars for stopping when it doesn’t know what to do.

First, post something false, prepare to be corrected. "In the US, it’s always considered the fault of the person who read ends someone" just isn't how things work.

Second, I'd expect Cruze to at least have civil liability if their cars glitch and brake spontaneously, resulting in an accident. Since spontaneous braking is something autonomous vehicles have struggled with (Tesla at least), yes, it's relevant.

I've been brake checked by a cruise car in SF. I had a dash cam and had I hit it I'm confident I would have been cleared and the robo car that stopped suddenly randomly in the middle of traffic would have been at fault.

They literally have signs on them that say "warning, stops suddenly" , not a joke, literally

This is not true at all

So if you are waiting at a stop sign, red light, and someone crashes onto you from behind is it your fault by law? Hard to believe

They said the person that rear ends someone is at fault, which in your scenario is the person who is who crashes into the waiting person from behind.

No, the opposite. That's you getting rear-ended, not you rear-ending someone.

Reply on news.ycombinator.com