As a programmer, I won't put myself into that thing. I'm also not happy to be an unwilling participant in this experiment. Given the state of technology and our capabilities to write bug-free software (Seriously, we can't even get things like buying airline tickets correct), I think it's way too early to approve those vehicles on city streets.
It seems pretty clear to me (if you've been to SF recently) that autonomous cars react to pedestrians quicker and are much more careful around pedestrians than human drivers.
You can certainly take the attitude "I will only agree with having autonomous cars on the street once they are provably 100000000x safer than a human" but I don't think that matches most people's view on what constitutes safety.
Also, that goes pretty strongly against the American ethos, where we let people make their own decisions in terms of what kind of things they buy, within reason.
We already know that SUVs are more dangerous for pedestrians, but I think most of us agree that making SUVs illegal for this reason would not be a good idea, and is against the American ethos.
> We already know that SUVs are more dangerous for pedestrians, but I think most of us agree that making SUVs illegal for this reason would not be a good idea, and is against the American ethos.
As a non-american that doesn't own a car it's not that unreasonable for me :) At least banning SUVs that are excessively unsafe for pedestrians in a collision. Or maybe just ban it in cities where some walkability is expected. I just don't want to die by getting hit by a car
As another programmer, and one with decades of experience building software running critical infrastructure, I have a very different view.
You seem to be expecting perfection, or a reasonable facsimile of it, but that's not the bar that's being set by the existing solution (people).
Using your example of even a pedestrian task such as buying airline tickets, humans generally have a worse error rate than a computer doing it.
Source: I live in the Philippines where manual handling is still de rigeur for everything from buying ferry tickets, to immigration paperwork, to car registration and driving licenses. The error rate is far, far higher than the automated systems that do these things in other countries.
Similarly, humans (in aggregate) are really not very good at driving cars safely. Software on the other hand, is only getting better at it as time goes on. It's perhaps debatable whether computers are currently better (again, in aggregate) than humans, but with the current state of the art my view is that they are.
I'd welcome these things with open arms in the Philippines compared to the average driver on the road. Or equally so in my home country of Australia, or the US, or anywhere where humans kill each other every day in fast-moving steel cages.
15 years ago I worked on robot tanks for the US Army, I currently am in machine learning and I live in Phoenix, so I was very interested in this and have taken several rides on Waymo (never been in a Cruise). The Waymo felt very safe- and is definitely trying to convince you of that fact. Most seats have a screen that shows the cars current situational awareness of all obstacles around it- you can even see pedestrians, parked cars, etc., and when you come up to a stop light it will display a little icon with the correct light so you know that it is reading the red/yellow/green correctly. It plans routes such that it rarely takes an unprotected left turn (though the current generation of vehicles seem to be more willing to do that- on the old Pacifica minivans I don't think I ever had one do that, the iJag's have done it occasionally though not often). They are clearly thinking about trying to make it feel safer, and convince riders of that safety.
Now, I will note that the Phoenix area is by far the easiest driving I have ever encountered in 25 years of having a license. Wide roads, little traffic, few pedestrians (I moved here from NYC!), little weather, and I don't see them out as much at night. They crucially also completely avoid the most unpredictable driving in the area, for example Sky Harbor Airport, and near construction sites (there is construction in our kids school parking lot, and the Waymo won't go in- they will only drop us off next door and make us walk). But this too builds my confidence in the system- the people who are operating it seem to understand that it has limitations, which suggests to me that they are wearing their engineers hats and not management or, God forbid, salesmen hats. Is it perfect, I'm certain not. But I definitely prefer it to Uber or anything like that- no tip, no worry about the driver doing something concerning, just gets me where I want to go.
It's significantly more absurd to permit highly distractable, slow-reacting meatbags to operate the multi-ton high-velocity death barges in the close vicinity of pedestrians and other meatbags-cum-death-barge-pilots.
"""The National Highway Traffic Safety Administration has released its latest projections for traffic fatalities in 2022, estimating that 42,795 people died in motor vehicle traffic crashes. """
"""The number of vehicle-miles traveled on all roads in the United States decreased by some 1.55 percent to approximately 3.17 trillion in 2022. "
Ignoring some important details, that means americans drive 75 million miles before a person is killed. That seems .... awfully damn good to me.
The US is #5 (of 26 countries for which the statistic is available) in fatalities / km. That's only "awfully damn good to me" if your goal is to kill more people.
What does our relative position amongst a group of countries matter to my argument? The best countries see 150M miles between traffic fatalities. If we roll out self-driving cars and they continue to work as well as they have so far, the US numbers will get better.
Why ignore the other columns? Is it just because that one statistic leaves out most undeveloped and developing countries? Having lived in China before, not one of the 26, I know traffic can be much more fatal than in the states.
Cars are dramatically safer for passengers than they used to be. Miles per death isn’t reflective of the skill/attentiveness levels of drivers.
As the sibling comment points out, miles driven per death for pedestrians isn’t necessarily being improved by the mass and visibility and safety factors.
It has deaths/mile, in that they provide deaths, and we know the miles.
I do not disagree the numbers are trending up (although, I believe rates/miles driven might show a different story, but I didn't grab the miles driven for years 1975-2021 to normalize).
Americans drive 400 MILLION miles before a pedestrian death.
Let that sink in. 400 million miles per ped death. That's just absolutely gob-smacking amazingly good for something involving direct human control of an enormous ton+ vehicle.
If a couple drives 10,000 miles each for 50 years, that’s 1 million miles. In a neighborhood with 75 houses, each inhabited by such a couple, 1 person will die from driving in those 50 years. And when they die, they usually had many more years to live otherwise. That’s to say nothing of injuries.
I believe my reported numbers include pedestrian fatalities. I would love to have better sources that made that clearer, as these are markes as "traffic fatalities"
Why? We have what, like 50+ years of certainty as to how dangerous people are as drivers? They’re a known quantity and we know we can keep making cars safer to reduce the danger to drivers and passengers.
Instead we’re going to YOLO a half-baked tech-bro fantasy onto the roads and hope for the best?
Also “meatbags” has the same tone as “sportsball” imo.
> The data so far is clear that autonomous vehicles are safer, and it isn't close.
The data so far on fatality rates is near nonexistent. There' a fatality every 75 million miles in the US with a human at the wheel, with significant proportions of those fatalities considered to be humans operating so far below the threshold for acceptable human driving we jail them for it. Waymo hit 1m autonomous miles this year.
I’ve seen many people in this thread insinuate driverless cars are as safe as human drivers despite the vehicles being subjected to a subset of conditions the average US driver is. I’m surprised to see such disingenuous statements on here.
But we absolutely don't keep making cars, or trucks, safer for pedestrians: We are actively putting people in vehicles with worse visibility, because said human drivers feel safer when they are driving in a larger, taller heavier vehicle that is more dangerous to others.
The argument is not that self-driving cars are very safe, but that we tolerate risks that are quite large when it comes to humans doing the driving. Risks so high that it might not take all that much for entire categories of drivers to be less safe than a computer.
If we need 50+ years of certainty for self driving cars, we will never have said certainty, because we'd not let them drive, ever. And yet, we have very old people, those that are easily distracted, and people who often drive impaired in the road all the time. We also put them very close to places where we have pedestrians, and let them drive with huge differentials over said pedestrians.
We aren't really making the cars safe enough as it is, and in the US we lack any practical roadmaps to make them safer, other than, 'never walk, and drive in an increasingly bigger car'. We are YOLOing every day, with large, single passenger trucks that will strike a pedestrian at chest height.
So yes, self driving research is a better way out than hoping to change human drivers, whose hardware and software are hard to upgrade. I won't necessarily assume a Waymo car is better than most drivers today, but I'd already trust it more than some that have licenses today.
Right, we have 50+ years where motor vehicle involved deaths are like 15-20% of all deaths for people under 50, not to mention how many more people are permanently injured due to motorists.
The status quo is really bad and we are doing a terrible job of making them safer for anyone besides the occupants of the vehicle causing a collision. So yeah, we should really focus on pushing technology that can reduce the danger of distractable drivers. That mostly should mean investing in public transit, walkability and speed-limited small vehicles like bikes or golf carts. But the bar for driverless cars is basically: do they speed? Do they drive on sidewalks? Do they kill a few people a day? No? Then they’re better than the status quo, because the status quo is awful.
They're a known quantity of death though. Human drivers are going to drive drunk, sleepy and distracted. We've been at this since 2005 with the DARPA Grand Challenge. it's not half-baked, it's not a "tech bro fantasy", and hope is not a strategy. It was, a decade and eight years ago when the shit barely made it across the finish line (and many did not), but in those 18 years, theres been some development work and some money invested in making it work. In those 18 years, some 180k people have been killed by drunk drivers.
There are growing pains, absolutely, but at 2:15 am, crossing the street next to the bars, which driver with a red light are you going to step out in front of, crossing the street.
I don't agree with your characterization of what's happening as "YOLO" and "half-baked". It's taken many many people many years to get to where we are now. It's pretty easy to do a miles driven and accidents caused comparison. Self driving cars appear to be wildly safer thus far.
What would make you happy with a self-driving safety record? Why not be excited about the future?
I understand when some folks think that the focus would be better spent on public transportation, but you just seem like a hater.
I think at least part of the issue is that most people view other drivers as `highly distractable, slow-reacting meatbags`, but themselves as responsible, reasonable, highly capable drivers.
Interestingly missing from these discussions is: What does the distribution of accidents look like across the population of drivers?
It's possible to introduce an automated driver that is better than an average driver, but if the only people that use it are above-average drivers then you could end up with more accidents.
Yeah it's a good thing that cities aren't filled with retail locations that sell liquids that, if ingested, turn humans into bad drivers that think they're great drivers. And it's a good thing that those hypothetical liquids don't form debilitating addictions that make the humans want them more.
Rebuttal to what? You accused them of being a tech cultist and then started talking nonsense about terrorist attacks that are already performed with cars and trucks and would probably be harder to do in a self-driving world, as well as fear-mongering about "hacking" when all modern cars can be hacked by ATPs anyway, and they're the only ones who'd be capable of pulling off something like what you're talking about. Nonsense doesn't require a rebuttal and neither do insults.
There's two sides ..either you're all for them or on the fence or against. Im on the fence with trusting Waymo more then Cruise as Waymo's been at it for more then a decade but am concerned about all things I noted.
I did not insult anyone it was tongue and cheek. As well i was just responding to the drinking and driving analogy which my argument and that are two different ones.
Plenty of cars today are already drive by wire and connected. Cars have had computer controlled cruise for a decade or more. There's absolutely no reason to assume the security of the car's systems is impacted by its ability to drive itself.
That was one car. The thing with hacking is that it's incredibly scalable. Imagine a million cars going haywire at the same time. It could be worse than being hit by a nuke, and it could be dropped by a single talented extremist.
I ride a motorcycle. I'll trust a machine to check its mirrors. I won't trust a person. I'll ride ten miles and see twenty people on their phones, holding a cup of coffee, turning around to look at their kids, whatever. I get cut off by people who don't look both ways. I get cut off by people who do look both ways and think they will be out of my way before I get close to them. I get cut off by people making turns that aren't in a lane that allows turns.
I'm not afraid of self driving cars because I know that even if they're not perfect, they will still be far better than the overwhelming majority of drivers on the road.
Bicyclist here. The biggest danger to me are trucks and buses that swing around corners without visible signals. Electrics can slap more cameras and enforce blind spots.
But the biggest of all — which this article repeatedly notes — is enforced speed limits. We could have that on “human cars” of course with GPS speed limiters, but that is of course politically impossible despite there not even being a coherent argument for why cars that can move faster than 75mph are legal to sell.
Drivers can pick and choose which laws to abide (see that incident just last week where David Simon made the case that speeding in school zones is just find); ultimately, a computer can be forced to do what it’s told. https://hellgatenyc.com/david-simon-got-a-speeding-ticket-in...
Nobody expects self-driving cars to be bug-free-- why would we?
All that matters is, statistically, that the cars are better than humans at driving. And that's something that will eventually be achieved through a combination of software engineering, hardware engineering, and spending lots of money to build trust. Some number of people will still die in accidents even if the entire fleet is self-driving.
You might, if you do not already, find a like-minded crowd on the Risks mailing list, where they argue about 0.0000001% risks like "streetlights that turn green automatically when emergency response vehicles approach the intersection, fail at some rate because the designer didn't anticipate two emergency vehicles approaching the intersection at right angles" (which very very rarely causes an accident).
I'm a software engineer and I put myself into that thing. Didn't die. AMA
But last week, a block from home, this old lady killed a 4 year old in this city. Perhaps the little girl was a consenting participant to this, but I am quite doubtful.
I'm with you. That's why, as a house builder, I'll never get into an aeroplane. I mean bricks are way too heavy to fly, the concrete will crack at the first sign of turbulence and how far can they really go before the mains power cables get cut? It makes no sense. At least, unless they're doing things differently.
This thing has a statistical safety rate. Which is either worse or better than your average Uber/Lyft/taxi driver.
But if it's better, it's better. And if you'd rather keep getting in vehicles with human drivers then they're statistically more dangerous, well that'll be your choice.
Buying airline tickets usually works fine, but I'm going to visit a client in upstate New York and a business class ticket BDA->LAX->SYR was cheaper than BDA->PHL.
In GA, yeah. On commercial passenger jets with a couple well-trained pilots and a multiple-times lower fatality rate per hour than cars, there's no safer place to be if you need to cross the continental US. I doubt a self driving car is anywhere near as safe as that, especially considering how unlikely violent crime is to get anywhere on a US domestic flight, for obvious reasons.
