Are self-driving cars already safer than human drivers?
arstechnica.com
arstechnica.com
1) It's very important to do an apples to apples to comparison when you're looking at accident rates, and you can't just do that by looking at averages for the US for all miles traveled. There have been issues with numbers for Tesla autopilot before where it looked like autopilot was safer than it was because Tesla compare autopilot against the average accident rates for all miles driven but in reality for the types of situations where autopilot was used, human drivers actually had fewer accidents on average.
2) Waymo has a low accident rate because its vehicles are operating very conservatively right now. If it could scale up and still operate the same way, that wouldn't be an issue, but its vehicles are already causing problems by getting stuck which suggests that it wouldn't be practical for them to operate a significantly larger number of vehicles the way they are operating right now. To scale up, they would need to drive less conservatively, and in that case they very likely would not be safe in their current state. IMO, this means it isn't really meaningful to say that they're "already safer" than human drivers.
They would likely have to setup in a Northern location and run their cars in that location for a while. Also need to update their simulations to account for snow and icy roads. So more complex driving models and more challenging road detection.
Sometimes there is light snow over the whole road and there are no markings. You just have to position yourself relative to street signs or ditches where you think your lane is.
Autonomous vehicles are good at making routine repeatable decisions based on clear data. I'm not sure how we can sense and input this sort of data clearly enough.
I think weather conditions can also predict road conditions pretty well. And the combine that with historical data on where things were slippery, I bet you can create a pretty good model of what conditions to expect across a driving environment.
Although in Canada, sometimes all roads in a city are just horrid - dozens of cars in ditches or hitting each other because no one can stop. And maybe in those cases, autonomous vehicles should refuse to drive until it is improved.
When someone new moves into town, someone will typically explain to them to watch for where the sun hits the ground in winter. Where that happens on a steep slope, avoid driving there when there is snow. Before they know to do this, a human will attempt to climb that hill, not make progress because of the ice, back down, and try another route. This is a rural area without much traffic, so conditions might change significantly between "real-time" updates.
So, we have human drivers "zooming out" in a couple of ways... in time/context (observing conditions that lead to certain routes becoming high risk) and in problem resolution (try another route). This is quite different from what is currently being tried algorithmically.
LLMs are so good because they have almost the entire digitized text output of humanity available for training. They may not always provide factually correct information, but it is dished out in near perfect English (or whichever language you prefer).
Personally i think if we want to recreate a human driver full AGI is needed. But, in the right conditions current autonomous cars will work well. They just need some external helpers, like being geo-fenced to a set of roads which have all the correct sign posts and markings.
Not every human driver can solve every situation that the best human driver can solve. Also, the most dangerous situations happen at high speed, which humans aren't great at handling unless we've had a lot of training. The main reason I'm optimistic about autonomous cars is that a system like Waymo's will only get better and better over the years. And newly produced cars will be just as good as "experienced" cars.
This is a mantra from speed limit fans. There are a lot of dangerous situations happening at low speed (bicicle coming from left in an intersection, children playing )
...that makes the human driver, who normally would get say 10k miles of practice every year will drive... 500 miles on average.
It really does need to be "car can do all of the driving", or else it will be putting people that have zero practice in those hard situations it can't handle.
Unlike in SW, choosing a different path might lead to accidents.
My car can keep the lane very good in a sunny day, when the sin shines from above (i.e. midday).
In heavy rain or snow, i can rely only on my senses.
For an automatic driving car to be useful, it must function correctly in 100 % of the cases (ok, 99,999 %).
Not surprising.
> To scale up, they would need to drive less conservatively
Can you quantify this? I think that self-driving cars should drive conservatively. Human drivers who are aggressive or risk taking are idiots even if they get to where they are going on time.
Driverless cars should drive like public transport buses (at least around here) / school buses. Conservative, slower than you would like and safe.
For example, if they need to make a lane change for a turn on a congested street, they will wait for someone to let them in rather than “pushing” their way in like a human driver (especially a bus driver) would. When that inevitably doesn’t happen, they are forced to go around the block and try again.
That exact behavior happened during the “Uber” ride I took on a Waymo, and it is possible they have improved since then.
If you compare a traffic light intersection where there is a clear moment at which you should stop or continue with a roundabout where you need to make a judgement call; being overly conservative would mean stopping at the roundabout until there is 0 traffic.
There was a discussion on this in ep 454 of Freakonomics radio.
I agree, but the specifics get tricky. It is possible to drive so conservatively that it’s unpredictable and therefore dangerous.
I’ve known some terrible conservative drivers in my time; I was a passenger with someone who stopped abruptly in an intersection despite having a green light because they were afraid someone else was going to run the red (while it’s fair to say the person who rear-ended us was at fault, it was an accident that would not happen with a more normal driver).
Maybe I’m nit picking on the difference between conservative and unsurprising.
I would much rather see comparisons to safe/prudent drivers vs the average. Try to sell me autopilot as being as safe or marginally better than the average driver and I'll never use that shit - the average driver sucks.
https://www.npr.org/2022/12/13/1142598789/drivers-crashes-dr...
Would you trust a self driving vehicle to attempt to drive your family into safety?
If not, then you probably got some food for thought about averages.
I get what you’re saying though. I’ve been driving for 20 years, have had a single accident (rear ended at a stop light, not really my fault), no moving violations, and have driven about ~300,000 miles total (sounds like a lot but I was a full time lyft driver for a bit).
I’m positive I’m an outlier but I’d also trust myself a lot more than some of the current autonomous vehicles. I’d still assume with AI assistance or partial autonomy, I’d be safer on average though.
Yes, that is the paradigm unfolding here. That is indeed one of the many critical attractions of self driving vehicles.
>The reckless drivers will shun self driving because they want to show off and have fun.
First, this is your unsourced head canon. People with drinking problems, cognitive impairment (both temporary and permanent), etc seem likely to be a far bigger percentage.
But it doesn't matter, because they won't get the choice, which again is one of the core aspects of self driving vehicles. Right now, human driving in America is a quasi-right because it's currently intrinsically tied to "personal arbitrary time/space mechanized transportation". I say "quasi-right" because while legally it's not formally a codified right or in the Constitution or something, as a practical matter if most of the population wasn't able to drive the country would fall apart. And that is recognized by government, both executive/legislative and judicial. Losing driving rights would be a major burden for tens of millions people and would disrupt their ability to get or retain jobs, in turn causing other societal expense. So traditionally government is quite hesitant to wield that stick and almost never preemptively.
Self-driving though would decouple human driving from someone having their own car and ability to have it take them where they want, when they want like right now. That in turn would completely change to political dynamics. Driving would truly become a privilege, and getting a license could become much more like class C licenses or professional emergency responder licenses or the like are. Much, much more serious training and testing, continuing ed requirements as with other professions, and much easier to lose. Because unlike right now, losing one's drivers license won't mean losing the ability to use a car. Anyone who acts recklessly at all, even if there is no damage of any kind, can just preemptively have their license rescinded until at a minimum remediation is completed. Because driving will just be a hobby.
That long tail of outlier clipping works on an individual human basis, too. How many accidents happen because an individual careful driver is, momentaritly, inanttentive?
It depends. I doubt it because the distribution needs to be considered. Replacing average drivers either an average system yields no benefit. If it's applied to both tails equally, then you haven't moved the average either, juar made it more homogeneous. You'd have to target the highest risk groups to make the impact. Many in those groups can't afford or won't adopt the tech - the elderly, low income (or less educated depending on the study), or young males mostly.
The better ways to target only the problematic tail is stricter testing, better education, and possibly permemant interlock devices for DUI offenders.
"How many accidents happen because an individual careful driver is, momentaritly, inanttentive?"
My guess would be probably about as often as autonomous systems screw up in novel environments.
If other drivers would not yield to robocar, it may get stuck on intersection forever. Robocars have no assertiveness to push themselves into traffic.
Maybe making the point of comparison with professional drivers in fleet cars would be more accurate, but I doubt that data is available.
I raise this because the pool of people buying a Tesla is different than the full driver pool. There is likely serious sampling bias in this pool of drivers due to demographics, geographics, etc..
I've noticed a lot of automation legally still require a warm body to take the blame. Seems like eating your cake and having it too.
Waymo are driving, there is no warm body, in accidents where a Waymo self driving taxi struck somebody that's on Waymo, you're just a passenger the same as if you were in a New York cab, or a Uber, or if an A320 you're sat in smacks into a post because the pilots took a corner too close. You're not the one who needs insurance, because you're not driving, Waymo is.
Even Waymo still has some safety drivers per the article linked. And if we are going to constrain our analysis to self driving pilot programs, then we need to constrain the comparison to the same geofenced area. Arguably there are not a lot of fatal accidents per 1M miles on the local roads of Phoenix/SanFran by human drivers either.
Do you have any source for that? As far as I'm aware, there simply isn't enough public data released to actually make an unbiased comparison here. So whichever side of this you're on, you can always twist the data to confirm your point of view. Only when Tesla and all other carmakers begin to hand out all of their data, we will get a useful comparison.
Once they scale up, they will be encountering more other Waymo cars on the roads, eliminating a lot of the accidents where someone else crashed into Waymo, or made an upredictable move. So then they could afford to drive less conservatively.
We are pretty sure that our roads would be much safer if all cars were self-driving, but that cannot happen overnight and there will be an interim phase when robo-cars have to mix with human drivers, which is the hard engineering challenge.
If it's a worse driver than me but better than average; I'd still be safer driving myself.
I'd be safer if I could force the idiots into a self driving car.
What I think is most interesting about your way of looking at this is whether we'll find those sub-domains where autonomous driving makes sense versus where manual driving will remain safer. And what those domains will be.
[Grade of Automation 3 services don't need a human "driver" but they're not safe enough that we can just leave them to be used unattended like an elevator, there's a trained member of staff somewhere, probably helping tourists or checking tickets or something - the GoA 4 services are unattended, like an elevator there are humans but they're far away in a control center and you're safe enough while you wait for a human to come to where you are and fix whatever extraordinary problem the automation couldn't handle e.g. there's no electricity so you can't go anywhere.]
https://mleddy.blogspot.com/2023/08/goodbye-fine-arts-elevat...
"First, other vehicles ran into Waymos 28 times, compared to just four times a Waymo ran into another vehicle (and Waymo says its vehicle got cut off in two of these cases). Second, Waymo was only involved in three or four “serious” crashes, and none of them appear to have been Waymo’s fault."
This suggests that if we were to replace all humans with Waymo, at least in the nice temperate snow-less/ice-less cities they are currently driving in, accidents would possibly be cut by 4x.
Very interesting.
Also, replacing all human drivers with Waymos isn’t possible. Waymos can’t do everything drivers need to do and don’t operate in all conditions, in all places, at all times. Until they do, these stats are not really meaningful.
For example, it's quite common for an autonomous vehicle to slow down or stop in a way that disrupts traffic. This isn't "unsafe", per-se, and in fact it's probably the car's safety protocols functioning correctly.
But (depending how frequently it happens) it's still a strong argument against deploying these things on public roads.
We need vehicles that are not only safer than human drivers, but less disruptive and more competent overall.
In fact, as a pedestrian and cyclist, I could not care less.
I mean, if you don't use something, you don't care about downsides that apply to it. That's not a great argument, unless you also only want to convince other people who don't use cars to not use Waymos.
The safety benefit doesn't even have to factor in, or only somewhat. Part of the drive for speed with human driven vehicles is that for most of us it represents pure waste in our lives. The two hours I drive to a job isn't some uplifting activity, just robotic drudgery. Yeah I can load up a podcast or something and try to get something out of the time (at the cost of slightly more distraction perhaps!) but that's limited. So of course I'm motivated to have that be as short as possible. Shaving off 15 minutes on a trip made a hundred times a year will add up to weeks of my life back to me, although of course at a higher risk of losing my life entirely perhaps?
Whereas if the robotic aspect was taken care of by a computer instead, I'd be free to make far more use of that time. Or even just sleep, which is still useful. If the car takes an extra half hour, that's not necessarily a problem in the slightest because it's not a wasted half hour. Being conservative will be much easier purely from a self-interested perspective. Avoiding all the other issues is very good as well but I don't think it'd require some self-sacrifice either.
>In fact, as a pedestrian and cyclist, I could not care less.
Well, you should.
Oh right - those don't happen in California.
Yes, many places are harder to drive in than SF and Phoenix, but they're harder for humans too, and Waymo & co will continue to get better as they get more training data, better sensors, and bugs get fixed.
It does actually. You just have to go into the mountains.
The article is specifically about driving in SF. It also states SF is one of US more difficult cities to drive.
Self-driving cars don't have to drive in snowy conditions; indeed, they don't have to drive in any conditions at all. It'd be nice if they could though.
Self-driving cars are the foreigners. In isolation they may be safer and more predictable. But they have driving patterns that "the locals" simply don't. There are bound to be more bumps and cutoffs as these two groups interact, only at high speeds with tons of weight.
It may not matter how conscientious their algorithms are if the locals keep bumping into them or get continually inconvenienced. The locals aren't going anywhere, for better or for worse. FSD needs to drive like a local, only better. That's a much harder nut for FSD to crack.
>Waymo and Cruise have driven a combined total of 8 million driverless miles
For context, the average American drives about 13k miles per year, so this is about as much as much data as a village of a thousand people produces annually.
The actual experiment that would produce relevant data: A substantial mass of driverless cars on a random set of roads under random conditions, including with a high enough concentration that the cars need to interact with each other the way humans need to interact with other drivers. (obviously we know how this would end, which is why nobody attempts it)
>"Let’s say that autonomous cars slash overall traffic-fatality rates by half. So instead of 32,000 drivers, passengers, and pedestrians killed every year, robotic vehicles save 16,000 lives per year and prevent many more injuries.
>"But here’s the thing. Those 16,000 lives are unlikely to all be the same ones lost in an alternate world without robot cars. When we say autonomous cars can slash fatality rates by half, we really mean that they can save a net total of 16,000 lives a year: for example, saving 20,000 people but still being implicated in 4,000 new deaths."
[0]https://www.wired.com/2013/07/the-surprising-ethics-of-robot...
Then they say 1 fatal crash every 100 million miles.... And then make it sound like we have to have self driven cars drive 100 million miles before we can tell if they are safer.
That article is all over the place.
The national average is 600k miles. For FSD (which has over 300M miles driven) it is 3.2M miles between accidents.
Oh not humans. Americans. Most developed countries are much safer. America has found a lot of exciting ways to make this worse, whether its choosing more dangerous technology, not prioritising safety in design, or just obvious stuff like oops we made it crucial for everybody to drive so now we have to let everybody drive even though that's unsafe.
1) You would need self driven cars to drive over 100 million miles to validate that, assuming that there have been no deaths from full 'self driving' (i.e. Cruise/Waymo rather than Tesla, which isn't level 4)
2) Depends what you call a crash - 1 impact per 5 years might not be entirely wrong... According to this link:
https://1800lionlaw.com/how-many-times-will-your-crash-you-c...
> The car insurance industry estimates that the average driver will file a claim for a collision about once every 17.9 years.
This matches me - I've had 1 crash in 15 years of driving which resulted in a claim (someone drove into the side of me), but there have been at least 2 other impacts which have not resulted in a claim (1 person bumped into the back of my car on a motorway and impacted but but no damage, and there was one incident where I accidentally scraped a car when parking and I agreed to pay the driver for repainting directly without going through insurance).
The issue is, that I can totally see a self driving car fuck a rare obscure situation up in the worst way — where even the worst drivers would probably do okay.
E.g. let's say you are in a tunnel and there is a fire, you car blocks the fire truck. The normal line of thinking would be to throw all existing rules of traffic out of the window and so totally illegal maneuvers that might even damage your or other peoples cars. A self driving car might be able to do that, but it cannot make the judgement needed to know when to throw all rules out of the window.
Now in that example the driver (if they exist) could take over, but there are similar examples where there is less time to decide and/or a switch to manual might not be feasible. Think of all situations where multiple dangers have to be weighed against each other, for example a rock slide where a street is partly blocked (enough for the AI to stop) but there is still rock coming down the mountain. A situation where a criminal gang blocks the road, in order to make you stop and rob you — this being a thing that could become easy game when self driving is wide spread enough.
The problem with self driving cars is that even if you statistically manage to be safer over the typical journey today, people still wanna know that it keeps them safe in very atypical conditions as well. And that as of now is not the case.
So? Unlike with humans, self driving networks of cars can learn from a rare obscure situation, then the entire network can be updated to handle it going forward, forever. Whereas "the worst drivers" (drunk speeder road ragers say) will just cheerfully plow into students in front of their schools yet again.
>The problem with self driving cars is that even if you statistically manage to be safer over the typical journey today, people still wanna know that it keeps them safe in very atypical conditions as well.
Meh. This is just a typical argument from incredulity/unfamiliarity. People have zero issue accepting significant risk given solid benefits and most people's driving most of the time is short and boring. That'll be Good Enough to get the ball rolling, and it'll only ever improve.
>And that as of now is not the case.
Where is the suggestion in the article that the current cars are ready to handle absolutely everything everywhere today? It's about incremental gains and safety in one particular very high population location.
This is not an argument that can be easily be wiped away by a class of people who wants to believe in new technology (and profits from its potential adoption). After all from the standpoint of most untechnological people the past decade has been a one where everything they knew got shittier and less reliable while claiming to be "smart". It got even to the degree where you need to create subscriptions for basic features of your damn car, to the benefit of exactly nobody other than the shareholders.
In that climate taking the leap to trust a new technology is something you need to be willing to make. And many people aren't buying the promises. Again, whether they are right or wrong isn't the point here, but some technology is better suited for incremental safety improvements than others. E.g. nobody gives a damn about roombas falling down stairs, but if it is a damn car with their family in it they are more careful about their decision.
There are also psychological aspects that play into this. Given the choice:
A) travel inside a box that has a 10% chance to kill you and your family without you being able to influence the outcome at all
B) travel inside a box that has a 15% chance to kill you and your family, but if it happens you are at fault and you can also prevent it from happening
Most people would choose B just because the agency they have gives them a better feeling. Now granted, when we reduce the probability of A and/or increse the probability of B there will come a point where most people will flip. In the case of self driving cars this is not a out numbers, it is about perception. And the point of my post was that the perception isn't there yet.
> With all that said, it seems that Waymo cars get into serious crashes at a significantly lower rate than human-driven cars.
The companies can absolutely develop their products outside of public streets. A few thousand acres and some purpose built infrastructure would do it. If that's not cost effective, well that's their problem.