Slashdot Mirror
Are Tesla Crashes Balanced Out By The Lives That They Save? (eetimes.com)
Friday EE Times shared the story of a Tesla crash that occurred during a test drive. "The salesperson suggested that my friend not brake, letting the system do the work. It didn't..." One Oregon news site even argues autopiloted Tesla's may actually have a higher crash rate.
But there's also been stories about Teslas that have saved lives -- like the grateful driver whose Model S slammed on the brakes to prevent a collision with a pedestrian, and another man whose Tesla drove him 20 miles to a hospital after he'd suddenly experienced a pulmonary embolism. (Slate wrote a story about the incident titled "Code is My Co-Pilot".) Now an anonymous Slashdot reader asks: How many successes has the autopilot had in saving life and reducing damage to property? What is the ratio of these successes to the very public failures?
I'd be curious to hear what Slashdot readers think. If you add it all up, are self-driving cars keeping us safer -- or just making us drive more recklessly?
But there's also been stories about Teslas that have saved lives -- like the grateful driver whose Model S slammed on the brakes to prevent a collision with a pedestrian, and another man whose Tesla drove him 20 miles to a hospital after he'd suddenly experienced a pulmonary embolism. (Slate wrote a story about the incident titled "Code is My Co-Pilot".) Now an anonymous Slashdot reader asks: How many successes has the autopilot had in saving life and reducing damage to property? What is the ratio of these successes to the very public failures?
I'd be curious to hear what Slashdot readers think. If you add it all up, are self-driving cars keeping us safer -- or just making us drive more recklessly?
Who cares what Slashdot readers think? This isn't something where opinions or anecdotes matter. Do (or read) a study, collect data. Then you'll have an answer.
What people think makes no difference. In fact, I think how many lives are saved vs. lost makes no difference. No body, not Tesla, not government regulatory bodies, not programmers, no one is claming the software is perfect. Some people are claiming it saves lives in aggregate vs human miles driven, but only data can tell us that. And even then, it determines on how you look at things. Do they save lives vs the average vehicle? How about lives vs other vehicles in the same price range, or vs other autonomous vehicles. Or even vehicles in similar weather conditions, etc... etc...
What really matters is how much potential there is for improvement. The chips these systems run on will follow a Moore's law trajectory, and the amount of data these vehicles learn from each other with every mile is even more insane. We can not possibly make human drivers 2X, 4X, or 10X better, but we can make these systems that much better. All it takes is learning with data, with next gen sensors, with better networking tech, better algorithms derived from it all, and lastly, better vehicle coordination from infrastructure. In 20 years, the answer to this question will become painfully obvious. We just have to let the technology carry us there, and listen to everyone on the way: be vigilant when the system is in use today. Be aware of what your vehicle is doing. You are still responsible.
The biggest problem with Tesla is not their technology, but their communication. They call their system "Autopilot", but backpedal that statement in their legalese and fine print. They say their car is safer, but only acknowledge accidents after investigative reporters uncover them (the attached article is a perfect example). Further, they always shift as much blame on to the driver as possible, while giving as few details about the crash as possible. This is poor communication. Tesla should be transparent about how well they're doing. It's to their benefit, as many people (myself included) would be more open to trying out potentially unsafe technology if the risks are clearly explained and can be mitigated.
Back to the original point of the question, "Are Tesla Crashes Balanced Out By The Lives That They Save", Tesla could easily answer that question if they wanted to, they have the data. Unfortunately, due to their secretiveness and poor communication, they only share self-serving pieces of what they know, and no one gets the full picture. Based on the way Tesla has conducted itself, we have to assume that these vehicles are unsafe unless proven otherwise.
Fast Federal Court and I.T.C. updates
Without being subsidized would GM or Chrysler have been viable companies?
An article over on Forbes already looked into this.
The TL;DR version is that Tesla's autopilot has 1 fatality per 130M miles driven, while the US average of all vehicle-related fatalities comes out to about one per 94M miles. That's 94M miles under all roads, all conditions, compared to Tesla's autopilot being driven almost exclusively on highways.
A few score may die now to save hundreds of thousands later. Have you driven on the freeways of America lately? People drift into your lane, they don't stay centered in their own lane. Drivers are looking at their phones while they're driving, no matter what the laws say. People are NOT as qualified to drive a car as a computer which checks its sensors hundreds of times per second.
Autonomous cars are the future, and Tesla is pushing that forward. There are going to be mistakes in the beginning, and people will die and be injured.
Driving is dangerous, but we don't outlaw cars because their utility outweighs the risk. Same here.
Dang, we've been found out. We thought we'd covered our tracks so well, despite all of us mysteriously being absent for the day -- we were worried that would kinda be a big clue, but till you came along no-one figured it out -- your brilliant investigations and whip-smart deductive powers have found us out. Foiled again! We'll just have to be more cunning next time. But I fear that with people with your remarkable qualities around, our plans for world domination can never come to pass.
The TL;DR version is that Tesla's autopilot has 1 fatality per 130M miles driven
Actually the important fact is that they have 130M miles total, one ugly front-on-front collision and the numbers would be completely different. The national average is statistically good across 32k deaths, but extrapolating from one death is folly.
Live today, because you never know what tomorrow brings
You don't see a news story every time a Mazda is crashed during a test drive. Stop giving clicks to this drivel.
I find it curious that AI-guided vehicles appear to be held to a standard- perfection- that is never expected of humans. This is especially important given that driving is the kind of task that humans do especially poorly: it requires extended attention to something that is, for the most part, repetitive and boring. Given those kinds of tasks, humans easily lose focus, where computers do not.
Given that the US has averaged 35480 deaths on the road over the last 10 year (https://en.wikipedia.org/wiki/List_of_motor_vehicle_deaths_in_U.S._by_year) I have a hard time seeing how AI *couldn't* make the roads safer.
-Z
-Z
We spend far more attempting to avert air crashes than car crashes. The regulators of both form of transportation have struggled with why they are pushed by political forces above them to have such different levels of concern for the same lives. People doing polls and focus groups, professors doing anthropological studies, say, have formed the opinion that it relates to control.
We chafe at having our autonomy restricted in cars - speed limits, four-way stops, seat belt laws, helmet laws, all unpopular, though such restrictions seem small prices to pay for your life. The cost of a highway interchange, at $50 million plus a million or more a year to maintain and replace, can be controversial though it would save a life per year in perpetuity, a couple or three million per life. We feel a death is a lot less anybody else's fault if we were in control of the vehicle at the time. On the other hand, a death in air traffic is just being tossed into the ground at 600 MPH by somebody else who screwed up. We really hate that a lot more.
So, yeah, autopilots are always going to have to do twice as well to be half as appreciated. It's a glitch in human nature. Sorry.
It would be hard to determine what it prevented when no one is reporting it. Some of the drivers may not have even been aware of the incident the car avoided.
It is so much more complicated than that. Here are two perhaps more reasoned points to consider...
First, fatalities aren't all that matter - injuries and property damage should also be considered. It would be awesome if someone could come up with a cost of accidents per million miles rating that puts a price on the lives and adds it in to get a single number.
Second, it needs to be considered systemically and with a long-term point of view. A transition that costs a few more lives in the short term but brings a vast improvement forth sooner in the long term could cause the total lives saved over, say, the next 20 years to be vastly higher with a short, dangerous transition than a long, "safe" transition. We all too often miss the advantage of paying a price up front.
We should also consider other advantages of rushing to the eventual arrival of fully autonomous vehicles. Full autonomy is almost undoubtedly going to result in vast savings to any society that utilizes it - no more insurance cost, a drastic reduction in accidents, a transition to a vastly smaller fleet of cars that are shared, million mile cars once the companies that are making them are also the owners and maintainers, fewer roads due to much greater efficiencies of road utilization, etc., etc., etc. Vast pools of both time and materials will be freed for other endeavours. Awesome.
This is something that can be easily figured out with statistics.
It's not easily figured out without data. And we have no data.
Accident rates and serious injuries per miles driven in autopilot vs human.
On the one hand, human drivers with a couple hundred million people driving billions of miles in the USA, rack up on average 1.08 death per 100 million miles. all miles. all weather. all types of driving. includes drunks. includes iced up roads. includes stolen cars. includes accidents caused by poorly maintained vehicles losing a wheel. it even includes suicides.
On the other hand, you have autonomous vehicles that haven'tracked up remotely enough miles to even be statistically significant. 1.08 death per 100 million... you are going to need billions of miles to get the error bars into a reasonable place.
But on top of that those autonomous vehicles are predominantly driven on the highway in mostly ideal circumstances in vehicles maintained by a team of engineers, with drivers who TAKE over if & when the situation gets complicated.
When you've got a billion miles racked up dropping kids at school in rush hour, with stop lights, crosswalks, pedestrians, cyclists, roundabouts, crossing guards, double parked cars unloading kids in heavy rain, sleet, snow, and fog.... etc...
Then take the human statistics and factor out accidents where the vehicles were in poor shape contributing to the accident (e.g. where an old worn tire comes apart on the highway, or stopping distance is impacted by brakes that are past due for service, or some idiot is running summer tires on snow... because google/tesla autopilot isn't coping with any of that right now. ).
THEN maybe we'll have something statisically valid to compare.
** as for drunks and suicides and stolen vehicles ... clearly that doesn't represent human driving ability; so having it count against the human driving stats isn't really representative of the human accident rate.
On the one hand automous cars might nevertheless eliminate those sorts of accidents. Drunks at least. suicides will find a way.
And we havent even considered the new failure modes of autonomous cars -- poorly maintained malware ridden IoT cars of the future, with remote hacks, and dirty 15 year sensors... are a far cry from the meticulously maintained test cars google / tesla etc running.
That is ignoring the additional problems with the complex chemistry, manufacture, and eventual disposal of batteries versus steel and aluminum engines.
The batteries are almost 100% renewable, so best you do "ignore" them, rather than reveal your bias and ignorance.
Almost certainly not better when everything is accounted for.
Lower energy per mile. Lower demand on fossil fuel. Win all around. Account for everything, and electrics/plug-in-hybrids are better, by far.
Learn to love Alaska
Gasoline on the road versus coal/natural gas/etc... in a power plant, then transferred at a slight loss through the electric grid, transferred at a loss to charge batteries, then transferred at a loss again into mechanical energy. That is ignoring the additional problems with the complex chemistry, manufacture, and eventual disposal of batteries versus steel and aluminum engines. Almost certainly not better when everything is accounted for.
Ok; lets pick that apart, only this time with actual knowledge instead of your "intuition".
First, Gasoline engines are around 20% efficient. That means you burn 5 gallons of gas and get 1 gallon of gas worth of actual travel. Plus, Gasoline cars do not have regenerative breaking (we'll get to that later).
Second, Electricity production using Natural Gas has 35% efficiency. So for that equivalent 5 gallons of gas, you get 1.666 gallons of gas worth of travel, but you have to add charging efficiency, battery to wheel efficiency and of course transmission line efficiency. I'll leave it as an exercise to the reader to look those up, but For rough numbers, the battery to road efficiency is about 87%, the transmission line efficiency is about 50%, and the charging efficiency is about 85%. There are some other minor factors, but the big ones above are the main items. So, you take and multiply all of the efficiencies number to get the final efficiency. With electric vehicles, that efficiency is about 13% all told.
So! now we know that gas engines really are more efficient! Not so fast, heres where that regenerative breaking I mentioned above comes into play. There is no known way to convert mechanical energy back into gasoline, but with an electric vehicle, the circuits to take mechanical energy and put it back into the battery when breaking are very trivial to make. That means that the amount of energy it takes to make an electric car with regen breaking go 1 mile is far far less than it is to make a gasoline car go the same distance when all other factors are included. This number is typically 1/4 of the energy on average. That means that of that original 5 gallons of gas worth of energy, the electric car only got .64 gallons, but that will take the car 2.58 gallons of gas worth of distance while the gasoline car only got a gallon of gas worth of distance. Regen breaking is the key folks, always has been. With electric vehicles, regen is almost automatically included. With all other types of propulsion, it is a nasty and complex problem that is usually solved by ---- you guessed it---- adding a battery and electric motor...
I wish I had a good sig, but all the good ones are copyrighted
In the very grand scheme of things, autonomous driving technology has the potential to save hundreds of thousands of lives every year.
The idea is that while autonomous driving can advance, become better and actually learn from past experience... human drivers cannot.
Things have been getting slightly better in recent years, but if you take overall statistics ranging from the 70s or 80s 'till today, the number is pretty constant. I mean, the number of accidents, and the number of deaths in crashes. Too many people die every year in car crashes, and a whole lot of them comes from problems we are tired of knowing about. Driving under the influence, speeding, not paying attention to road signs, underestimating the severity of handling a x tons metal box at high speeds. Ramping up fines, making it harder to get a license, educational efforts, changes in law, among several other measures might have helped a bit, but not enough. And there doesn't seem to be much better solutions for human drivers.
Autonomous driving has the potential to drastically reduce numbers when it eliminates the human factor. There will always be crashes, accidents will still happen, and I don't think autonomous driving can ever reach a point of perfection... but if done right, it could eliminate a whole lot of erratic and problematic behavior behind wheels.
But this is about the overall technology, not about Tesla in particular. It's a huge and drastic change, and if I'm honest about it, I think Tesla is rushing things out, not taking lots of stuff in consideration, and turning the whole thing into hype and selling point - with big risks of making it a step back instead. It doesn't take many car crashes while using the autonomous thing for people to start avoiding the technology altogether.
Tesla is selling it as if they had the complete solution already, but it's really the first steps into the technology, which is a really questionable strategy. It's skipping ahead using consumers as testers, quality assurance and research and development, instead of doing it like other manufacturers are doing it - in controlled environments, by employees.
So no, if a Tesla car crashes and kills the driver because the technology isn't working as it should, it's their responsibility. Nothing balances out. Saving the lives of others won't bring back the lives of those who were killed, it won't fix things for the families of those who were lost, and it's no excuse for putting out a technology prematurely. But the technology itself is worth investing and worth insisting on.
We should also consider other advantages of rushing to the eventual arrival of fully autonomous vehicles.
Oh, absolutely. I'm not inherently against self-driving vehicles: the benefits are massive and far-reaching. There are immense economic pitfalls that need to be navigated, but in the long run it's a net gain. As a rough snapshot of where we stand today, however, it's fair to compare Tesla's fatality rate with traditional vehicles (even while it is more nuanced than that).
We all too often miss the advantage of paying a price up front.
We do, but we cannot forget that sometimes the price is people.
They can get a zero on both sides by doing nothing. If they can't beat that they shouldn't be in the game.
Sheesh, evil *and* a jerk. -- Jade
Without being subsidized would GM or Chrysler have been viable companies?
Yes, they would have declared bankruptcy, restructured, and continued on......which is actually what they did in the end anyway.
"First they came for the slanderers and i said nothing."
So what you are saying is... that according to you the only morally acceptable kind of car company is one that doesn't make cars.
Well I'm sure the very, very, very far extreme fringes of the environmental movement will agree with you - that is if you go live with them in their hippie communes in the woods - but the rest of the world will probably collapse if we tried that. Better to try and build greener and progressively safer cars. There are times when you can and should demand perfection - but this is one of those cases where perfection will never exist, so you can and should demand improvement, which is exactly what Tesla is doing.
Bad things happening sometimes does not mean it isn't improvement. It just means it's not perfection.
Unicode killed the ASCII-art *
First, I mostly agree with you.
However, the issue i have with drunks in the stats is essentially that the addition of autonomous cars into the system makes the human statistics better; as the 'least able humans' would avail themselves of the autonomous option to get home.
This really counts as a point FOR autonomous cars; but it needs to be noted because it suggests the autonomous systems in some sense don't need to be better than the average human drivers... they just need to be better than the worst human drivers (drunks and people who are half asleep, as examples) to be a net gain for society.
That said, I have a lot of misgivings... from surveillance capabilties (ugh), remote hacking (remote theft - not just of the car... but of the contents... even remote kidnapping); not to mention the chaos you could create with a botnet of compromised vehicles...or hell even just a few well placed stalls could bring a city to its knees.
And then those cars are going to age out; you look around today and you'll see some 35 year old driving a 1996 Hyundai Tiburon beater he paid $1000 for; the cars seen better days... but the drive train and the supension are maintained... and it can still be perfectly safe because the driver is perfectly competent.
Would you trust that car to drive itself? With 20 year old software and 20 year old sensors? Is it still going to be ok? Is it even going to know how not ok it is if its not ok? If one of the sensors is cockeyed will it know? What will 20 years of dirt, moisture, temperature changes, condensation, insects and rodents, acid rain and UV exposure do to the performance of the autonomous sensors and electronic systems? What gremlins will that car have?
Is the autonomous performance of a test car meticulously maintained by a team of engineers even relevant to what the real world is going to be like? Even Tesla's today are all upper class status cars owned by enthusiasts... will you trust its autopilot the same in 20 years when the car is being hawked on craigslist as a beater for $1,200 by a guy who tells you that the big gouge down the side should buff out and is trying to avoid admitting that the car was written off at least once and has 'rebuilt status'
You would get even bigger savings.
Ummm, No.
The best of the lot: Flyweheels are fundamentally limited by how much mechanical energy can be stored safely in a flywheel. A flywheel significant enough to store the energy of a 2 ton motor vehicle moving at 75 MPH is going to require so much mass that it will actually impair the maneuverability of the vehicle (unless it is fully gimbaled, in which case, it is going to weigh a lot more, and will require a hugely complex slip-ring arrangement, which will take up a huge amount of space).
One of the most powerful flywheel systems ever built was a 133kwH pack for railway use. It occupied the better part of a locomotive, weighed 25 tons, and could accelerate the locomotive to cruising speed (although not much more). Compare that to Tesla's 80kwH pack that fits handily inside a passenger vehicles with room to spare for other amenities such as passengers. Flywheels have had 100+ years of research and that is the best it gets. Batteries on the other hand have had around 20 years of real research, and there is still a lot of room for them to continue improving.
I'm sorry to have to tell you that if you got a mechanical engineering degree hoping to design vehicles, you wasted your time. The future of transportation is electric, and not even a trillion dollar automotive industry could prevent it forever.
I wish I had a good sig, but all the good ones are copyrighted