Elon Musk: Tesla's Autopilot Software Could Save Half a Million Lives Every Year (fortune.com) 265
An anonymous reader writes: In the wake of a deadly crash involving a Model S that was driving with its Autopilot software turned on, Tesla CEO Elon Musk issued a few interesting remarks on the technology to Fortune. Notably, the publication recently ran a piece attempting to portray Tesla in a bad light by noting that Musk sold more than $2 billion worth of Tesla stock just 11 days after the aforementioned May, 2016 accident. And all the while, shareholders were kept in the dark up until recently. "Indeed, if anyone bothered to do the math (obviously, you did not) they would realize that of the over 1M auto deaths per year worldwide, approximately half a million people would have been saved if the Tesla autopilot was universally available. Please, take 5 mins and do the bloody math before you write an article that misleads the public.
except..... (Score:2, Insightful)
Re:except..... (Score:5, Insightful)
Re: (Score:3, Insightful)
Re: (Score:3)
Maths & Journalism don't mix (Score:3)
Where is this maths (Score:2)
Re: Where is this maths (Score:3)
I didn't RTFA but Tesla tracks all cars in real time. They have observed that cars on Autopilot are half as likely to have an accident (as measured by air bag deployment).
Re: (Score:2)
Change the name of the feature!! (Score:5, Insightful)
Pilot-aid would be better and might have saved an extra life.
Re: (Score:3)
Rolaid would be better and might have saved an extra life.
FTFY
Re: (Score:2)
Pilot-aid would be better and might have saved an extra life.
How about drunk-aid? Which begs the question, at what point is autonomous driving good enough to allow drunks back behind the wheel, and whose to blame if there is an accident?
Re: (Score:2)
Re: (Score:3)
I'd say at the point when they can legally get into the passenger's seat instead. And do so. So long as you need a human driver to be able to override the autopilot, they should be legally competent to drive.
Once you cross the line from "autopilot" (aka simple-situation driving aid) to a fully autonomous system such as Google's self-driving cars are pursuing, then you are simply a passenger, and it shouldn't matter if you're too drunk to stand, any more than it would if you hired a taxi in that state. At t
Re:Change the name of the feature!! (Score:4, Informative)
Autopilot is actually an accurate name for it.
Autopilot was primarily invented for aircraft and even today, autopilot will still happily fly an aircraft into terrain without human interaction if you let it. There have been numerous CFIT fatal crashes of aircraft with over 9000 deaths. Each of these incidences brought more knowledge of how to improve technology to help prevent future occurrences (I expect the same to happen with autonomous vehicle technology). Autopilot was never intended to replace the human pilot or alleviate the responsibility of the human pilot to maintain constant situational awareness. Likewise, autopilot in the Tesla was never intended to alleviate the driver of the responsibility to maintain continuous situational awareness. The driver actually has to agree to this when using it.
I think Hollywood may have warped people's perception of what autopilot actually is and its limitations.
Ideal vs. All Driving Conditions (Score:5, Insightful)
The Tesla autopilot running under ideal conditions (with a human backup) compared to a human driver under all conditions are not equivalent, and we cannot directly compare their failure rates. Beware of naive statistics.
Re:Ideal vs. All Driving Conditions (Score:4, Insightful)
It's pure bullshit. Highway driving has about 1/4 the death rate of all driving.
Highway driving is all the Tesla 'autopilot' can do. It's more dangerous than human driven for now.
Back at you, summary author.
Re:Ideal vs. All Driving Conditions (Score:5, Interesting)
Not true. The current US death rate for highway driving is 1.08 deaths per 100 million miles driven. Tesla's average at the moment is 0.769 deaths per 100 million miles driven on autopilot.
Re: (Score:2)
I've been speaking American English all of my life and in my experience "highway" pretty much exclusively refers to "divided highway". There may be regional variations in usage, as well as distinctions in technical communications.
Re: (Score:3)
Calm down Elon. You're about to go cornholio.
Re: (Score:2)
Calm down Elon. You're about to go cornholio.
He tends to start spouting stuff when reports come out regarding Tesla's repetitive missed delivery targets. Deflection
Not feasible, he's shirking responsibility (Score:2)
This would only be true under ideal circumstances in which everyone and everything worked flawlessly in tandem, and that just isn't the real world any more than the opposite statement. Suck it up, take responsibility, and be a man, Elon. Do the right thing and suspend public trials of this tech until it's truly ready.
Re:Not feasible, he's shirking responsibility (Score:5, Insightful)
Re:Not feasible, he's shirking responsibility (Score:5, Funny)
Tesla claims you still need to pay attention to the road around.
Its like having your wife in the car. She corrects you when you are doing something wrong, warns you when you get out of line, and expects you to pay attention. What a wonderful feature!
Re:Not feasible, he's shirking responsibility (Score:5, Funny)
Re: (Score:2)
It is very dumb to compare airlines and cars. Airliners at cruising altitude don't require split second reaction most of the time. Autopilot disengages, pilot takes over few seconds later. No cross traffic appearing unexpectedly out of nowhere. It has nothing to do with cars other than silly name Musk chose for marketing. Today's cars have traffic aware cruise control and lane keep assistance. The same terminology is used in Tesla manual, as for all other automakers. Now only Tesla went into "ludicrous irre
Re: (Score:3, Informative)
Why would he suspend trials - the current version is already safer than a human. The US average death rate when driving on a freeway is 1.08 deaths per 100,000,000 miles. Tesla autopilot's current death rate is 0.769 per 100,000,000 miles.
Re:Not feasible, he's shirking responsibility (Score:5, Insightful)
The US average death rate when driving on a freeway is 1.08 deaths per 100,000,000 miles. Tesla autopilot's current death rate is 0.769 per 100,000,000 miles.
Come back with deaths per mile of people driving high end, less than 10 year old vehicles, and exclude miles driven in snow, ice or other treacherous conditions and also eliminate passenger deaths. That's just for starters.
Re: (Score:3)
Re: (Score:2)
That the death rate in America for all roads.
Freeways/divided highways are about 1/4th that.
Re: (Score:2)
Re: (Score:2)
With a sample of 1 death, the two are statistically equivalent. Maybe after we ~10-20 Tesla deaths with autopilot we can properly compare, but for now...
Re: (Score:2)
Only if it's affordable (Score:5, Informative)
Re: (Score:3, Informative)
How many years will it take for the automated car to be affordable for the common person?
The marginal cost is very low. It is mostly software, which has a marginal cost of zero. Then there are a few sensors. Hi-res cameras cost less than $5 each (which is why they are in $20 cell phones). Radar units used in adaptive cruise control are less than $1000, and dropping in price. If your car already has ACC (as many do) then the additional cost for full self-driving is minimal. It is likely that you will save more on insurance than the extra cost for hardware.
Re: (Score:2, Interesting)
So what? The cost of producing the FIRST copy of the software is sky-fucking-high. And every time you change your hardware platform, you have to spend a vast amount to certify, test, bug-fix, and fully integrate your software with that new, revised hardware. This isn't a fart app that "oh well, if it crashes now and then, no big deal." The requirements for mission-critical realtime software systems on which lives dep
Re: (Score:2)
> The requirements for mission-critical realtime software systems on which lives depend are (and *should be*) incredibly high.
Why? As I see it they should follow the same criteria as pharmaceuticals: at least as safe and effective as the current standard. Which in the case of driving systems is heavily peppered with drunks, idiots, distracted soccer moms, etc.
Re: (Score:2)
Now cost out the robotic brake and steering systems.
Re: (Score:2)
Should be cheap enough - scale models of the systems are already mature in the high-end RC car market and mechanical actuators aren't exactly expensive. In fact many new cars already have such systems in place as safety aids - all that's missing is the judgment to take over full-time instead of just in emergencies where it appears the driver won't avoid a crash on their own.
Re: (Score:2)
Re: (Score:2)
I'd like to see that starting with ECUs for normal cars.
Re: (Score:3)
Why? Aircraft systems require extensive certification because:
1) They typically carry dozens to hundreds of people who are liable to die because of a single failure.
2) A single crash has a fair chance of causing extensive additional death and destruction wherever it hits
3) The industry has mature regulatory capture by a tiny pool of businesses building avionics components, all of which would much rather pay ridiculous certification expenses than face competition from any promising upstarts that might other
Re: (Score:2)
How many years will it take for the automated car to be affordable for the common person?
With services like Uber, it's not going to take very long. In the case of the "average family", Uber is already cheaper for some trips than the bus.
And let's do the math, a full time Uber car typically does 180,000 miles every three years. But a fully automated Uber car could work around the clock and should do much more than that -- thereby requiring more frequent replacement of the car.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Become standard (Score:2)
Collision avoidance technology are becoming more widespread over time.
What used to be only available on high range Volvos and Mercedes has now trickled down and even the smaller and cheaper VW Up! have LIDARs used for "City Safety" (=automatic brake to avoid collision with pedestrians and with other vehicle at in-city speed ranges) as a standard option.
And that is the car currently available as the cheapest option of the fleet of some car-sharing companies.
It *is* getting affordable.
(Well for a certain cate
Re: (Score:2)
Re: (Score:2)
Are you sure? Telecommuting can be done from around the world as easily as across the street. Are you ready to compete on equal footing with Indian and Chinese applicants with a much lower cost of living? And then there's Africa which is rapidly closing the communication and education gaps, and is known for its numerous ingenious free-thinkers which are so often in short supply in the Asian cultures.
Re:Only if it's affordable (Score:4, Interesting)
Re: (Score:2)
Please solve the problem of shopping for groceries if I can't drive to the grocery store and no one from the grocery store can drive to my house.
I'm all for reducing traffic, but I kinda hate starving ...
Re: (Score:2)
Please solve the problem of shopping for groceries if I can't drive to the grocery store and no one from the grocery store can drive to my house.
I'm all for reducing traffic, but I kinda hate starving ...
At one time there were small grocery stores every couple of blocks. Only when people proved willing to drive for a simple task like shopping, did they all go out of business and consolidate into a single neighborhood super market that's nowhere that anyone can walk.
Re: (Score:2)
Re: (Score:2)
Call me when your plumber "telecommutes" to fix your faucet. There are millions of jobs out there that demand that you (or somebody) be present. Plumbing is just one of them.
Youtube fixed my bathroom faucet. I just provided the hands and tools.
Re: (Score:2)
The fact used cars don't duplicate phone features is a plus not a minus.
In 10 years this car will be an orphan with no parts available and no existing manufacturer. They will be about as common as Deloreans were 10 years after they stopped production.
What Math? (Score:3)
He continued, “Indeed, if anyone bothered to do the math (obviously, you did not) they would realize that of the over 1M auto deaths per year worldwide, approximately half a million people would have been saved if the Tesla autopilot was universally available. Please, take 5 mins and do the bloody math before you write an article that misleads the public.”
Are these projections from peer-reviewed research published in a proper journal? Are these projections based on public Tesla claims? Is this Elon Musk pulling numbers out of his trunk?
Considering these are real lives of actual people at stake I hope Tesla did some serious research before selling these to the public.
Re:What Math? (Score:4)
He's pulling the numbers out of his trunk.
If the Tesla that killed its driver was full of people, it would be 4 deaths in 130 million miles, twice as bad as the world-wide average, 3x worse than the USA average.
Re: (Score:2)
He's pulling the numbers out of his trunk.
Wow, good thing Teslas have a big trunk, that's a big number.
Re: (Score:2)
Considering that we world wide have far less than half a million deaths to cars per year ... his trunk is really big!
Re: (Score:2)
That level of wealth alone... (Score:3, Insightful)
That level of wealth alone would save far more lives. Anybody who could afford a Tesla could also afford clean drinking water, air conditioning, medicine, proper nutrition, etc. Musk is just taking in one figure and ignoring the fact that so much of the world is driving a run-down beater that doesn't have anti-lock brakes, or they're just driving motor scooters which are far more dangerous, or they're driving nothing at all and hauling water from toxic wells because of POVERTY. How about Musk buy a helmet for every 3rd world motor-scooter rider, then get back to us on this?
Re: (Score:3)
Re: (Score:2)
You're putting words in my mouth. I didn't say he should abandon the technology--only that his reasoning is specious.
Re: (Score:2)
Re: (Score:3)
If these statistics aren't a credible threat [asiancorrespondent.com] then I don't know what is. True though-- people tend not to wear helmets unless there's a ticket involved. For some strange reason, a chance of losing a few days wages correlated with the sight of an available officer is greater motive than the chance to lose your life correlated with seeing accidents. So then I guess I would have to add that Musk should involve himself with lobbying for helmet laws overseas, or maybe what I really should have said is that ad
Re: (Score:2)
I'm not so sure - autopilot seems a fairly accurate description to me: a driving aid designed to handle the minutia of maintaining a steady course through low-risk/low-variable situations so that the pilot can stay relatively fresh and focused on risk assessment until needed. An autopilot is categorically NOT a replacement for an alert pilot ready to take over the controls at any time. Though arguably there's a widespread enough misunderstanding of what autopilots actually do that makes it inappropriate.
Re: (Score:2)
How about "autoconvoy". With lane assist + adaptive cruise, it's a convoy. 10-4?
Re: (Score:2)
Does that say "less sophisticated" to you? It says "more" to me - your average aircraft/ship autopilot just assumes there's nothing else around it, and screams bloody murder if it detects anything on an even vaguely intersecting course (assuming it even has any detection capacity at all) , forming a convoy would be completely beyond it.
I don't believe the claim (Score:5, Insightful)
I think there's a huge unrealized danger to these quasi-autonomous cars because people will treat them like a a fully controlled car and do things they shouldn't (e.g. read the news paper, watch a movie, doze off, etc.). Right now the drivers of these expensive Tesla cars are not representative of the larger driving public. If we put this technology into 100% of the cars on the road I predict the number of accidents due to imperfect AI will rise significantly because of driver inattention. It may still prove to be an improvement over human controlled, but I doubt the numbers of lives saved will be what Musk claims.
Give me a car that will take me to work while I nap in the backseat. I have no interest in being on the road filled with semi-autonomous cars.
Re:I don't believe the claim (Score:5, Insightful)
Give me a car that will take me to work while I nap in the backseat. I have no interest in being on the road filled with semi-autonomous cars.
Sorry, but history seems to indicate that humans don't adopt technology this way. We just brashly try shit way before it is remotely safe to do so, and then regulation follows when necessary. Hell, seat belts weren't standard on ANY car until 1958, and the very first seat belt law anywhere in the world didn't happen until 1970. Automobiles have historically been death traps, with a continuum to relative safety now. This will probably continue going forward, until our descendants view our relative death traps as we do the Model T. Automation will almost certainly make cars safer, but it won't be a binary operation. It will be a long slog through imperfect implementations.
Re: (Score:2)
Re: (Score:2)
"We"? No. We are exactly the same species as our parents and grandparents. We'll keep right on killing each other in new and technically interesting ways. Musk can point to his 1 fatality in 134 million miles track record as being better than the 1 in 94 million human average - fair or not. He's selling cars, after all. By the time regulators catch up, they'll be requiring autopilot because the accident rate will be consistently better than humans. We'll keep right on being human.
Re: (Score:2)
Totally agree with you protest_boy.
I drove a Tesla for a week with the Auto Pilot function and found it dangerous.
Because it relies on white lines for steering, a soon as conditions become non ideal, it gives up.
Volvo said semi autonomous driving was a bad idea, and I agree with them.
You often have hundreds of milliseconds to take corrective action, and half assed autonomous driving system (tesla has no LIDAR) will leave you lapsing concentration for extended periods making you a risk on the road.
One Idiot Dies (Score:2, Insightful)
..and half of people think its the fault of the tool he was (mis)using.
Frankly Tesla's autopilot isn't all that special. It's just a combination of lane keeping, adaptive cruise control, and brake assist that pretty much is available from any car company.
There will never be an automated system if any kind that won't cause deaths.
Re: (Score:2)
Actually, it's not inattention per se, but there's considerable evidence that *all* of those driving aids cause people to drive more recklessly once they get accustomed to having the "safety net" available, until their driving safety declines to almost where it was without them.
As such, it does seem reasonable to expect that a semi-autonomous driving system will be at least as bad, and likely worse since it facilitates complete inattention until it's to late to develop the situational awareness needed to av
Re: (Score:2)
Where is that 'considerable evidence'? The traffic fatality rate was about 40% higher 10 years ago, 70% higher 20 years ago, 250% higher 30 years ago, and 350% higher 40 years ago.
Bullshit Elon (Score:2)
Just because when your auto-pilot is activated, there have been half the number of deaths per kilometre doesn't mean you an save half the people who died in car accidents.
It's only used on highways. That's a road with no pedestrians to kill.
If keeps you in your lane and attempts to keep a safe following distance from the car in front of you (unless a truck pulls out infront of you, it keeps driving at it at full speed and kills you)
In Britain 60% of deaths occur on country roads, not highways, where street
Re: (Score:2)
Its actually highways with a median which is even more restrictive in terms of unexpected things scenarios.
Listening to general car podcasts I've heard a few stories from reviewers that the autopilot will suddenly bail, say mid-corner which doesn't sound particularly safe to me.
Never go to be work "as is" (Score:4, Insightful)
I love the idea of self-driving cars, but I think it's going to be hard to get it to work acceptably "as is", that is, the way the problem is being approached.
Much of the decision-making must remain "onboard" the car but I think self-driving vehicles will be vastly improved with some feedback and control signals from the road or a locale-specific traffic guidance computer.
In other words, in addition to its own decision-making software, the vehicle should also be receiving some sort of signals or guidance info from the road in the area it's currently passing through.
Sort of an air-traffic control system where responsibility for air traffic is handed off from control center to control center as the plane makes its way from point A to point B. The difference is that this guidance should be completely automated, and be an adjunct to what the car does, not its primary means of navigation. I'm see this primarily as speed and road condition management info.
I know, I know- what about hackers? Yeah, that'll be an issue for sure, but it can be mitigated by the use of some solid encryption routines and boundary-monitoring, i.e. to make sure that a Bad Guy(tm) doesn't hack the controller and tell all the vehicles in its area to all speed up to 100mph or whatever. Or to tell 1/2 to speed up and 1/2 to come to a dead stop. Some things shouldn't be able to be overridden, such as max speed and collision avoidance.
In short, I think autonomous vehicles would be better (and probably safer) if they not only thought for themselves, but also were receiving some sanity-checking and guidance info specific to the road or area they travel on.
Re: (Score:2)
AFAIK, many new cars already do this, and they have even more rudimentary "AI" than the Teslas.
Lazy sourcing: https://www.google.com/search?... [google.com]
Re: (Score:2)
*Move fast and break things
Re: (Score:2)
Autonomous vehicles will never take off if the prerequisite is to first create a centralized traffic control system. What I think we'll see instead is autonomous vehicles taking cues from human driven vehicles via the V2V (vehicle to vehicle) communication system that will roll out in the next few years.
Tesla's vehicle already kinda does this. They used the car's GPS units to build a map of highway lanes, based on how the humans drove them, and then feed this map into their autonomous system. One can easi
Re: (Score:2)
FUD ..... (Score:3)
The truth of the matter is, Tesla pretty much HAS to come out swinging, defending its self-driving technology, or else it's easily "game over" before it even really gets started for them. Somebody had to release the tech for the general public to use first, and Tesla took the chance. (The other car manufacturers have been far more conservative with things, offering only "emergency braking / collision detection and avoidance" or just parallel parking assist.... individual components that would make up a "self driving car".)
That said? I agree with the folks here saying his stats are way off the mark and unrealistic. Since you can't even use his technology right now when not on a highway, it's not even an option for saving any lives in collisions that happen on smaller roads.
I think it was Mercedes or maybe Audi who commented that the Tesla system uses cameras and computer AI to determine if something is in the car's way. Their system used radar in conjunction with cameras, which sounds superior to me.
Re: (Score:2)
Tesla doesn't HAVE to come out swinging. That's just Elon's way. He's a huge whiner. And it doesn't ever seem to matter if someone's criticism of his product is valid or not, just whines away like a little baby. Embarrassing, really.
Doing the math (Score:5, Informative)
It's all marketing hype and mere armchair statistics.
Fortune doesn't know how to do the math, I don't know how to do the math, Musk doesn't know how to do the math, but perhaps a few readers of this comment could do the math.
It would take 275 million miles of autonomous driving to have any confidence at all that an autonomous car is safer than a human driver.
Ars Technica reported on it [arstechnica.com], and if you want to see the math, the RAND corporation, who are kind of experts at the math, have a detailed report [rand.org] available, which explains the math.
Basically, while the marketing engine can claim that autonomous driving is safer, it's not even possible to have any proof of it within any reasonable level of statistical confidence.
I mean, sure, we try to make driving safer, and assisted driving may help, but please, let's be realistic about where we're at.
Re: (Score:2)
Re: (Score:2)
Can we at least get some data on the number of people who drown [thenationa...awyers.org] in their cars vs. a floating Teslas [arstechnica.com].
marketing genius (Score:2)
Yeah, someone who's last name rhymes with "tusk".
Holy shit. And if they took a train instead, it could save approximately ONE MILLION LIVES.
Seriously, I understand that in an age of Martin Shkreli and Star of David dogwhistles, subtlety
It's the eternal technology problem (Score:2)
Powerful tools in untrained/stupid hands.
That's the problem every time, be it with computers, nuclear fission, cars or whatnot.
Look what people are doing with the Tesla "Autopilot". Pure and utter reckless fooling around. No wonder people die.
From all we know it's pretty certain the man was watching a f*cking DVD while being the responsible handler of an automobile.
That alone should cost you a drivers licence for a lifetime!
I'm glad he only killed his own stupid self and not somebody else. That would've b
Re: (Score:2)
It's much worse than that.
Highway driving is by far the safest, about 1/4 the death rate of all driving. So for the USA highway driving death rate is about 1 per 380 million miles.
Based on current data, everybody driving Teslas would more than double the highway death rate.
Fix the design flaw, but still relatively safe (Score:2)
Before we talk about numbers let's just say that yes the car should have stopped or at least applied breaks and tried to stop Period. That it doesn't recognize an elevated obstruction as a road hazard is a design flaw. An automated car or automatic braking system should see the path in front of it and brake. That the Tesla appears to disregard anything above hood level needs to be fixed either in this model or the next. The car needs to know its clearances and be able to sense its surroundings and proj
Re: (Score:2)
I've heard multiple claims that this driver has repeatedly posted films of themselves completely ignoring the road while autopilot is engaged, (even taking naps), and was busy with their phone during this accident. If true, it seems very likely that they weren't paying any attention to the road this time either, so it's quite possible that had they been paying attention they could have stopped safely.
Basically - this was one of the worst-case irresponsible drivers using a driving aid as though it were a fu
Re: Fix the design flaw, but still relatively safe (Score:2)
To me if it doesn't make the problem worse, but can with more experience and improvements make it better then it makes sense to deploy and learn from and improve.
Seat belts killed people. Saved a lot of people too. But at some point we realized lap belts could do more harm than good. We didn't get rid of seat belts, we installed shoulder belts instead of just lap belts. Then seat belts plus air bags.
This was a tragedy to be learned from in order to improve safety not go backwards.
Re: (Score:2)
Re: (Score:2)
Anyone have a link to a article on the other one? I can't seem to find any other fatal incidents involving the autopilot.
Re: (Score:2)
Re: (Score:2)
While it really should to be able to detect a semitrailer parked in front of it AFAIK no one died in that incident.
That's not a bug it's a feature (Score:2)
Re: (Score:3)
He has a bad case of the 'Chomskys'!