"The instructions signaled to motorists that automated vehicles would not be a Wild West where companies can try anything without oversight, but were also vague enough that automakers and technology companies would not fear overregulation."
That ought to set back automated car evolution by twenty years. It's always fun to watch lawyers trying to write rules for industries in which the technologies change before the ink dries in the federal register.
Government regulation of the Internet has been sparse and technologically-neutral. The CDMA Safe Harbor comes to mind. Not to mention the fact that Congress has gone to great lengths to deregulate any part of the Internet that changes rapidly. Is there an example you're thinking of?
This is a hard problem because driving software is so complicated with many special cases. Life and safety code in general is hard.
Do you want to mandate they apply aerospace coding technology? Like the other guy said, this could set them back 20 years. Mandate a traceability culture like ISO9000? How about third party audits? Escrow the code and mandate telemetry for NTSB to examine after an incident?
Even if there are a few bugs, they'll still be safer than humans. Google's up to 1.5 million miles driven with no fatalities. The human metric to beat will be 1.2 fatalities per 100M miles.
Google's own designed cars are, not the Lexus models. But their stats are junk stats in another way: Test drivers simply intervene super regularly to avoid any sort of autonomous incident.
This is a very hard problem to solve. The longer the interval between interventions the bigger the chance that the 'driver' will not be able to intervene in time because they no longer have situational awareness.
I think the best judge of self-driving aptitude, though, would be "average contiguous self-driving miles driven". The total miles driven figure doesn't mean anything if a driver has to intervene every time you approach an intersection or something. Every time the test driver intervenes, reset the clock.
> Google's up to 1.5 million miles driven with no fatalities.
Has the Google car driven on a Bulgarian road late at night, rain pouring down, with no road-markups, with lorries coming your way and blinding you every couple of minutes? I did that yesterday, and after those moments when I was trying to guess "what is the actual road?" I was thinking that there's no way in hell a computer program would be able to manage that unless full-AI becomes a thing.
I did that last night in the USA. Well it wasn't raining -- but an unpaved, washboarded road with no lane markings of any kind, blind hill crests and turns, and just barely wide enough for two cars to pass.
While it's certainly hard - and remember, people don't manage that either, at least not as safely as we will require self-driving cars to operate - I don't see why it'd need "full-AI". Recognizing the road is not a matter of general intelligence, it's a matter of having very good but very specific recognition abilities.
Besides, remember that a self-driving car won't have to operate with just a couple of eyes stuck inside a cabin. Lorries won't blind radar.
Do you have any idea how self driving cars work? They currently have sensors that are much different from your eyes behind a windscreen, they don't get blinded by light and rain doesn't bother them.
One thing I find surprising is that most geeks seem to be all for Self-Driving cars, yet are heavily against government surveillance. Yet self-driving cars are not only going to offer up much more surveillance data, but also offer a potential 'kill switch' backdoor that would allow nefarious agencies to drive you into a brick wall at 80mph and make it look like an accident.
Once manual cars go away completely, you will not only loose control of your data, but also the certainty that your motor vehicle isn't going to act against you.
I suppose this depends. If you have open source technology that isn't reliant on a cloud service to figure out how to drive, potentially we could be okay. Though a lot of companies are treating cloud tracking as a given right now.
If the security services target you, and truly wish to kill you, current methods will more than suffice.
The Israelis like magnetic mines, the Russians polonium, the Yanks... Well to be perfectly honest I'm not sure, but it almost certainly involves some combination of clandestine airpower, special forces operations, OCO, and morbid obesity.
Right now, the chain of devices that can track and kill you is enormous, from your router to your standard ICE Jeep. You're most likely already being tracked on traffic camera, the electronic toll system, cellular towers, passing Wifi, and the endless set of embedded Linux devices that you don't even consider.
If you are targeted, there may be many obstacles to your removable from the board, including political, economic, or legal ramifications.
I fully agree. The crux of my argument is that self-driving cars aren't inherently any more hackable or trackable than existing vehicles with existing available data connections through cell, emergency communications, OTA updates, or usb if local.
One could, in theory, use the fly-by-wire steering, traction control, or ABS systems to simulate a pretty convincing crash. It would be fairly straightforward electronically through the onboard processors in existing cars, and would be practically untraceable.
Yes, I would say a car that can do anything from centrally provided instructions is inherently more hackable than a car that is disconnected from everything and can't do much more than keep its speed on the highway.
That's fair. But what I see is that the car fleet will head towards OTA no matter what we do.
Forgive me for my cynicism. If you've ever worked with embedded systems... Well the state of consumer security is so bad I'd be shocked if someone hadn't already been killed in this way.
The Bluetooth alone is a nightmarish enough vector already. And local installation is very possible.
> The Israelis like magnetic mines, the Russians polonium
I don't understand how either of those work, but they don't sound discreet. The ability for a state to assassinate people at will and make it look like an accident is entirely different from "yeah, but the government could also shoot you with a drone at any moment".
The magnetic mines are exactly what they sound like. Someone drives by on a motorcycle, slaps an explosive on the side of a vehicle, and drives away. Not very subtle, but apparently very successful in taking out Iranian nuclear scientists.
Polonium is actually fairly clever. It's a radioactive material that acts as a slow-moving poison.
It took a while for Western physicians to diagnose, so many domestic leaders, political dissidents and even foreign heads of state (Arafat may have been killed in this manner) fell victim to it.
The "kill switch" is more likely a way to control the movement of the population. With the flip of a switch, whatever areas can effectively become "no-go" zones.
Surveillance will probably be a big factor too. The government will surely know who's in the car and where they're going...all in the name of "safety" or something.
It's no wonder that government types are so gung-ho for this...all in the name of "safety" of course.
I'm a proponent of self-driving cars, but I do really struggle with this. Nefarious governments are one concern, but my larger concern is the complete joke that is computer security in 2016. For the most part this means the worst case is people lose some private data; bad news, certainly. But car companies have already proven they can't handle security, and now we're building Internet functionality into cars. What happens if some sociopath like Auernheimer decides to kill a few people for laughs?
Will they kill more than 16000 per year? Because that's how many it would take to make it worse than today, even assuming that self-driving cars only halve the current numbers.
I agree with you, but I have confidence that those security issues can be resolved -- by just removing the network connections, cellular backhauls, etc. if necessary -- while the alternative represented by the status quo has basically ceased improving.
We've spent decades with passive, and more recently active, safety improvements, and the results have been good ... but in the last few years the rate of improvement in fatalites-per-mile has tapered off. It seems that all the low-hanging fruit is basically gone from that particular tree, and we may well be pushing into a zone where cars are perceived to be safe enough that further safety improvements might lead, perversely, to less safe behavior (the "football helmet" problem). At the very least, making modern manually-controlled vehicles significantly more safe seems like a very serious challenge.
On the contrary, we know that autonomous vehicles can be much more safe than manually-controlled vehicles, just by virtue of not being run by a person who may be drunk, high, emotionally impaired, tired, distracted, medically incapable of driving, etc. Right there you remove a bunch of leading causes of crashes. Given our failure to eliminate those problems completely (we made some progress on drunk driving, sure, but not much on the others, IMO) given decades of effort, we should consider that building secure computer systems might be a more tractable problem.
It will take a lot of effort to ensure that the car manufacturers don't do obviously boneheaded things (like allow over-the-air software updates), but I have more faith in our ability to quickly fix machines than to change human behavior.
Self driving cars are no easier or harder to track than normal cars. They don't need constant over the air connections to some central server, so it's just the standard license plate readers or built in tracking systems.
PS: On star does everything your suggesting and it's already in million of US cars.
To be honest I'm less concerned with nefarious governments, since they already have countless ways to track you and kill you if they really want to. And for the most part these days they don't even have to be discrete about it because of how little ability people have to push back.
I'm more concerned about what corporations are going to do when they have complete control over your movements. Especially with someone like Google getting in on it. They'll have control over what information you have access to, your communications, all your data, and now also your ability to move around. And once they have this data the US Government (at least) will have complete access to it too.
Although to be honest I'm even less concerned about that then I am the big problem. What happens when our countries economy completely collapses because a huge chunk of the current workforce is no longer needed? And I'm not talking utopian UBI day dreaming solutions. I mean real, tangible, short term solutions to dealing with 20%+ unemployment? In a country that can't even get behind something like universal healthcare ...
> I mean real, tangible, short term solutions to dealing with 20%+ unemployment? In a country that can't even get behind something like universal healthcare ...
We'll do the same thing we did in the 30s when unemployment threatened to become socially destabilizing: we'll create work for those people. Either in the form of some WPA-ish scheme, or mandatory military/national service, perhaps. In the US it will likely be the former, because there's a sort of social collective memory of the concept in the context of the Depression. Having done it once basically greases the skids involved in doing it again.
What I think people don't really appreciate is how mild recent "depressions" have been relative to the 1930s; we haven't seen 20+% unemployment in this country since WWII, so lots of things seem politically impossible. They aren't; it's just that the necessary conditions for them don't exist right now. But if we were facing the same set of problems again, there's no reason to think that we wouldn't reach for the same set of solutions.
Though personally, I don't see any of the automation technologies that are on the horizon right now leading to 20% unemployment. Most automation technologies have typically underperformed expectations in terms of headcount reduction, with niche exceptions in certain industries. I suspect automation will follow that pattern: you'll see a few industries or job functions get decimated (long-haul truck drivers, potentially), a bunch of new jobs get invented (remote backup driver, automation systems technician, etc.), and a bunch of other jobs get reshuffled, expanding or contracting based on costs (lower transportation costs might lead to more shipments, more need for loading/unloading workers, supply chain management types, etc.). Plus, the time horizon on many of these systems is approaching a significant fraction of a person's career. Someone in middle age, driving an OTR truck today, probably won't have too many problems riding that particular horse all the way into the sunset of their retirement (although at the end of their career they might feel a bit like John Henry, looking for those particular jobs that the machines have a hard time with). Someone who's 18 and looking for a career ought to beware, though. And it would be fair to ensure that public funds for education and job training are matched to expected job availability in the future -- it seems too often that we train young people for jobs that are on the verge of disappearance, and then blame them for taking the bait. That obviously shouldn't happen.
Very rarely to things have upside with no downside. "Geeks" are clearly aware of both the upside of autonomous vehicles, and the downsides presented by advanced, always-connected tech and the monitoring that comes with it. Thus, the advocacy for privacy, strong encryption, no backdoors, and the fight against government surveillance is the responsible approach to advocating for a bright, advanced future where tech serves us, rather than consumes us.
Maybe we will end up with the highways portrayed in tek war and other scifi, where you don't do the driving on limited access roadways but can or have to do it on side streets.
it certainly would be easier to mark visually and electronically limited access routes like interstates. that in turn would make it simpler to create safe self driving cars.
still I am not impressed as to where the tech as implemented on roads with current cars is today. they only work in ideal situations, the exact opposite of most safety equipment. its obvious they cannot rely on visual acquisition of objects, Tesla's system failed spectacularly in that area, so what will win out? RADAR? Laser? Combination of both?
45 comments
[ 0.23 ms ] story [ 97.1 ms ] threadThat ought to set back automated car evolution by twenty years. It's always fun to watch lawyers trying to write rules for industries in which the technologies change before the ink dries in the federal register.
Do you want to mandate they apply aerospace coding technology? Like the other guy said, this could set them back 20 years. Mandate a traceability culture like ISO9000? How about third party audits? Escrow the code and mandate telemetry for NTSB to examine after an incident?
Even if there are a few bugs, they'll still be safer than humans. Google's up to 1.5 million miles driven with no fatalities. The human metric to beat will be 1.2 fatalities per 100M miles.
I thought Google's speed limit was 25 mph? What's the human metric for fatalities at 25 mph?
Has the Google car driven on a Bulgarian road late at night, rain pouring down, with no road-markups, with lorries coming your way and blinding you every couple of minutes? I did that yesterday, and after those moments when I was trying to guess "what is the actual road?" I was thinking that there's no way in hell a computer program would be able to manage that unless full-AI becomes a thing.
Besides, remember that a self-driving car won't have to operate with just a couple of eyes stuck inside a cabin. Lorries won't blind radar.
Once manual cars go away completely, you will not only loose control of your data, but also the certainty that your motor vehicle isn't going to act against you.
The Israelis like magnetic mines, the Russians polonium, the Yanks... Well to be perfectly honest I'm not sure, but it almost certainly involves some combination of clandestine airpower, special forces operations, OCO, and morbid obesity.
Right now, the chain of devices that can track and kill you is enormous, from your router to your standard ICE Jeep. You're most likely already being tracked on traffic camera, the electronic toll system, cellular towers, passing Wifi, and the endless set of embedded Linux devices that you don't even consider.
If you are targeted, there may be many obstacles to your removable from the board, including political, economic, or legal ramifications.
But technical means will not be one of them.
I'm not sure either, but setting up a traffic accident wouldn't attract too much scrutiny today.
One could, in theory, use the fly-by-wire steering, traction control, or ABS systems to simulate a pretty convincing crash. It would be fairly straightforward electronically through the onboard processors in existing cars, and would be practically untraceable.
Forgive me for my cynicism. If you've ever worked with embedded systems... Well the state of consumer security is so bad I'd be shocked if someone hadn't already been killed in this way.
The Bluetooth alone is a nightmarish enough vector already. And local installation is very possible.
After all, all cars do need servicing...
I don't understand how either of those work, but they don't sound discreet. The ability for a state to assassinate people at will and make it look like an accident is entirely different from "yeah, but the government could also shoot you with a drone at any moment".
Polonium is actually fairly clever. It's a radioactive material that acts as a slow-moving poison.
It took a while for Western physicians to diagnose, so many domestic leaders, political dissidents and even foreign heads of state (Arafat may have been killed in this manner) fell victim to it.
Surveillance will probably be a big factor too. The government will surely know who's in the car and where they're going...all in the name of "safety" or something.
It's no wonder that government types are so gung-ho for this...all in the name of "safety" of course.
https://en.wikipedia.org/wiki/List_of_motor_vehicle_deaths_i...
We've spent decades with passive, and more recently active, safety improvements, and the results have been good ... but in the last few years the rate of improvement in fatalites-per-mile has tapered off. It seems that all the low-hanging fruit is basically gone from that particular tree, and we may well be pushing into a zone where cars are perceived to be safe enough that further safety improvements might lead, perversely, to less safe behavior (the "football helmet" problem). At the very least, making modern manually-controlled vehicles significantly more safe seems like a very serious challenge.
On the contrary, we know that autonomous vehicles can be much more safe than manually-controlled vehicles, just by virtue of not being run by a person who may be drunk, high, emotionally impaired, tired, distracted, medically incapable of driving, etc. Right there you remove a bunch of leading causes of crashes. Given our failure to eliminate those problems completely (we made some progress on drunk driving, sure, but not much on the others, IMO) given decades of effort, we should consider that building secure computer systems might be a more tractable problem.
It will take a lot of effort to ensure that the car manufacturers don't do obviously boneheaded things (like allow over-the-air software updates), but I have more faith in our ability to quickly fix machines than to change human behavior.
PS: On star does everything your suggesting and it's already in million of US cars.
I'm more concerned about what corporations are going to do when they have complete control over your movements. Especially with someone like Google getting in on it. They'll have control over what information you have access to, your communications, all your data, and now also your ability to move around. And once they have this data the US Government (at least) will have complete access to it too.
Although to be honest I'm even less concerned about that then I am the big problem. What happens when our countries economy completely collapses because a huge chunk of the current workforce is no longer needed? And I'm not talking utopian UBI day dreaming solutions. I mean real, tangible, short term solutions to dealing with 20%+ unemployment? In a country that can't even get behind something like universal healthcare ...
We'll do the same thing we did in the 30s when unemployment threatened to become socially destabilizing: we'll create work for those people. Either in the form of some WPA-ish scheme, or mandatory military/national service, perhaps. In the US it will likely be the former, because there's a sort of social collective memory of the concept in the context of the Depression. Having done it once basically greases the skids involved in doing it again.
What I think people don't really appreciate is how mild recent "depressions" have been relative to the 1930s; we haven't seen 20+% unemployment in this country since WWII, so lots of things seem politically impossible. They aren't; it's just that the necessary conditions for them don't exist right now. But if we were facing the same set of problems again, there's no reason to think that we wouldn't reach for the same set of solutions.
Though personally, I don't see any of the automation technologies that are on the horizon right now leading to 20% unemployment. Most automation technologies have typically underperformed expectations in terms of headcount reduction, with niche exceptions in certain industries. I suspect automation will follow that pattern: you'll see a few industries or job functions get decimated (long-haul truck drivers, potentially), a bunch of new jobs get invented (remote backup driver, automation systems technician, etc.), and a bunch of other jobs get reshuffled, expanding or contracting based on costs (lower transportation costs might lead to more shipments, more need for loading/unloading workers, supply chain management types, etc.). Plus, the time horizon on many of these systems is approaching a significant fraction of a person's career. Someone in middle age, driving an OTR truck today, probably won't have too many problems riding that particular horse all the way into the sunset of their retirement (although at the end of their career they might feel a bit like John Henry, looking for those particular jobs that the machines have a hard time with). Someone who's 18 and looking for a career ought to beware, though. And it would be fair to ensure that public funds for education and job training are matched to expected job availability in the future -- it seems too often that we train young people for jobs that are on the verge of disappearance, and then blame them for taking the bait. That obviously shouldn't happen.
it certainly would be easier to mark visually and electronically limited access routes like interstates. that in turn would make it simpler to create safe self driving cars.
still I am not impressed as to where the tech as implemented on roads with current cars is today. they only work in ideal situations, the exact opposite of most safety equipment. its obvious they cannot rely on visual acquisition of objects, Tesla's system failed spectacularly in that area, so what will win out? RADAR? Laser? Combination of both?