U.S. eliminates human controls requirement for fully automated vehicles
reuters.com
reuters.com
The only thing here is that the government rewrote the rules to no longer assume those things. It's not saying those things aren't necessary, just not required. If the car can meet all the safety standards without it, then that's fine.
It would be like saying today that you can't have a computer without an on/off switch because in the 70s every computer had a physical switch.
As aside, that’d be great.
And besides, I almost never used the switch on the computer anyway. I was always told to put it into a power strip and use that, because it was a lot cheaper to replace the power strip than the broken power switch on the PC!
Instead, I've turned on the feature in the BIOS to boot up when power is applied. Now, to turn on my desktop, I just flip the power switch on the back. Feels satisfyingly retro.
Anyway, I'd rather avoid accidents.
I had a Dell that once got stuck and the usual long press of the power button was not switching it off (probably it was sw/fw controlled). I had to take it apart and disconnect the battery, otherwise I would have lost my working day waiting for it to die.
P.S. Yer 'bout to miss your flight."
Actually, removing the steering wheel is probably a crash safety gain. Manufacturers had to invent a way to retract the steering wheel in crashes since it caused so many torso injuries. Steering wheel injuries were also the spur to invent and deploy the first airbags.
I assume such countries changed the laws to remove that requirement once EVs were no longer a novelty.
With the presence of an automation kill-switch and manual human controls the driver can always take back control.
Without those, the driver is at the mercy of the hacker.
Vehicle automation falls under the SCADA sub-domain of cyber-security. SCADA stands for Supervisory Control and Data Acquisition, and the cyber-security in SCADA systems is light-years behind. Pro-tip: going into cyber and want a good certification (if such a thing exists)..then get CISSP-SCADA.
Some relevant articles:
[1] https://physicsworld.com/a/how-to-hack-a-self-driving-car/
[2] https://hackernoon.com/how-to-hack-self-driving-cars-vulnera...
[3] https://blog.tesu.edu/should-we-be-worried-about-the-hacking...
Isn’t this envisioning a car with no passengers who can drive? Agree there needs to be a physical fallback. But fallback to a parallel system running on ROM might be the safer option.
Admittedly computer systems have come a long way in ten years, but given the attitudes and technical expertise of a lot of the people I worked alongside, I would say there's a reasonable chance the needle hasn't moved very far.
They're excruciatingly good at getting oil out of the ground and getting it to market, but they're not going to be winning capture the flag at Defcon or Blackhat any time soon...
Which raises the question: "Why would we engineer vehicles to accept external network connections?"
Any communication to/from a vehicle should only be initiated by the vehicle itself. There's no reasonable criteria under which a vehicle should have its systems available to external sources.
Or am I missing something important?
We absolutely want to remotely control cars at some point, how else will you send a self-driving taxi to a customers or "ask" a car to comeback to the garage for a recall on the underlying hardware or for general maintenance?
But any connection poses vulnerability - for example, security certificates to be sure you're talking to the right person become obsolete over time as processing power gets cheaper, and an update can fail and brick the car.
Exactly. Which is why vehicles (like most other IOT devices) shouldn't connect to the Internet at all.
Please define a scenario where internet access would be required (i.e., not replaceable with other, less vulnerable communications mechanisms) for a self-driving vehicle.
I can't think of any.
Sure that means maybe waiting two minutes for the car to move, but that's not a huge deal, especially if you know that happens and request it two minutes in advance, but it gives anyone in the car the chance to cancel the request if it's invalid.
The point is that the car won't just take remote instructions but will gather them and allow the occupants the chance to override them.
Sure, that single server becomes a juicy target, but that's where the defense in depth comes in, where the car still doesn't execute the command until the user accepts it.
Sure, client/server certificates are great. But as soon as there's a BGP hiccup/attack, a fire at the relevant data center, a moron with a backhoe, among other eventualities, you're screwed.
Autonomous vehicles (as well as those with drivers) need to be, as the name says, autonomous. Not dependent on some online resource.
I'll say it again: Vehicles should not be dependent/accessible/updateable over the Internet. Full stop.
They could know about it the way humans do: get stuck in traffic or stuck at a road closure and just burn time waiting for the issue to correct itself.
But wireless radio communications allows us to do so much better.
When people think about attack surfaces in devices they often think about the things making repeated connections out. The thing is, every piece of reachable mutable code in the device is an attack surface, not just the software making connections.
You may have heard of ATM car collectors that attach to real ATMs, or people that have hacked an ATM to collect cards. Your smart car connects to a number of systems throughout its life: your dealer's diagnostics, every time you charge, every time you play music through the bluetooth connection on your phone. Each one of these is a potential attack vector.
Let's say one of these things is using JAVA, and somehow does not have an updated log4j [1]. Now that thing can be used to compromise other subsystems of your car. And these attack vectors do not have the antivirus and other protections your desktop computer or laptop do.
As bizarre and backward as it sounds, I think the best bet is burned immutable firmware on a physical module that can be easily swapped out by an authorized dealer (or with a key of some sort to allow people to service their own cars). This would reduce the vulnerability quite a bit, though not eliminate it. It also would eliminate over the air firmware updates, which are insane in my opinion. It's kind of how if you want a more secure computer you burn a live CD with a minimal hardened distro of linux, rip out the hard drive, make sure you have a fast CD, and have the computer boot to CD. This is part of what I used to do for a DMZ PC. There's always a risk your firmware will be hacked, but that is harder, and can be checked on startup depending on what you have on the system [2].
[1] https://www.trendmicro.com/en_us/research/21/l/examining-log...
[2] https://crypto.stackexchange.com/questions/59965/tpm-and-rem...
You're suggesting that if a self-driving car has a feature to remotely summon it, such a feature should need a passenger inside to click "accept"?
Especially if it's being summoned for some sort of service. I'd argue no vehicle should be on the road sans occupant to be honest.
What? You think that it's totally okay to have a vehicle titled to you auto -accept a request that could be coming from any moron out there?
How many levels of optimism are you on?
So someone napping in the passenger seat is fine, but empty isn't? Why?
> Especially if it's being summoned for some sort of service.
Especially? When it's doing some sort of service is when it's the most ridiculous to have someone hanging around doing nothing.
If you don't want a summon button, just say so. Don't say summon should need a passenger.
> You think that it's totally okay to have a vehicle titled to you auto -accept a request that could be coming from any moron out there?
I didn't say that.
I don't think even a ride share would go to anyone, but I wasn't even suggesting that, I was talking about any kind of summon at all, so by default that would be the owner.
Who is this "we"? You might, but I don't.
That's a disaster waiting to happen.
What's more, under the circumstances you describe, there's no reason such vehicles need to be internet accessible. Or have we forgotten that communication doesn't require the internet?
> What's more, under the circumstances you describe, there's no reason such vehicles need to be internet accessible. Or have we forgotten that communication doesn't require the internet?
What kind of remote communication these days doesn't require the internet?
Text messages?
Special short range base stations with 56k modems?
Satellite phone?
Pretty much any low power radio application (CB/HAM radio, key fobs, walkie talkies, POTS lines and dozens more), or higher power (although those require FCC licenses) don't require the internet.
Besides, I'm referring to the software that makes the vehicle operate. That software/system should never be exposed to remote sources. Otherwise, you're pretty much begging to be locked in your car and taken on a joyride which ends with you rolling off a cliff.
> Otherwise, you're pretty much begging to be locked in your car and taken on a joyride which ends with you rolling off a cliff.
If hacks while driving are your concern, you could have the car isolate from the internet when the key is inside it and/or someone hits a stop button. If hacks before driving are your concern, I don't think any of your suggested radio applications are safe either. And I don't really see why half of them are better than an internet connection in the first place.
The difference is that there aren't 25,000 (conservative estimate) people looking to hack/scam/destroy stuff on local radio channels.
Which means that someone would need to target me specificly, not run across the poor/no security on the internet link from my car while scanning the 'net.
Besides, the car proper (ICE/electric motor, steering, brakes, etc.) and the software that runs it shouldn't be accesible by anyone unless they have both physical access to the car and a reasonable amount of security (key fob, password, etc.) to be allowed to muck around with the car's operating software.
Infotainment systems, sure. But those need to be on separate, air-gapped hardware.
If they don't have significant range then they don't solve the problem.
Right. Which is another reason not to allow access to the "make it go" software unless you have physical access and authorization.
Where are the “low-chip-count” new cars with VFD radio displays, dials, traditional keys, no screens and especially no cellular modem?
You state the security risks of connected vehicles, so I’m not sure why you lead with these incorrect statements.
Where is the requirement (or even a good reason) that a vehicle be connected to the internet?
Which is the same question to be asked about ovens, toilets and a raft of other stuff too.
I'd say "none."
Maybe in your vehicle. But certainly never in mine.
Any vehicle whose software can be remotely modified is one I'll never buy.
As I said, that's a disaster waiting to happen.
What? To install nightly builds? I think not.
I wouldn't put software into production (or even beta) that required weekly updates. And that's just for software to make a website or other app go, with no risk of injury or death.
Why would anyone want to be driven around in a vehicle by software that's so buggy it needs weekly updates?
How often do you update the firmware in production systems? And a commercial self-driving car definitely needs to be production quality, no?
There are plenty of reasons. Whether or not they are good depends on your opinion, but for me at least good reasons include:
* Triggering my home alarm if someone interferes with the car while it's parked in my driveway.
* Remote tracking/disabling if stolen. Not actually that useful directly but if most cars have this then fewer people will steal them.
* Automatically notifying emergency services in the event of an accident (you could argue this doesn't need internet, but 5G is IP based).
* Being able to get OTA software updates.
* Being able to listen to Spotify and use Google Maps for navigation.
* "Smart insurance". Not everyone's cup of tea, but regular insurance for me is £100+/month, so being able to cut this by 50% by remotely verifying that I drive like an old lady is a pretty good deal to me.
My car is old enough not to have an internet uplink, so instead I have an aftermarket head unit and an OBS-II 4G dongle thing. Obviously it doesn't do all of the above, but it's useful nonetheless.
Why should that require the Internet? An analog radio can do exactly that.
>* Remote tracking/disabling if stolen. Not actually that useful directly but if most cars have this then fewer people will steal them.
This is the worst reason. As we've seen, this has led to disabling of cars on the highway by auto finance companies.
What's more, no software that actually controls the operation of a vehicle should be remotely accessible -- to anyone.
>* Automatically notifying emergency services in the event of an accident (you could argue this doesn't need internet, but 5G is IP based).
That's not a bad idea. But why should something like that be linked to the driving/operation software?
>* Being able to get OTA software updates.
That's the worst idea. No software in a vehicle should allow remote updates. That's a disaster waitiing to happen.
>* Being able to listen to Spotify and use Google Maps for navigation.
Why should such functionality even be linked to the operational software of a vehicle (self-driving or otherwise)?
> An analog radio can do exactly that.
An analogue radio can do anything that an internet connection or digital radio does, but it's sort of redundant to reimplement all the functionality that modern IP gives you just because you don't like the internet.
In particular, my alarm doesn't have any programmable analogue radio, so I'd presumably need some sort of module to interface with my alarm or to change my alarm for one that supports the analogue radio that the car alarm uses. With the internet-based approach, I have a little script that reads my car's API and triggers my home's alarm with its API.
I can also do things like sending me a push notification on my phone if something happens to my car when it's parked in a public car park.
Also, analogue radios aren't necessarily safe either. In most cases, digital technology has better encryption as it's easier to implement. In the UK, signal amplification attacks are becoming an increasingly common way to steal cars, which obviously wouldn't be possible using an internet-backhaul. Digital also tends to use less power and is less prone to interference.
> no software that actually controls the operation of a vehicle should be remotely accessible -- to anyone.
Remote disabling of cars in operation is obviously not good. However, MOST cars implement lockouts by having it so if you turn them off they can't be turned on again, which is a lot safer - and works more like an immobiliser which is a well proven technology. Both this and immobilisers sometimes have faults, but it's probably one of the least likely things to cause a fatal accident on a car.
> But why should something like that be linked to the driving/operation software?
Well I wasn't arguing strictly for linking to driving software, just reasons to connect a car to the internet. But I think the reason these systems have a connection to the operational software is to detect what's actually happened to the car - has the airbag gone off, what was the speed of the impact, what do the sensors say, etc. I don't believe there is any intent that they can control the car, but as soon as you link anything digitally there's always the possibility.
> That's the worst idea.
For most people, it's a great thing. I'm not sure there's ever been an attack caused by a remote update system in a car, but at least in computers, remote auto-update has been one of the best things for security and stability in the last few decades. There are a ton of theoretical issues with it, but in practice the benefits outweigh them.
Imagine your car's fly-by-wire software is discovered to have an unintentional acceleration bug. Remote updates means everyone's car is patched immediately. Traditional updates mean maybe if you bother taking it in for your annual service to an official dealer, it might get patched if they remember. You might be one of the 0.001% who would put it on a USB pen and patch it yourself or obey a recall, but almost nobody else would bother.
Who says I don't like the Internet?
Quite the opposite, in fact. But not all communication should be done via that network of inter-connected networks.
No wireless (and in some cases, even wired) communication can ever be completely secure.
When there's a device (in this case a vehicle) that can cause injury, death, destruction and/or other mayhem, taking steps to ensure that it can't be controlled remotely, as long as humans are in the mix, is a necessity not a burden.
You may be happy to put your life in the hands of a script kiddie maliciously updating your vehicle's operating software, but I'm not.
Sure, there are many conveniences associated with allowing remote access to vehicles. But when I get behind the wheel (or potentially in a passenger seat of a self-driving vehicle), my top priority is not to die.
Allowing arbitrary connections to a device that can kill me contradicts that priority.
My cell phone can do most of those things in my current car, and also has 0 ability to steer me off a cliff.
If you're running them as a taxi service (which I think all the major self-driving companies except Tesla are), you need a way to get stats on the fleet and to direct a car to go pick someone up.
As of now, taxi services seem to be the focus of the self-driving industry because the LIDAR hardware (among a couple other things) is too pricey for consumers to justify owning. That's going to require some way to communicate with the cars. Cellular data seems a reasonable way to do that. I don't see a remarkable advantage over building their own boutique network of transmitters just to say they aren't technically on the internet.
You could theoritically still hack that.
If you hit the big red button going 65mph on the center lane of a crowded highway, it could be bad. Especially if the tech starts to feel safe enough that cars can travel close enough to use the draft from cars ahead.
I think an emergency brake works for trains because they’re isolated.
Oh wait..
The difference is just that it’s higher level controls than a steering wheel.
A "pull over" button could work, but still relies on the automated driving capabilities.
A third option would be to dial an "expert driver" who takes remote control of the car to get you to safety, bypassing both the automated driver and the passengers' inability to drive. Lack of connectivity could be an issue, and it opens the door wider for hackers (already an issue though). However I have to imagine that in-transit connectivity will be increasing dramatically once people want to work or watch Netflix during their commute.
I think it’s sensible not to optimize to the person with the most trauma.
I hope she reconsiders naming her daughter for the girl who died tragically.
My sister was named for a cousin that, as a baby, fell out a window to her death, a decade or so before my sister was born. When my sister was a couple weeks old, my parents realized that they didn't want to think of that tragic story every time they said my sister's name. So they changed her name to one without the emotional baggage.
There are other ways to honor someone's memory, that don't affect your relationship with your new child.
Spoken like someone who has never been a parent.
I do agree tho, “no steering wheel, no responsibility” should be explicit in law.
If the owner is responsible for any deaths caused by their car, the owner will have an incentive to choose a car that has a good safety record, and to keep that car in proper working order.
Conversely, if a corporation is responsible, they can only be punished with fines for causing a fatal accident, and those fines will be considered a cost of doing business. Also, the corporation will have a lot more money to invest in lobbying for laws (and judges) to make those fines as small as possible.
Let us not forget the infamous "Pinto Memo":
https://www.spokesman.com/blogs/autos/2008/oct/17/pinto-memo...
Well, I meant “no responsibility for the passengers”.
The owner should have some limited liability as you say, but primarily liability should be with the manufacturer. The way I view it AVs should have similar liability as an elevator. The manufacturer and the maintainer should have liability, but never the passengers.
Ideally, the manufacturer keeps enough sensors on the thing such that the car doesn’t move without the proper maintenance. But at that point is it truly ownership?
My preference is to just have many services competing that make them relatively cheap.
I think the real concern that the car is actually programmed explicitly to kill you if it has to, to ensure it's own safety or the safety of it's riders or something.
If a car is programmed that way, how does that benefit the car company (or the programmer)? It's just a different family that sues them if someone dies that didn't have to die. Are you thinking they could advertise it as "prioritizes your life more than random other people!" That's just begging for a lawsuit.
And they especially wouldn't program the car to protect itself above a human life. Why would you think it would? The car is not nearly valuable enough to outweigh the cost of a lawsuit.
Comparing a car programmed for self preservation versus a human seeking their self preservation only really makes sense if the car and human are both equally capable of assessing threats.
At the moment I don't think cars are at the same level. Maybe I'm wrong. Humans are fallible too of course.
But that's not what you said. You said you were concerned about its agenda, that it would be intentionally prioritize the lives of its passengers over other lives (and even suggested it would prioritize the preservation of the vehicle itself more highly)
you said....
>I think the real concern that the car is actually programmed explicitly to kill you if it has to, to ensure it's own safety or the safety of it's riders or something.
My comment was about that....the "agenda" issue, which is 100% what your comment was about.
To be clear (if I wasn't in a previous message)
There is NO incentive for it to be explicitly programmed to do that, since the company could be equally sued by pedestrians as it could by passengers.
There IS incentive for a human to behave that way, since natural selection built a self protection motivation deeply into the human brain.
There is also very strong incentive AGAINST prioritizing protection of the vehicle itself over any human -- passenger or not. For both human drivers and driving software programmers.
Anyway, you seem to be changing your argument to one of competence, which to me is a completely different discussion. (I think self driving cars are better at some things and humans are better at other things, but eventually cars will be better -- or at least equal -- for the vast majority)
And no, this isn't "junior-developer grade software routine." They also don't want the car slamming on its brakes in the middle of traffic because a plastic bag blew by in front of the car.
I don't know what to say.
We can't do that with humans; instead, we have probabilities and known badly-designed roads that incentivize (or don't disincentivize) driving in a way likelier to cause death. The only way to fix those issues at scale is redesign the roads, often for less efficient traffic throughput.
Normally I walk and have confidence the human will stop and not want the consequences, or I at least make eye contact with the human driver for reassurance they will stop
This time I ran, mumbling “self driving car fuck that”
Many cars with self-driving capability do not use Lidar, I can't tell which ones do, the very prominent Teslas do not, and they've have hit people in obviously avoidable circumstances that have nothing to do with "being too predictable in a human world", just total fuckups, and I've seen videos of what the Tesla cars see compared to what they do, and I'm not satisfied with them.
So for most they may have to build trust, for me they have to rebuild trust, because all the companies have pretty much gone the nightmare route making us all cannon fodder.
When at an intersection, you know another car sees you by looking through their windshield and judging whether the person is distracted or if they're aware of your presence.
It would really benefit motorcyclists, pedestrians, and other drivers if there were some mechanism for self-driving cars to visually and outwardly acknowledge your presence.
How often do you wave or signal other drivers with your hands when you both arrive at a stop sign at the same time? What's the replacement for this in a world where half of cars don't have a driver?
No eyes no trust. Full automation is going to elevate the precedence of fallible motor vehicles over people with every right to the road even more. Great.
The implied "locked eyes therefor I am safe" is NOT a thing. I apologize to the inexperienced if I implied that. You got to watch the whole process. But in the distribution of the MV danger types, not paying attention at all is much higher IMHO.
Everyone knows this? I had this happen on my own street that I live on with a neighbor two houses down. Yeah, I was rude.
Having an indicator that the car "sees" you would be very interesting.
https://www.businessinsider.com/jaguar-land-rover-eyes-on-ca...
Are sunglasses on a car driver unfair too?
Or those Jurassic park like vehicles
In fact, many of the other companies in the space are looking down their nose at Tesla's experiment because Musk thinks he can do the trick without LIDAR. He might not be wrong (humans do it without LIDAR), but it's a hell of a gamble that results in a sensor bundle quite different from the sensory tapestry other self-driving vehicles can take advantage of.
Couple that with the fact that other companies are employing professional drivers to test their hardware on public roads while Musk is crowdsourcing it to anyone rich enough to buy one of his cars, and one can get the sense why those who are actually concerned about driver and pedestrian safety are real nervous about Musk's foray into this space.
The idea that you can make a deadly self-driving car, and then jail the driver because he could not react in 0.2 seconds when your car suddenly swerved is the greatest handover of money and power to corporations since the times of slavery.
So make the choice clear - either tha car has controls, or it doesn't. And if it doesn't, there will be no temptation to blame the passenger
I think this is forcing a false dichotomy between having a steering wheel and driver liability for failures in the autonomous driving system. I think we can have a world with steering wheels and manufacturer responsibility... we just might not get there for a while.
But to the parent comment, it seems insane to me that a consumer would accept full liability for some company's software as they would for their own driving. It seems obvious to me that the manufacturer is liable. If my taxi driver gets in an accident I'm certainly not at fault.
You just need to make it impossible for the driver to take control of the vehicle unless it's fully stopped and in a safe state.
This change to the law makes it so that if someone else wants one that can't do that last 5%, it's not literally illegal to sell them that car.
> if the vehicle is autonomous, it should be illegal for it to have a steering wheel.
These levels still provide improvement in quality of life, but for a cheaper price (because it's easier to implement).
That way, the state of the switch lets you know who caused the crash.
If there was a low-end car that could drive my kids to/from school without a licensed driver present, I'd buy it in a heartbeat.
So would any other parent who has the money, and regularly has a conflict between where work would like them to be, and where they need to be for their children's school. (Yes, I'm familiar with and have used rideshare. Uber's liability won't cover taking kids, HopSkipDrive will, but they cost a lot more and every so often you get the situation where they don't find a driver. The convenience of, "Don't have to worry about it" would be worth a lot to me, and would give the children a lot of freedom.
What? Do you go to jail If someone kills a pedestrian with your car? Do you have to pay compensation for injuries?
I don't know what jurisdiction you are in, but here the one who cuased damage is the one who pays.
To provide some extra context, it's also possible to be criminally liable for what another driver does after leaving your car if they commit a crime that you had foreknowledge of. Here is one tragic example of a case where this happened:
https://truecrimedaily.com/2017/06/08/exclusive-ryan-holle-c...
"U.S. regulators on Thursday issued final rules eliminating the need for automated vehicle manufacturers to equip fully autonomous vehicles with manual driving controls to meet crash standards."
It does sound like "U.S. regulators are removing regulations.", which does look like they are acting against themselves..
There could be dedicated lanes assigned to the autonomous cars and platooning-protocols that ensure that the cars use their automatic capabilities only on relatively safe highways, only queued between other autonomic vehicles.
Should be a more predictable environment and thereby safer
I guess we can always drive with a USB-plugin gamepad if there's a glitch or a central system outage?
This would actually represent progress if they eliminated the requirement for side mirrors, which from what I've read of good aerodynamic designs is responsible for as much drag as the body of the car.
Of course fully automated vehicles will look nothing like typical cars. There's this really common tendency to think way too small, to confuse the current prototypes with a final product.
The mid-term future for SDVs looks more or less like a train compartment on wheels.
A bus?
Yes, autonomous public vehicles seem likely - like a Flexi-bus - but people like to travel without random members of the public, private hire of small vehicles will definitely be a thing unless the public get priced out of it.
This seems way too high to me. https://en.wikipedia.org/wiki/Automobile_drag_coefficient says side mirrors contribute 2-7% drag.
Still worth removing them of course, I'm just correcting you to prevent people from picking up wrong numbers.
This rules allows car manufacturers to work on new fully automous cars without controls that are designed to function as campus shuttles or taxis in a limited scope. There is still a lot of work to be done regulating the use self driving vehicles, but this removes a safety regulation hurdle that is stopping states and car makers from experimenting with different approaches.
I've had electrical failure in a car before. The alternator died and wouldn't produce enough power at under 2000 rpm. My stereo and lights were flickering and engine was sputtering. In a steer-by-wire system, I'd have lost my steering. I'm just glad I was driving a stick shift, giving me more control over engine speeds so I was able to make it home and not have to get a tow.
Regulators shouldn’t just take for profit companies’ word that a vehicle is automated. There should be a standard battery of tests which a vehicle must pass to be considered automated.
Allowing companies to get rid of controls just because they pinky swear the vehicle can handle driving safely in all foreseeable conditions is a sure way to end up with needless death and injury due to the same dynamics that were written in the book Unsafe at any Speed.
The safety and regulatory framework should exist before the activity is allowed rather than waiting until the inevitable deaths and injuries and relying on the families of the deceased to petition congress.
https://knowyourmeme.com/memes/youll-own-nothing-and-be-happ...
The WEF thinks you shouldn't anyway.
The practicality of manually driving is not going away. Just this week my neighbor was driving up and down his driveway to transport some heavy packages received.
This is like removing human control on an autonomous Russian roulette revolver that shoots to the side if it detects a person in front of the barrel, then praying for the best.
Regulatory capture means it wont be high and the potential profits to be realized from first mover advantage here means there will be a significant incentive to "move fast and break things".
So to speak.
I'd feel safe if an executive who OKs a software update that kills someone risks significant jail time. If that means no self driving cars for 20 more years then I am totally fine with that. That would simply mean that they werent safer as promised.
If executives were jailed for car accidents people we would have trains and trams in every corner.
When it comes to AI droving, noone has a fucking clue what it will do in next 50 milliseconds. Teslas literally mis-classified a semi trailer for an underpass
The opportunity for shenanighans is endless
It doesn't make sense to have universal product liability in that way: there are too many creative idiots.
It's very disingenuous to compare it to a knife.
I don't object at all to holding manufacturers accountable for defective products; but it makes no sense to say "if your product is involved in an accident, you go to jail", which was the premise of the comment to which I responded.
Citation needed.
> was the premise of the comment to which I responded.
It clearly wasn't.
The status quo is 139 people dying every day in the US in traffic accidents.
If we replaced all cars with self-driving ones, an executive could approve a software update every single day that kills only 100 and it could very well still be a net benefit to society.
These discussions always seem to implicitly presume that we're currently at zero deaths. We're not--we just don't hear about them. Every year we delay self driving cars is more people dying, not fewer.
I never presumed that and never would.
I do presume that if we watered down criminal liability for reckless driving then we'd get way, way more than 139.
The current regulatory framework surrounding self driving cars is rather like making drink driving subject to a $30 fine. I would be surprised if it starts out safer than humans.
If you have not paid your taxes, get in to your car, and it drives you to the local court-house or police station, this is a new risk, no?
More specifically, what IF or rather WHEN there is an incident that could've been prevented by human interjection, but wasn't?
Other aspect is, why are we only ever discussing these separately? What is the safety of a automation-assisted human? Can we get to the point where safety is better than either system alone?
But for far more individuals there will be a net increase in safety, moreover, since traffic accidents don't only imperil drivers, pedestrians and safe drivers will also experience an increase in safety due to average drivers using just-above-average automation.
> What is the safety of a automation-assisted human?
That's where we're at today with systems like Tesla's Autopilot.
I don't really know how to think about it. Accident rate, on a per-mile basis, falls out of some combination of "hardness of the driving problem" (freeways have many fewer accidents than city streets, I think) as well as propensity to take risks (driving at night; in bad weather; at higher speeds; while distracted) as well as driving skill.
What, by definition, people will accept is an accident rate similar to that for professional drivers; e.g. taxi/rideshare/livery. According to a report by Uber, that's about half the number of fatal accidents on a VMT basis versus the general population in their case.
Reduction of average statistical risk is what really matters and will be what is primarily responsible for driving down real world fatalities and life altering injuries.
I for one am not going to stand back and let fear of new technology and unfounded preference for human driven action over automation stand in the way of making a statistical impact on reducing the 1.35 million auto fatalities that happen globally every year. To propose that possibly reducing that number to 1.30 million is not good enough and we have to hold back rollout until the tech is better than the "best driver" is disgustingly sociopathic.
Just don't stand in the way of people who choose to use self driving cars because you're more afraid of it that of extremely lethal and incompetent human drivers.
Anyone claiming they're safer now is either selling something or doesn't know what they're talking about.
I figure deaths per million miles is a good stat
In most of the US (not where I live, thankfully) the "choice" is use a car or stay home.
Not much of a choice.
Go and move to West Memphis, TN (or, as I said, with most places in the US) and try to live without a car.
Not really a choice. Which is why I live in a place with halfway-decent (well, the best in the US anyway) public transport.
What's the choice? I don't have any idea what you mean
Absolutely. At least until you're fired from your job because you don't show up. Then you'll (depending on where you live) a few months to maintain that "choice" before the landlord throws you out on your ass because you "chose" not to go to work.
Please.
Interstate highway? I'd absolutely trust it to cross the country while I napped.
Downtown urban cores where I can expect bicyclists in the road and delivery trucks parked in a lane? Not a chance.
Anywhere that involves making an unprotected left turn? Gambling with your life, there.
1. isn't a steering wheel the only plausible failsafe control for somebody supervising these experimental machines? Not sure what other control interface would be better in life-or-death instants/moments, plus any driver can carry over their preexisting skills in that situation. If better emergency controls than steering wheels exist, fine, I guess.
2. Isn't the question not
safety(autonomous vehicle) > safety(human driver)
safety(autonomous vehicle + human control fail-safe) > safety(autonomous vehicle alone)
This Wikipedia article kind of shows the mishmash of automated/half-measures/full measures situation:
https://en.wikipedia.org/wiki/Automation_of_the_London_Under...
If there is a fully driverless (not driving itself with a human still in it) truck - as companies tell us is their goal - it does not increase or decrease safety to have or not have a seat and a steering wheel in it.
The argument here is that building an autonomous Russian roulette revolver is a stupid thing to do, which is orthogonal.
I have some experience in navigating NHTSA when bringing new automotive tech to market. The following reflects my experience.
When working on cutting edge tech that have associated hazards (cars hurt people) what companies want is a clear set of standards to meet, as this goes a long way to showing due-diligence when things go wrong and lawsuits start showing up. But with new tech there are no standards, no generally accepted way of doing things. So how do you know if the system is capable/safe-enough to deploy in volume on our roads? How do you know you've done enough due diligence to confidently launch new tech without putting your company at financial risk? Let's face it, a big enough software bug (and associated liability) could potentially put Tesla completely out of business. When dealing with hazardous new tech, the threat of recalls and lawsuits weighs heavily on a company's decision making process.
When it comes to new tech, NHTSA is in no position to answer these questions. And they know it. If they knew better then they'd be the ones driving innovation. As it stands under capitalism the innovators are all working for private companies hoping for the big pay out. Knowing this, it would be counterproductive for NHTSA to say "we know better than to let a self driving car on the road". They don't know better and they don't want to get in the way of innovation.
So instead NHTSAs job is to setup a regulatory framework and set of rules to help these companies bring tech to market, while helping identify/limit bad actors. When lessons are learned (ex. ABS systems provide a clear benefit, and should generally do X Y and Z) these lessons get integrated back into the regulations as a guide to all players in the space. The regulations and standards improve over time as more-and-more millions-of-miles are driven and the tech/algorithms mature.
In my experience, NHTSAs attitude seems to be best summarized as: Innovation is good for America and the last thing we want to do is stop you from developing cool new things. But we have no idea what your new tech should/shouldn't do. The risk is all on you, your company ultimately bears the legal responsibility for your products. You'll have to determine if your product is ready for market all on your own -- and failure carries a high price tag. Now... if/when the tech matures, and the industry seems to generally agree that "new tech X" should include "X Y and Z" as part of the safety/risk equation, then we'll add that to the regulatory framework. But until the tech matures and the clear path forward is known we don't want to stifle innovation.
This is about the best you can do in this situation. ADAS is new, and no one knows where we'll end up. No one, including you and me, knows if these systems can be deployed safely. Personally I don't trust ADAS systems, and believe the financial risk would be too high to risk my company. But what do I know?
It seems that the message being sent here is: "We're ready to start dealing with cars that don't have human controls. Best of luck to all those who dare play in the space."
The market will still decide if these vehicles sell. I wouldn't buy one, would you?
Plus, it's never too late to crack down. If there are a lot of accidents, you can bet members of Congress will thrive on the outrage.
So, there's really nothing to see here. Just the government getting out of the way, a little bit, for now.
Absolutely! In a heartbeat. Because I know it's a tool to enhance my driving, not replace me, and I still need to pay attention.
I already have an add-on for my car that gives me level 3 on the freeway (no hands on the wheel needed). I still keep my hand on the wheel though and I still pay attention, but it definitely helps because I don't have to use brain power staying in the lane or maintaining speed. I can give 100% of my brain power to looking for danger.
But the point is, no manufacturer would sell one without a wheel either because they know it wouldn't sell. This is just a change to not require one to pass safety tests. The car still has to actually pass the safety tests though.
Now, it's as simple as that!
I think we oughta take the men out of the loop.
There could definitely be a number of use-cases beyond taxis/individual cars. This is why top-tier regulations like this need to be broad. You can refine it at the vehicle category level.
The average human is often too busy texting, or eating to pay attention.
Humans do very stupid things in cars, it's a part of why I won't really consider living anywhere where I have to own one. Holy crap, you'll see people literally just rear-end each other at stoplights because you want to keep itching closer.
One of the most common complaints about automated cars is they go The speed limit, I recall once I was literally at the exact speed limit for the highway and I flicked on the cruise control. Immediately a driver behind me started honking and flicking me off.
Like holy shit I'm not from around here. I'm not going to risk getting pulled over because you're 5 minutes late to brunch or something.
The average human is such a horrible driver. I don't think Tesla bots can do much worse. The times the Tesla AI fails are very rare and far between, the times some idiot human decides they're going to rear end someone, or change lanes to late are very common.
All hell the robocar overlords
We have to do something about inequality or we will have totalitarianism. It will either be in the form of a totalitarian fascist or communist revolution, or a totalitarian police state to prevent such a revolution.
Truck driving employs 1% of the US population and if those jobs are lost to automation, there's no chance of them coming back like they did throughout other recessions. It won't just be the drivers, it'll be their friends, families and children that will be unemployed, unhappy, and very motivated.
The right does a good job guiding the anger of the working class toward issues that have nothing to do with inequality like COVID, race, immigrants, trans people, etc. Meanwhile the left offers up a totally out of touch message that doesn't resonate at all, or offers them welfare without understanding that these people consider welfare to be something to be ashamed about. They want productive work.
You know... we have a housing crisis in this country. A lot of these people could find work building adequate housing for the US population, but that would threaten high real estate valuations...
We also have a lot of infrastructure that needs building, but we may have to wait for the generation that was brainwashed to believe that infrastructure (except more roads) is communism to retire.
There are plenty of jobs but we won't create them. We'd better get on it.
1) the main expense is the land, the houses themselves are cheap, so anything that causes existing-but-undesirable houses to now be in a commuter zone increases useful supply
2) self-driving vehicles will allow people to commute while asleep or doing their hobbies (this is a terrible idea with current tech, people already get in trouble for doing both in existing self-driving cars)
3) remote working (we already have a lot of digital entertainment and socialising)
If I’m wrong, I expect it to be because all this makes meatspace location into an even bigger status symbol than it is already.
This doesn't eliminate the traffic issues and energy consumption associated with long commutes. Self-driving cars are a substitute for the efficiency benefits of density.
If anything, the expansion of remote work has just pushed the housing crisis into more housing markets.
FWIW I prefer density but isn’t this good? The reality of yesterday was too few zip codes with too much demand and not enough good will to increase supply.
Instead we now have a lot more zip codes that can make decentralized decisions about building more housing. Hopefully they vote for more development and use it as a way to stimulate jobs in their local construction industry.
How can this be a bad thing?
People just don't think in that long a timeframe but there are localities which do and hopefully it works for them and we all get to learn the timeframe for benefits (or the lack of benefits) from such a change.
https://www.theurbanist.org/2020/08/12/portland-passes-sweep...
This why why I am making the point that we shouldn't rely on remote work (of self driving cars) to solve our housing crises. The only good solution is increasing density, (and I suspect we agree on that.)
It's great if people want to take advantage of remote work to live I'm new housing markets, but they should realize the consequences of those choices and work to mitigate them.
How are traffic issues relevant to housing costs when no person is wasting time in the driver seat?
Traffic can greatly extend commuting times and thus is directly relevant.
(And that’s separate to having a self driving RV, which also radically changes the home marketplace)
I don't think "self-driving" rv's do anything to revolutionize the housing market. What they do is revolutionize travel (it's like having a private luxery train car that can go almost anywhere.) Parking an RV can be an expensive and competive in and around urban areas (speaking from vanlife experience.) Having a self-driving RV driving around all night while you sleep also isn't cost effective and won't scale.
What I see as a revolution in the rv/housing market would be affordable, all eletric RVs with sufficient solar capacity to recharge their range over the course days. Combine this with starlink and/or 5g and you create cheaply location flexible housing.
Yeah and? The limit is in the order of 250 miles. Have you seen how cheap houses get 250 miles away from cities? London to Bradford is about that distance, and the average property prices in those two are £629,400 [0] and £58,673 [1] respectively, nearly a factor of 11.
Or, if you prefer American reference points, SanFran to the edge of Nevada; or the White House to anywhere in Delaware and Maryland, most of Virginia and New Jersey, and half of each of Pennsylvania, Virginia, and West Virginia; or Boston to the Canadian border; or LA to Ensenada, Mexico.
Don’t tell me there’s nowhere affordable in all that. Especially not in the large number of tiny places that aren’t important enough to be labelled on the map when it’s zoomed out to 250 miles.
House costs are mostly the land cost. More range = more land = more supply => cheaper houses.
[0] https://www.standard.co.uk/homesandproperty/property-news/lo...
[1] https://www.moveiq.co.uk/blog/buying/10-cheapest-places-buy-...
There was an attempt in 2016 to prepare a program to retrain coal miners in center Pennsylvania and other traditional mining regions, as the general market, automation, and decrease in available coal from mining had made that a high-unemployment space. Job retraining was on the table.
People handily rejected it. They didn't want to do something else; they wanted to mine.
The children of professional drivers may find something else to do, but professional drivers are going to, on average, be most comfortable driving, and if there's no driving to be done we're on our way to system shock.
It's like a guy in a clown outfit saying the nobody wants open heart surgery beacuse everyone turns him down when he asks.
I don't believe that autonomous cars are actually feasible in environments that aren't arid/flat Arizona that doesn't have winter so it's kind of a moot point. There isn't and will not be an autonomous car that can drive on winter, snow covered roads with emergent (incorrect) paths that humans flock and follow. If the car does the right thing it'll kill people. But 1000/day? I think they can probably come in under that.
You do need a number though. People don't like to think about it but you can't value a life at infinity.
You might think anything less than the current incident rate. But that’s looking at it from a group perspective. From an individual perspective, I’ve never been in an accident in ~25 years of driving, so my requirements might be higher than others to feel like autonomous control is safer.
There are other vehicles besides autos (aircraft, boats, ships, tanks). Need to be more specific.
Maybe ejection seats requirement should be an added clause.
Why doesn't Bond get cool cars anymore?
I suspect women have a greater relative comfort with automated vehicles versus those controlled by a stranger than men, though I’m not aware of any solid data for that (certainly haven't seen any for your claim, either.)
Exactly.
Legacy regulations governing the occupant compartment are not yet a practical impediment for self driving cars because they haven't got that far yet.
Realistically what the removal of this regulation means is that some hypothetical future company with an autonomous yard spotter can drive the thing 500yd down the public way from warehouse B to warehouse C without the legal department getting antsy.
They're probably not possible with the road system we have but we redesigned our road system to accommodate cars(arguably with terrible results for non-cars) and we could probably redesign it again for autonomous cars. For example by requiring pedestrians and cyclists to wear radio transponders.
Which parts of driving would require AGI?
Being able to read the intent of other cars and preemptively act if necessary. For instance, being able to predict that the harried mom on her phone toting 3 screaming kids in her beat up minivan is likely to change lanes without looking (I passed her preemptively), or that the car with out-of-state plates is likely to miss a confusingly signed exit and try and cut over at the last minute without looking (I slowed down preemptively, to let the car abruptly veer across three lanes of traffic ahead of me).
Both of these examples require an understanding of the world that goes far beyond mere driving skills.
An average 20 min ride increased by 10% would be 22 mins. I’d gladly trade an extra few mins in the car to 1. Be safer. 2. Get back 20 mins to check the news or play chess or worst case work.
Self driving cars aren’t just going to be a reality, in PHX they kinda are a reality. It doesn’t seem far fetched to think 5 years (high estimate - 10 years) from now Waymo will be scaled to the top 25 cities in the US.
I have the opinion we don't need AI to be perfect, just on par or better than humans from an aggregate perspective.
Only if there is a mathematical proof that eventually it will happen, someone might proof the reverse that current type of tech will never be good enough for general purpose driving and will work only in some specific conditions.
Absent souls, human drivers are existence proof that minds made of matter can drive cars.
That doesn’t stop being true even if the current preferred architecture is all wrong, as we can and do change those with ease.
So I was thinking "eventually" as somewhat in next 100 years, otherwise even if you need a soul you could say shit like "eventually we will evolve to the next plane and become gods and we could create brains".
Could it be impossible that we hit a tech limit for AI? I mean there is no sure what that we will invent something better then digital, silicon based computers. I am not that pessimist though, but is possible we get smart and realize that we could just put tracks and have self driving little trains on them to travel around: 100% safe (we could have actual scientist/engineers code the software for them) and much more green.
IMO we will probably hit the single-atom limit in consumer devices this decade, though researchers are already suggesting multiple q-bits in a single atom so perhaps we won’t even stop when we get 1Å process chips.
The main limiting factor right now is that we don’t really understand the structure of our intelligence. If you will permit an analogy: perceptrons and convolution kernels are like bricks, while current brain scans (MRI, EEG, etc.) are like nighttime satellite photos of city lights — we’ve got a gap in the middle that would tell us we need steel reinforced concrete to make skyscrapers and why want to build them in the first place.
This is a rather intuitive mathematical proof that neural nets can emulate all functions in many dimensions. Knowing integrals (Calc 102) would help.
With this information it’s either that self-driving is eventually possible or human brains are not mathematically representable (i.e. something akin to a soul).
Meanwhile, the data collection/storage is getting better (all Teslas are continually testing FSD and reporting failure cases, synthetic generation of “good and unique” perfectly labeled data[0] is only a year or two “old”). Sensors/cameras are improving (and have a lot more room to improve). The compute hardware is getting better and cheaper (analog computers[1], TSMC 2nm and beyond, Apple/Google/Teslas neural cores, etc).
Additionally, given Waymo is already in production without humans in PHX it’s arguable that we already have “autonomous vehicles”. We now just need them to improve slightly and for Waymo or similar to test in the top 25 cities in the US. Which is mostly a matter of capital and time not technology limits. Maybe weather will be an issue for a few more years/decades - but even if Waymo only operated in Summer, I’d be happy to reduce my Lyft costs - and with enough data and maybe better sensors the neural nets will be able to handle weather too. Especially if the car just goes slow.
[0] https://youtu.be/j0z4FweCy4M
[1] https://youtu.be/GVsUOuSjvcg
[2] https://www.cnbc.com/2022/01/08/heres-what-it-was-like-to-ri...
Yes, but the problem is we don't know what function we want to approximate. So it is like interpolation you give it some points and you get an approximation that is good enough around those points but probably very wrong if you are far away of the points.
We don't even know how many layers and neurons are needed so it seems to be a brute force approach, throw more data and more hardware and when stuff goes wrong maybe throw more data. But one day a Tesla might go on a different environment and maybe people have different hats and goes to shit because those hats were not in the training data, we don't have like in other sciences bounded errors.
I think of this as an unproven assumption, not a given.
Well, I mean, duh. You’d be a fool to bet against seeing a unicorn on an “eventually” timeline. Try to collect and I say it’s not eventually yet.
Unicorns can't, won't ever exist.
The idea of the existence of self-driving cars is definitely on the cards. This is en route.
[1] https://abcnews.go.com/Technology/faa-drone-regulations-blow...
As far as the safety, these things can be indemnified and that will pave the way. Don't see any problem insurers can price accidents and injuries from autonomous cars. Also certainly plausible they will be able to be made safer than human driven.