Hacker News Reader

So now, in addition to everything, there will come a day when a person is stranded some-place as the driverless car couldnt exactly find them

Ahh, if you think this is a new problem, you're too young to remember the fun of calling a cab by pay phone to some obscure address, then alternating between going inside and calling the dispatcher and then running out to the street to try to flag down the cab.

At least with a driverless car I can pinpoint my location on a map when I call it.

Person winds up being in an unsafe environment and needs immediate assistance.

Existing car-share (and cab) drivers are also willing to drop you off in an unsafe environment if that's where you told him you want to go. While sometimes they'll stick around to make sure you make it to your door (if you're female), that's by no means guaranteed.

im 42, so no not too young - and I appreciate your point.

I feel that people are latching onto their opinions about driverless car tech and not talking about the wider implications. That is the point, I am poorly, making, apparently....

I just think that not all the questions have been asked about what it means to be a "driverless car" and that we need to run through all the risks -- and thats the one thing I find really lacking in any of these conversations at all.

did you not read the terms of service when you opened the car door? they are not responsible.

NEXT NEXT NEXT NExT OK INSTALL YES

I'm just imagining opening the door and getting a 10 second long 2000wpm recording of someone reading the TaC.

The silkscreened-onto-the-headliner bright yellow non-removable warnings weren't enough.

TBH, your objection seems like something that anyone would think of in ten minutes of considering pitfalls. To think that no one at Waymo (or Cruise, or Lyft, etc.) has considered it, you must have a very dim view of their competence.

Nope, not question their competence - I’m questioning where the answers are to the dialogue we should be having surrounding ai cars???? Where the fuck are the answers, what body owns them?

Presumably each company will have their own approaches to these issues. They don't prima facie seem to be issues that require legislation. In that sense, "we" are not having a dialogue about this at all, any more than "we" are having conversation about cyclist detection, because "we" are not building self-driving cars. Waymo, Cruise, etc. are building self-driving cars, and we are not privy to their internal deliberations on these issues.

Thank you for this point.

My mention of “we” though are the consumers who will be paying these companies for rides and presuming they have already thought about these safety issues, yet “we” have no privvy to an understanding about who has approached the issues nor who has the best approach and neither a standard to refer to that “we” can expect the companies to uphold.

That’s the whole point of “us” starting to ask questions.

“So, uh, you want me to fly in your plane, to what standards are these planes built to?”

Oh look, “we” have an FAA who manages all that, regardless of who made the damn plane.

I'm mostly curious what you imagine happens now in your scenarios with owner-driven cars and taxis. It's not like they now lead to better outcomes.

I'll just be happy that I don't have to make sure the driver isn't microsleeping on a late drive home (it's happened more often than I care to count). Some countries have very poor regulations on working hours for people operating vehicles.

Edit: I don't mean to sound like I'm singling out ride sharing/hailing companies, it's an across the board thing with human drivers.

Replace your driverless car with the driverless metro in Copenhagen, where you're only likely to see a metro employee in the busiest stations at peak times.

Danish society hasn't yet collapsed. I think you're overstating the issues.

Is that said metro attached to rails -- or does it take you along "lines" where you have to get out at a designated station and then walk 2 miles to you house -- or does it pick you up and take you directly to the market.

Your comment is bullshit comparison if youre talking about comparing freaking rails with vehicles whos routes are 100% random per rider.

> Person gets into the driverless car way too intoxicated and passes out/overdoses/dies in the car -- need physical help?

> So now we have driverless cars effectively classified as emergency ambulances - so assuming the CARNOC is paying attention, they then drive the car to the hospital - and have to have a mechanism to contact said hospital

> Imagine the healthcare implications where if someone hasa seizure, ODs, cardiac, etc... what is the response time from the CARNOC, and getting them to a hospital ==> then what types of lawsuits will these companies see? what type of insurance will they require?

The comparison is useful because all of the above scenarios are possible in the Danish rail system.

There would be several bystanders in a typical Copenhagen rail car; they would be able to call for help. This wouldn't necessarily be the case in a driverless car; if the only passenger has a medical emergency, it would be up to the vehicle to detect that and take appropriate action.

Not necessarily in the scenario they presented. While I haven't been in Copenhagen specifically, I've been in plenty of train cars late at night (think last few trains before service closes) where I'm the only person in the car.

Just like a random onlooker could see you from the platform, someone in another car could see you in your driverless one and call emergency services. The two scenarios are very similar in the end. It's just that the train car has a higher (but far from 100%) chance that you will be noticed more quickly.

The Copenhagen Metro runs all day and all night, so it's pretty easy to find an empty or almost empty train -- near the end of a line at 3am on Tuesday should work.

The scenario certainly also works for trains in general -- perhaps even more so, since an older, manually driven train is less likely to be filled with CCTV cameras linked to a control room. (The metro has this, but it seems they're only monitored once someone presses the emergency button.)

Ok - thank you for that explaination, however, the danish rail system is probably a model which any and all of the driverless car companies shall fail to emulate in that regard because they are going to fail at having a top notch NOC as they hire a bunch of people who don’t understand or value that type of service level.

But thank you for clarifying.

Mentioning rails and "100% random" turn up an interesting point - people are comfortable with autonomous vehicles when they behave in predictable, understandable and widely known ways. And it looks like the Waymo cars and other autocars in general will do just that. Once people learn to recognize an autonomous driving style, whether they'll be concerned or bored depends on how predictable the style is, not the fact that it's autonomous.

hmmm - should we require all autonomous vehicles to be "micro buses-on-deman"d until they are trusted - and follow beaconed bus lines rather than random routes for a while until trusted?

> So now we have driverless cars effectively classified as emergency ambulances

That's quite a huge leap you made there. Care to back up and explain how a drunk person calling an uber magically turns the uber into an emergency ambulance?

They manage to climb in, they OD/go into arrest, cant climb out.

What then? So the uber now has to either get a person or take that car to a hospital - how do they know to do such?

So is uber running a massive call center that watches every driverless ride and monitors the situation?

Are they monitoring if people get into a fight in the car?

That actually happened to me, where I took an uber pool, and the two other riders got into a physical fist fight in the car and the driver had to pull into a gas station and kick them out.

How to monitor a potential rape? How to monitor unless your watching literally every ride.

> So the uber now has to either get a person or take that car to a hospital - how do they know to do such?

It does? Why? What if the person passed out in their own, manually driven car? Would you expect that car to automatically call an ambulance?

Just because you theoretically could have some kind of elaborate system that detects when the user falls unconscious and carries them to the hospital, doesn't mean that's _required_.

> Are they monitoring if people get into a fight in the car?

No. Why would they be? What makes you think that's necessary?

They will have to handle the case where the passenger doesn't exit the vehicle somehow or another. But the fact that it is occurring in an automated vehicle shouldn't make anyone worse off.

Ya, this sounds like an easy problem to solve.

If the passenger doesn't exit the vehicle after a set time, start a video call to a support central (perhaps the person is unable to get out because the doors won't unlock? or they fell asleep?) The attendant could then instruct the vehicle to go to a pre-specified location nearby where a human can intervene, call another vehicle or notify the competent authorities.

There are certainly cases where going to another location will be impossible. If the passenger opens the door and does not exit, for example, there is no way to close the door or to move the vehicle.

But I agree with the overall point, while occasional incapacitated passengers or manual intervention may occur, these are not insurmountable problems.

I suggest you do the math on this though since you think its so simple:

Ride is avg $10, vehicle is ~$30,000 satellite uplink is ~1200 per year. BOFH guy watching is ~$65k per year... insurance is ~$2400/ year...

so if we take that, we have an annual cost of (assuming life of care is 3 years) 10000+65000+1200+2400 - so ~80K per year per car... and / BOFH among his volume to monitor N cars....

but we extract the BOFH from the car model.

so we have a per car cost of 15K per year

and we have an avg rider value of 10 per ride, and a ride volume of 1,500 rides per month / BOFHs required / insurance claim / lawsuits

and we can suss out the model pretty well... but there are a lot more inputs.

and if this is going to be a service to public from multi-parties, we'd better understand this model....

and they'd better be standard on the CARNOC

What do you mean why?

This is literally the stupidest response on this thread. Why? Because presume I am paying Ajedi32 to drive me home. So the presumption is that Ajedi32 is responsible for my well being to arrive Home.

So I don’t arrive home. I get dumped on the side of the road or I die in the presumed care of Ajedi32.

Don’t mess around with Your reputation, Ajedi32 because you want to defend stupid ai driving cars, have some self awareness and vision.

Yes, you're paying the car to drive you home. You're not paying the car to monitor your vital signs, or to break up fights between you and another passenger. Those have nothing to do with the car's primary function.

If you were paying me to drive you home you might reasonably expect that I would also do those things, since I'm a human and am therefore capable of performing tasks completely unrelated to the one you're paying me for. But the car isn't a human, and there's no reason you should expect it to behave like one in all respects. (For example, I am also capable of engaging in small talk with you. Would you expect the self driving car to be capable of performing that task, just because I can?)

Basically, if you OD and pass out in the back of a autonomous car, there's no reason to expect the outcome to be any different from if you ODd and passed out in a public restroom. And you wouldn't be justified in blaming the car for that anymore than you'd be justified in blaming the restroom.

These aren't hard questions in the long term though.

Non responsive passenger->follow protocol that emergency services prefers (or even legally requires).

Fist fight->stop car, inform passengers that car won't transport them further.

Rape->turn over recordings of vehicle interior to police.

I suppose for the last one you have to take it as a given that they are making such recordings (probably on a rolling basis), but there you go.

It's not like these situations are any thornier because an automated system is involved. Passed out person is better off with monitoring (at least in health terms), fist fight without car, same difference, rape without monitoring, no video evidence.

> Rape->turn over recordings of vehicle interior to police.

Well, Uber may have some difficulty complying, but I'm sure their competitors will be fine.

I appreciate your calm response.

So, then the idea for a unicorn is that we need a "fitbit for health monitoring while riding in a self driving car" as an app on the apple watch.

No fitbit required. The interiors will have multiple video cameras of the interior. An automated system that flags unusual behavior, then alerts a human would be fine.

For emergencies where the passenger is more responsive, an "onstar"-like emergency button that calls emergency support staff would probably suffice.

I like where this thread is going. If the car is responsible for the passenger and what happens to them inside the car, they could conceivably act as an ambulance. Maybe it could even diagnose you on the way to the hospital.

How do they know you need to be taken to a hospital? They will have internal cameras and when you fail to exit at your destination after being invited 3 times to do so, they will enable that camera and somebody in a remote location might take a look and make a call - or the car might automatically take you to the nearest hospital without even consulting someone if you fail to exit after a timeout.

How to monitor a potential rape? That thing probably has 15 cameras on the outside and might already be saving the footage for law enforcement review. And you actually removed one potential rapist from the scene by not having a driver - sadly this is a real (though thankfully infrequent) issue.

video camera monitored by a machine learning system, which escalates unknown or edge cases to a human dispatcher for a decision.

how many poor outcomes are required to train the MLV on what constitutes shitty situations?

How many lawsuits come from that?

"So now we have driverless cars effectively classified as emergency ambulances - so assuming the CARNOC is paying attention, they then drive the car to the hospital - and have to have a mechanism to contact said hospital"

They don't necessarily need to drive the person to the hospital. They could just call a real ambulance and wait in place for it.

Yeah, I suspect the liability issues arising from trying to take the person to the hospital would prove too steep. Ambulance + EMT assistance is a safer bet, even with the initial delay.

These are companies with big purses and lots of human-factors personnel, and lots of lawyers that can write huge terms-of-service agreements, and cover legal costs and damages in lawsuits until they get it right.

There are few better test beds for driverless cars than Phoenix metro, from technical, legal, and customer-concern viewpoints.