Hacker News Reader

Comment by Noos | Hacker News Reader

I'm worried about them because they will replace human error with computer error, which although rarer, is systemic and harder to predict and deal with. Human error is bad, but I don't want to wake up one day and find a Heartbleed-level error in self-driving cars, or the equivalent of a hardware-level recall.

I'm always surprised people are so accepting of it, to be frank.

View on HN

If fewer people are killed in auto accidents, it’s a win even if the type of error is different. Dead people is dead people.

Why are you surprised? People are cool with massive metal murder machines flying down roads just because it’s a human driving it and not a computer? I seriously can’t wait for all us flesh drivers to be replaced with computers. The thing is that over time computers become better and better drivers. Humans don’t, actually we just get worse as we age and stay on the road anyway. Our human reflexes and biology also limit how fast we can react and subject us to road rage, stress, fatigue and other problems that can’t be debugged. Why is it worse to have software that we might actually be able to fix?

Where does blame get placed when there is a computational error and my self-driving car kills another human-being? Am I still liable under my insurance policy even though the error was out of my control? Is there any conversation going on about insuring self-driving vehicles and responsibility when an error does occur and someone is killed?

There are a lot of conversations happening about this. In short, right now the autonomous features available to consumers are considered level 2 so the driver is required to be alert and ready to take control at all times (similar to using cruise control).

https://en.wikipedia.org/wiki/Autonomous_car_liability

Thanks, fitzroy. That is interesting I'll have to read more about this. It's just my opinion but at this time I can't see self-driving vehicles as a viable option for anyone. It just seems like another layer of complexity added into an already complex situtation.

In the case I need to assume control of the vehicle am I going to be fighting against the autonomous system for control? Not only will I need to understand the current environment and what others are doing around me; I also need to understand how the system is computing the situation and maintain control based on my observations and my observations of what the system is trying to do.

GM has already announced vehicles with no steering wheel.

This will perhaps remind all of the Runaway Prius drivers something that people stopped teaching; Remove the key, It's the kill switch.

It has been talked about from the start. I remember a documentary from the late 90's about self driving cars (it may have been BBC Horizon) and the liability issue was discussed.

Maybe the answer will be Uber-like car fleet services that self-insure.

> Where does blame get placed when there is a computational error and my self-driving car kills another human-being?

Why do you assume the courts are incapable of handling this, or the law doesn't deal with it? Most likely you won't even own a self driving car, you'll be getting into a car owned by Uber or Google and they'll have to face liability, the same way you aren't liable as a passenger in a taxi if the driver gets into an accident.

On the victim side, you're actually more protected because a self driving car is less likely to be uninsured. If an uninsured driver hits you, you're just out of luck, they may go to jail but nobody is paying your bills. In the near term Uber or Google will have loads of insurance and loads of money to pay out claims.

There was no assumption. Just a simple question. If we are using assumptions, I assume it won't be as easy for the courts to handle initially as you suggest. It's a newer technology therefore there won't be any/many precendents. I'm not a lawyer but it seems quite obvious that this type of thing could get complex fast.

If fewer people are killed in auto accidents, it’s a win even if the type of error is different. Dead people is dead people.

Well, I think the GP was sketching a scenario where fewer people might be killed until one day when a lot of people are killed by a bug/hack/zero-day that suddenly generates huge crashes.

> is systemic and harder to predict and deal with

Is it actually harder to deal with? About 1/6 of fatalities are drunk drivers[2], and they cause another 1/6 of of car accident fatalities[1]. Almost half of all front-seat (driver + front passenger) fatalities aren't wearing seatbelts while on average 90% of Americans do wear seatbelts[2]. There's no way of solving either these problems without identifying people individually and improving their behavior, which is a pretty intractable issue.

1. https://www.cdc.gov/motorvehiclesafety/impaired_driving/impa...

2. http://www.iihs.org/iihs/topics/t/general-statistics/fatalit...

I would argue that aggressive education campaigns (similar to anti-smoking campaigns which drastically reduced the number of smokers) could help with seatbelt awareness. Showing how far someone crashes brought a windshield even at 35 mi/h can be pretty effective at getting the point across.

I don't think we will have much choice about accepting the risks of computer errors/bugs. Transportation systems generally are probably reaching the the limits humans ability to manage them, we will have to turn them over more fully to computers at some point.

Reply on news.ycombinator.com