Hacker News Reader

Comment by camjohnson26 | Hacker News Reader

This idea gets brought up every time autopilot is discussed here but it’s not really relevant. If autopilot were safer than human drivers there would be an argument for allowing it, but there’s no evidence that it is. Sure Tesla claims that it is based on very specific, cherry picked internal metrics, but independent analysis shows there’s just not enough data to say. Also the class of errors that autopilot makes is completely different than a human. Most human car crashes are preventable, either driving under the influence or distracted driving. I would much rather be in a crash with a drunk driver where liability was clear than have my car drive underneath a tractor trailer in the middle of a clear day just because I hit some neural network edge case.

View on HN

> I would much rather be in a crash with a drunk driver where liability was clear than have my car drive underneath a tractor trailer in the middle of a clear day just because I hit some neural network edge case.

Preferring to die by human error instead of shitty software doesn't make logical sense to me and feels like your trying to justify something irrational. Autopilot either drives safer than me or doesn't. And I wouldn't let a drunk friend drive me either.

> but independent analysis shows there’s just not enough data to say

Do you link a trustworthy analysis? I can't remember reading one that wasn't from a highly polarized source.

This makes a lot of sense, and is actually seen a lot in the wild. Having control feels good for us, and I would argue that's probably why there is such a visceral response to drunk drivers---they're not in control. You can empathize with a mistake, but something just sort of happening with no real impact or control over it feels wrong.

That being said, I'm on board with the self driving cars, we should approach it with a logical head and really try to figure out what we should do, not what feels good.

>Preferring to die by human error instead of shitty software doesn't make logical sense to me and feels like your trying to justify something irrational.

The poster, camjohnson26, did not say die, the poster said crash.

Camjohnson26's point as I understood it was to compare two situations:

1. A crash that occurs because one of the drivers was extreme drunk,

2. A crash that occurs because of an autopilot in one of vehicles did not make the current action.

In the first case most people and the legal system will likely conclude that the drunk driver was at fault. In the second case it will be more difficult to determine who was at fault. This second case may add legal, insurance and ethical complications that camjohnson26 wishes to avoid. For instance what if your insurance decides you are at fault for activating the autopilot under unsafe conditions?

> what if your insurance decides you are at fault for activating the autopilot under unsafe conditions?

Seems reasonable to me. The driver is always responsible.

Which driver? How do you determine it was actually the autopilot that caused the crash? What if it wasn't the autopilot?

>Preferring to die by human error instead of shitty software doesn't make logical sense to me

It certainly does if both auto pilot Tesla’s and standard vehicles are resulting in death, but the law will provide a remedy to your survivors and heirs in one instance and actually turn around and hold you liable in another instance.

After all the post was about liability and clearly nothing to do with how the poster prefers to die, which you somehow read into it.

Average rates appear to be 12.5 per billion miles, and it appears Tesla Autopilot is at 3 for over a billion miles with the system engaged. That's a quarter of the usual fatality rate, for anyone counting. Saying "there’s no evidence that it is" isn't true; the above statements (and supporting facts) are evidence.

You aren’t controlling for other factors. Luxury cars and new cars have lower rates.

”That's a quarter of the usual fatality rate”

I beg to differ. https://en.wikipedia.org/wiki/Transportation_safety_in_the_U...:

”Comparing motorways (controlled-access, divided highways) […] the United States had 3.38 road fatalities per 1 billion vehicle-km on its Interstate-type highways, often called freeways”

I think that number better reflects where Tesla’s autopilot is used. Also, if you remove motorcycle deaths from that number, I can’t see how the Tesla could be significantly safer, let alone four times less deadly.

I think you should also correct for vehicle age (older cars and lighter cars are less safe). You could (and, IMO, should) also correct for vehicle size/price, as Tesla drivers won’t be interested in whether their car is safer than other cars on the road, but in whether it is safer than what they could have bought instead.

And of course, there’s the elephant in the room. I expect that Tesla drivers prevented at least half the deadly accidents that would have occurred if Autopilot did all the highway driving.

People keep saying (in defense of it) that the Tesla Autopilot does not drive the car for you. As such, it makes no sense to compare its fatality rate to the fatality rate of humans who are driving cars. It seems arguable to me that all those Autopilot fatalities were excess fatalities.

Reply on news.ycombinator.com