Hacker News Reader

Comment by falcolas | Hacker News Reader

Why not use it? IMO, it's because I don't think that the difference is enough. And the wrecks which are being reported are for stupid things.

Not to mention, there's only one datapoint for Autopilot fatalities, which makes any attempt to use it as a statistically significant point useless. Just one more death in the coming months this year would make the number jump significantly above the averages reported in this article. And with the number of potentially deadly autopilot wrecks reported lately, it's inevitable.

View on HN

(mostly copied from my comment on the other story)

Given the number of Autopilot miles driven, you'd only expect one death so far, so that we have one doesn't really tell us anything.

However, injuries from car crashes are (fortunately!) much more common - roughly 60 times as likely. So you'd expect to have about 60 Autopilot incidents resulting in injury by now. I don't really keep up with Tesla news but, based on the reaction to the previous story (where the guy wasn't even injured), I'm going to assume the actual number of injuries is tiny.

Tesla or Elon recently announced that the accident rate for autopilot is half of human driving. I didn't read past the headline but I'd hope such a statement was based on a comparison that controlled for the type of driving environment and didn't exclude accidents that occurred moments after autopilot gave up and threw the wheel to the driver.

You'd really have to look at the number of accidents per mile where people were driving in a single lane on the highway, not the overall number of traffic accidents.

If you have better numbers, feel free to mention them. The deaths per mile is for highway driving, and I assumed the deaths:injuries ratio is the same, but that could be wrong.

I disagree with the "single lane" though: if Autopilot enforces driving in a single lane, and that turns out to be safer, then that's all that counts.

We need "humans driving in a single lane on a highway" for a control group for the experiment - the number of miles driven by a human in equivalent situations as driven by the Tesla autopilot. It may be possible to "control for other factors" as they say, but I have not seen that done.

If you don't have a proper control group, the Tesla autopilot will be significantly biased for success because it is dealing with much simpler driving conditions than "in all conditions" human driving.

I am saying that these aren't the best numbers we have. They just aren't sufficient to make absolute conclusions from. If all accidents happen while turning and autopilot never turns then autopilot would never make driving safer.

>Given the number of Autopilot miles driven, you'd only expect one death so far, so that we have one doesn't really tell us anything.

It's really telling when it is expected that the Autopilot will kill someone.

What is the current death toll of Google's tech?

Around 32,000 people die in the USA each year in traffic accidents, so yes, I didn't expect Autopilot to magically reduce that to zero.

The Google tech that is driven by professional drivers who take over when required, hasn't killed anyone yet. Do you honestly think there will be zero deaths in a Google vehicle when it's released to the public?

I don't know if a Google vehicle will ever be released to the public.

I would not expect an Autopilot to kill me.

It depends on how you drive, computer controlled cars are subject to physics like any object. Consider if you are standing on any city street corner, where the speed is 30 miles per house. If you jump in front of a car just passing you, and it hits you, you'll be dead, and no auto-drive or google drive or any drive could protect you.

I believe a Google car will see you on the corner, project your trajectory, and act to avoid a collision.

No more telling than it being expected that our primary mode of transportation kills dozens of people every day.

Reply on news.ycombinator.com