Hacker News Reader

Comment by kctess5 | Hacker News Reader

I keep seeing the "130 million miles of Autopilot control per fatality" statistic, especially in the context of "look these cars are better than humans" but I think this is incredibly disingenuous. First of all, you can't just take a sample size of 1 for an effectively random event and then do (total miles driven) / 1 = 1300000000 miles/fatality and pretend like that has any statistical significance.

Furthermore, during those entire 130 million miles, there was theoretically a human also sitting behind the wheel ready to take over if the system failed. If there was no human behind the wheel, the cars would not have been able to attain 130 million miles of drive time without a very large number of crashes. When humans drive, there is no intelligent being ready to take over whenever we make a mistake as is the case for Autopilot.

In my opinion, using this "statistic" is a very lazy way of hiding behind the numbers in dealing with a very non-obvious issue.

View on HN

Not to mention, those 100 million miles include non-divided roads, motorcycle deaths, adverse weather conditions, and cheaper, older cars.

https://news.ycombinator.com/item?id=12087603

There's an even greater problem. Aren't they are comparing humans driving on ALL roads to autopilot driving on highways? Most accidents occur at intersections, not on the turnpike.

> Furthermore, during those entire 130 million miles, there was theoretically a human also sitting behind the wheel ready to take over if the system failed.

Yes. That is the point, though. When Autopilot is used the way it is supposed to be used, it doesn't seem to decrease safety, and it may actually increase it.

I agree with the point in your first paragraph, though.

I would assert that you cannot make that claim, as there is no statistically significant evidence in support of it.

I wasn't making any kind of strong claim. I said "doesn't seem" and "may". My point is, whatever evidence we've got so far, it doesn't seem to indicate a decrease in safety. If anything, it seems to point (however feebly) towards a safety increase. IOW, the data so far doesn't support, in my view, the stance that Autopilot in its current form is unsafe, as long as it's used in accordance to the instructions.

The issue is that 1 data point does not point to either side. You can neither make a strong nor a weak claim for either side, because the sample size is too insignificant.

If I flip a coin twice with one head and one tail as the result, we cannot say whether or not the coin is fair. Even with both heads or both tails as the result, we still do not have enough information. It's best to just say that we need more data.

It's 1 data point if you look at crashes, there are 130 million data points if you look at miles driven. It's by far the largest amount of data available on Autopilot driving out of all manufacturers.

I agree we have no statistically rigorous argument yet that Autopilot increases safety. I agree with you and others who say we need more data. My only point is that there is no reason, IMO, to stop gathering that data in real-world conditions by banning Autopilot, as some media articles are suggesting (not anyone in this thread), because we have no evidence so far that it decreases safety.

As a person who in robotics, believe me when I say that I am interested in these technologies being developed and tested. That said, I am aware that a major PR battle is the last thing that the industry needs right now.

I think that Tesla is doing a disservice to the rest of the industry by being overly zealous with their marketing and naming department. "Autopilot" is a bit of an exaggeration that needlessly oversells what is effectively a lane-assist feature. By saying that it's "autopilot" they set an expectation of a high degree of autonomy, and thereby influence how their technology is perceived and evaluated both by users and the press. Fatal accidents caused by incorrect use of a "Lane Assist" is a very different headline than an accident caused by "Autopilot."

Tesla is not the only manufacturer with some kind of active cruise control, they just get a lot of attention because they like to pretend they are selling autonomous cars. I am just crossing my fingers that their recklessness will not cause a significant setback to autonomous car technology in general. Tesla hand waving away safety concerns by quoting dodgy apples-to-oranges "statistics" doesn't exactly make me feel warm and fuzzy.

Agreed.

Thank you, I'm glad to see a fellow robotics engineer speaking out against this. Oftentimes I see this debate cast as "pro-driverless-car" vs "anti-driverless-car", but the debate is more nuanced than that. It's up to people who understand the technology to point out what is actually wrong here. We're not trying to hold back progress; we're trying to promote progress in a safe and responsible way. No one needed to die here.

edit:

> Tesla hand waving away safety concerns by quoting dodgy apples-to-oranges "statistics" doesn't exactly make me feel warm and fuzzy.

Same here. I've always liked Audi's philosophy and their slogan really spoke to me: Vorsprung durch Technik - "Advancement through technology". After their emissions scandal, it seems that their strengths do not extend to ethics. I was hoping Tesla would replace Audi in my eye, but they are turning out to be just as morally bankrupt, IMO.

Reply on news.ycombinator.com