Hacker News Reader

Comment by smallgovt | Hacker News Reader

Not to dismiss the tragedy of this incident, but it should be expected that self-driving cars kill pedestrians -- just at a rate lower than what's expected from human drivers.

Perhaps there's a better metric to look at, but I'd like to see number of deaths caused per miles driven.

If Uber's self-driving cars are killing more pedestrians than a human driver would, we have a huge problem, but I'd be willing to bet they're at least an order of magnitude safer in this respect.

View on HN

You would lose that bet. As another comment pointed out[1]:

> The NHTSA reports a fatality rate of 1.25 deaths per 100 million miles[2], twenty five times the [4 million miles] Uber has driven.

So they should have driven around 75 million more miles before getting their first fatality, in order to remain even with humans. Not to mention they've been driving on the clearest/sunniest roads in only a few cities.

Of course a sample size of one is not enough data, but I'd say we should err on the side of "we have a huge problem".

There should have been so many safety precautions in place that nobody should have died from this yet.

[1] https://news.ycombinator.com/item?id=16620736

It'll likely never get passed, but we truly need an "acceptable death" metric to be passed to protect the companies (and apparently people too, since some in these comments what the developers held responsible) from deaths. In my opinion, a law should be passed that per mile driven, 5-10% of the current deaths per mile is allowed, and it should begin decreasing after 10 or so years to something low (maybe 1%, maybe .01%, we'd have to wait and see).

People will always make mistakes, the benefit with self driving cars however is that mistake should only be made once. Whatever bug caused it to be considered at fault can be patched and will never occur in that exact circumstance again. Meanwhile with humans you could have that same case occur over and over again.

With how much money people can win in lawsuits nowadays, all it would take is a handful of cases to totally destroy a manufacturer of self driving cars.

That sounds like the market correcting itself, while you want to protect the likes of Uber from market forces, with new laws. Putting that aside though, good luck trying to get re-elected after passing a law like that. Few people are technocratic and emotionally detached enough to agree with you, from the looks of it, not even on this site.

How on earth is a lawsuit a market force?

Did you even read the first 5 words?

>It'll likely never get passed

Can you stop trolling my posts? You aren't adding much of anything.

In general, I like having a diverse set of options available in terms of software implementations (eg: multiple C++ compilers to choose from). It does worry me a little bit, however, that so many companies are trying to implement their own self-driving cars from scratch. If you just look at how buggy new OS releases tend to be, how many blatantly obvious bugs you run into every month, it's clear that many software development teams don't properly test their stuff. Some companies, like uber here, will act irresponsibly and fuck up, resulting in the loss of human lives.

I would guess it's inevitable that in the medium/long-term, there will only be one or two companies developing and selling self-driving tech to every car manufacturer. Small players won't have enough data, everyone will be too afraid of the risk, the amount of government regulations surrounding self-driving cars will increase. More regulations will protect people but it will also mean it's much harder to start gathering the data you need and testing your product.

There is no data available at the moment to draw conclusions that self-driving cars bring less fatalities, and rightly so, because they have not been tested on a beta stage. You are only guessing and hoping.

There is certainly at least SOME data. At the very least, let's find out how many miles Uber's self-driving program has driven. Then, we can, at least, compare deaths per miles driven to a human driver.

There is, and the data doesn't look good.

There's typically 1 death per 100 million miles driven.

Uber's only driven a couple million miles.

Human drivers kill humans at a rate of 1 per 100 million miles. Collisions occur at a rate of like 1 per 200k miles.

AI drivers have killed 1 human having driven a mere ~5 million miles (not even counting Tesla autopilot). Disengagements occur at a rate of 1 per 5k miles.

The evidence strongly, strongly points to self-driving cars being more dangerous than humans today.

Fewer people will die (probably). The problem is that different people will die, and you can't celebrate with the families of the people who were slated to die and didn't.

And different people will "kill". It won't be the drunk drivers, or the road ragers, or the people texting on phones, it will be anyone unfortunate enough to be in the car at the time. A businessman. A politician. A cop. A teacher.

And to say that "it's not their fault" and "it could have happened to anyone" will be cold comfort to those who just watched their car mow down an innocent, and there was nothing they could do to stop it.

That’s an excellent point, one I hadn’t considered. Of course any number of accidents happen today to people whose only mistake is not being prescient, or at least not driving sufficiently paranoid to avoid something that arises unexpectedly.

Certainly this occurs today. "Tomorrow" it simply becomes the norm.

Whilst I imagine it is horrible to be in a car that drives someone over, I don't think the driver deserves the brunt of the sympathy in that situation.

Not the lion's share, but it could happen to anyone, and I have to imagine the psychological impact never truly goes away.

Recognizing that being a witness to a violent death, especially if its your vehicle (despite not being in control), has a significant impact on people takes nothing away from the victim's family.

Sympathy is not a zero sum game.

It's not that simple. Who is going to take responsibility for such incidents? The company that owns the car? The person behind the wheel (if any)? Or the engineer that designed the algorithm?

I would think it will be pretty standard product liability. With some assumptions (proper maintenance, software updates, etc., used in accordance with directions), it's hard to see liability for deaths and injuries not resting with the manufacturer. What's unusual for a consumer product is that, with the somewhat exception of pharmaceuticals, there aren't a lot of examples of consumer products that routinely kill people even when properly used.

Cigarettes?

Well, yes, cigarettes and alcohol (the latter in excess). Or even sugary drinks in excess. But there was, of course, a huge settlement in the case of cigarettes and the dangers are fairly well understood. There also would seem to be a qualitative difference between links to health problems/mortality over the course of years and failures that lead to someone's immediate death.

I'm pretty sure you don't want to compare right now. Self-driving cars are just too new.

I think the rate of overall vehicle crash deaths in the U.S. is less than 2 per 100 Million vehicle miles. Pedestrian traffic is a fraction of that.

Reply on news.ycombinator.com