Hacker News Reader

Comment by hackbinary | Hacker News Reader

I look forward to the day when most cars are autonomously driven, and it is news that a human was driven and the crash was caused by the human.

Human error accounts for something like 95% of all crashes.

https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/...

http://cyberlaw.stanford.edu/blog/2013/12/human-error-cause-...

View on HN

I wonder if that is actually going to happen. Since going to school I'm aware that autonomous systems that "take over responsibility" are an unsolved ethical problem. Delegating this away to the manufacturer or maybe even the 3rd party who delivered the code doesn't make it better. Adding many layers to responsibility usually s*cks. At least for me it's also why I don't like corporates, it's not only that you have less control over your work - you don't even know who pushes the buttons. As it turns out, in many cases it's nobody concrete. It's some kind of weird group dynamic that is oftentimes in control. In case autonomous cars become common, analysing this kind of complexity will be an awesome kind of busywork.

Kind of a silly question but in the case of a "normal" car accident, don't they just come down to compensation costs. Either for repairs and damages, the individual involved or even medical costs. Couldn't the likes of Uber simply underwrite their own insurance and fight their own cases in court where appropriate?. If self-driven Ubers are as safe as we expect all self-driven cars then wouldn't this be a "cheap" option anyway?. I don't know where you would start with more serious accidents, if Uber is at fault.

If everybody sticked more or less to the rules and if there are no severe injuries/very costly damages, then it's really just compensation costs.

I don't know the exact situation in the US (also I'm no lawyer or something), but in Germany when some kind of Negligence is involved people can lose their drivers licence temporarily or forever, may have to pay fines (on top of compensations) or even have to go to jail. (Edit: also a costly Medical-psychological assessment may be needed afterwards to be able to drive again in the future.)

Coming back to the costs, it's also interesting to say that insurances usually cover sums 1 - 20 millions. That's how much damage a single person can easily do with a car.

It's said that it's a plus that self-driving cars are somehow interconnected and learning from each other. You may argue they are thus one entity and may have to be insured like that. Ok, but that's maybe a bit too esoteric now... ;)

I think that a car accident is the wrong example to reason about autonomous cars and their accidents. A better example would be a toaster that starts a fire, in that case the important questions are, was the manufacturer negligent and was the owner negligent. If neither, then the insurance company will just regulate the damages and nobody goes to jail.

That is also a model that works well for autonomous cars, probably with some kind of licensing of the car and a mandatory insurance. The thing is, we do not need to deter a autonomous vehicle from speeding or driving under the influence, we can directly program it to respect speed limits and not drink.

>was the manufacturer negligent and was the owner negligent

Yes, but in the case of the toaster, it's pretty much the case of some sort of human error on the part of the owner (the cord was damaged, they jammed something in and it got stuck) or it's product liability on the part of the manufacturer.

It's an interesting aspect to these sorts of autonomous systems that you can expect to have accidents, even fatal ones, at some rate and that may have to be considered acceptable because "stuff happens" sometimes. There aren't a lot of examples where this is the norm in consumer-facing products. Pharmaceuticals probably come closest.

The main thing is the "risk", which in insurance speak means something like probability times costs of bad things happening. For a toaster this is less than 1 cent per year. For a human-driven car that's 1000 $ a year.

To be more concrete: the probability that a car owner gets involved in an accident at some point is actually really high. I've already been involved in 2 car accidents, not driving myself though, one of them was very serious. On the other hand I've never heard of someone having a toaster catching fire. And even if it would catch fire, you are probably watching the toaster and can put it out easily - maybe even by throwing the toaster out of the window.

This is only news because autonomously driven cars are still so rare.

It is mostly we hold automated machines to a higher standard. If 5% of all flights would not arrive would you fly?

Depends how depressed I was

Your comparison doesn't really make sense.

If human pilots had a 15% error rate, and autonomous planes had an error rate of 5% - which would you fly? It's basic math - you pick the one that is most likely to get you there safely.

Most people have no issues getting into cars - despite the fact that they're one of the largest killers. So we've already established that for most people, the convenience (or their love affair) of cars is enough to overcome the current risk of death/injury - if we could decrease that, is that not surely a net win?

> If human pilots had a 15% error rate, and autonomous planes had an error rate of 5% - which would you fly?

Neither.

If human pilots have an error rate with a large variance, and you can choose your pilot (or fly your own plane) the decision is less clear cut. Especially in a world where human accident statistics include a significant proportion of people exhibiting reckless behaviours that many jurisdictions will take their licenses away or even jail them for.

I'd want autonomous cars to be a lot safer than the average human driver before they were allowed on regular roads without a human behind the wheel. In fact, given that humans and machines are likely to make systematically different errors such that the human can solve most of the driving issues that the machine has and the machine can solve most of the driving issues a human has, the case for getting rid of the legally-responsible human driver altogether (humans ought to have the right to be drunk or sending emails whilst in their own vehicle; taxis ought to be cheaper?) isn't particularly strong.

I would choose neither.

You're making a subjective judgement that humans are fundamentally less safe than autonomous drivers.

Like most HN commenters you are making this assertion without any facts, as there isn't sufficient data to make the assertion. Other than drunk driving diversion, the case isn't as clear cut as you may think.

If you want to reduce fatalities, reducing SUVs and drunk driving is the end goal, as rollovers and alcohol related crashes are the most preventable fatality sources.

Are you implying that 5% of all car trips result in a fatal crash? That's preposterous.

Every system has a quality of performance. In production (let's say steel) we can produce with a degree of quality. , for statistical models we have a degree of confidence, etc. Every machine we currently build will fail to a certain calculable degree. The less you maintain the machine the higher the degree of failure will be.

You are assuming that after this crash their rate is still lower than humans, however it might not be.

Their rate is only lower than humans because humans take over whenever the software faces a problem they can't deal with, indicates a problem or behaves erratically.

In Uber's case, that happens a little more than once per mile driven http://www.recode.net/2017/3/16/14938116/uber-travis-kalanic...

Yeah I'd agree, however it was higher than humans then not only is it not currently viable it is unsafe to test on public roads.

I also look forward to that. Self-driving cars can save millions.

On the other hand, car hacking is a credible threat. We already have a Michael Hastings incident.

Reply on news.ycombinator.com