In the news

It again raises the ethical questions about AVs. If they cut the number of deaths by say 90% but people are still dying on the roads is that good enough? Society 'accepts' a pretty high price in death toll for the benefits of vehicle transport controlled by humans - where do we set that limit for robots?

The driver said it was like a flash, the person walked out in front of them

Like a flash, the middle-aged homeless lady pushed a bicycle laden with shopping bags into the middle of the carriageway. Literally, a split-second, like magic...

I don't think the ethical questions are that difficult really.

This 90% isn't an end game position. This is a trolley problem in a scenario where the track continues and an endless sequence of trolley problems with multiple switching positions that, if managed well, will get us to as close to zero road deaths as we can ever possibly get.

The question isn't "is it good enough?" when the alternative that we currently have is to leave those 90% to still die. In the balance of "good enough" questions, "is it good enough to sacrifice 9 people because we couldn't save the 10th?" should be the absolute loser every single time. I can't imagine why you're arguing the case against that.

Where do we set the limit for robots? It's hard to answer that question without first establishing the parameters of regulatory control, liabilities and responsibility, standardised safety protocols, information sharing protocols and so on. But if 90% doesn't fit into that limit against our current piss poor performance, we have definitely fucked up. Honestly, I might balk at 5% or 10% improvement. Largely on the grounds that much of it may be very difficult to accurately attribute. At around 15%, I personally start feeling comfortable pushing the yes button.

In my mind, the question should be; is this good enough for now? That's because, as stated, this isn't an end position. Achieving that 90% more doesn't mean you stop, it means you can have more resources to work on eliminating that other 10%. You can't actually work on that remaining 10% if you haven't dealt with the that starting 90% in the first place.

I can't imagine why you're arguing the case against that.

I think it’s the belief that tech interventions should be perfect on launch

That said, getting accidently killed by an autonomous robot in public sucks balls in a way that the same outcome by human error probably doesn’t.

There isn't a scenario for which self-driving cars are the answer. All it will result in is there being even more cars on the road, because now the one barrier to being able to drive one will be removed. It's an agenda being driven by those who stand to profit from them.

Also, full AV is the end state. This tech has lots of trickle down effects. Look at the existing collision mitigation systems in use on current cars. Lane keeping systems for motorway driving which reduce fatigue and presumably crashes. All saving lives right now, every day. Let’s not shut this shit down but have strict controls on how they experiment on public roads.

“it’s very clear it would have been difficult to avoid this collision in any kind of mode (autonomous or human-driven) based on how she came from the shadows right into the roadway,”

1. Don't rely on stereoscopic cameras for object detection.
   
2. Don't drive at the limit past low-visibility interfaces.
   
   

Their algorithm is overconfident, probably embodies the spirit of its creators.

I'm not arguing either way, but as Howard said people will see robots causing deaths on the road differently to human-caused road deaths. A 15-20% improvement might well save 1000s of lives but I don't imagine that would be acceptable to the Daily Mail etc etc as it's still 'Robot Car-nage on our streets'. I was just wondering where the line is / will be.

Don't they tend to use Lidar (although if that was the case I'm not sure why coming from the shadows would have made a difference, unless that was just a layman's interpretation of the situation).

The alternative scenario is that when you have cars that can park themselves and come and find you we'll end up with something more like a glorified taxi fleet.

I'd say the big barrier to more cars is cost, not being unable to drive (particularly in big cities).

"the answer"

No.

But there is a scenario where AVs are part of the answer. I realise humans are shit, but we are still capable of implementing multiple measures, at the same time, in conjunction with each other.

AVs should be there to mitigate the risk of the driving journeys we can't eliminate.

I can't see how they can do this. I don't buy the line that they will reduce the risk.
Computers can never be as good as a human at recognising and reacting to risk in our almost infinitely variable streets. Robots only work in strictly controlled environments and they are only as good as their programming and processing and can only work to the scenarios they have. They are shit at pattern recognition, language ability, abstraction, creative thought, 3d processing, anticipation, imagination and so many of the other thought processes that make up the main part of driving safely.

Robots are brilliant at making the car work, but that's only 10% of driving. They can't be brilliant at reading the road and conditions, anticipating what might happen, recognising developing hazards and all the rest of it. If driverless cars are to become a safe everyday reality it will have to be on strictly controlled roads away from pedestrians, cyclists, etc. That basically means banning cycling on the road, introducing jaywalking laws, etc.

I'd say the big barrier to more cars is cost, not being unable to drive (particularly in big cities).

Think of how much an average car is used in relation to it's value ie £/hour.

Now think about the changes in how we own and use things nowadays.

Driverless cars would go hand in hand with an Spotify-Uber-style system. If cars were utilized to even 50% of their ability then the cost would plummet to an almost meaningless level.

I can't wait for our robocar overlords. I also don't think it'll mean more cars on the road as less people will own them if they become a decent, autonomous taxi like service with a bunch of interlinked cars picking people up where needed and not parking on every fucking street when not in use, I'm sure some people will want their own still but that too would get less as owning your own car becomes less normal. The more robocars the better too if they're all sharing what they can see, if this car missed magic bag lady jumping out from the shadows, it may not have been a problem if the 18 cars in front of it had clocked her and was broadcasting that to everyone else in real time. Big scary complicated bit of road? Stick up some permanent sensors adding even more coverage to all local cars, would also mean less traffic, even with more cars, as they all work together at busy periods, detouring where needed and not forcing their stupid selves out into pointless places that just block up roads for others. Robot cars killing people is obviously bad, but it's already safer than most dickheads and people have been getting killed by all kinds of slightly more autonomous than it used to be machinery for hundreds of years, at least now it tries to avoid doing so.

recognising and reacting to risk in our almost infinitely variable streets

Are most accidents caused because people exceeded their capacity to do this? Or because they were well within their capacity to recognise risk but were distracted/impatient/tired/not looking/dazzled by the low sun that no-one could possibly have forseen?

There are lots of situations robots might be rubbish at navigating, but the default failure mode should be to slow down, give way, and if necessary stop and let the human figure it out. They don't suffer from fatigue, impatience or distraction, and almost literally have eyes looking in every direction simultaneously.

Robots are brilliant at the bits of driving humans are bad at, and frankly lots of humans aren't very good at the bits where they should notionally have the advantage.

When I was talking about more cars I meant in absolute numbers. It may be that the number of journeys increases but I think the number of cars doing those journeys will drop.

Computers can never be as good as a human at recognising and reacting to risk in our almost infinitely variable streets....They can't be brilliant at reading the road and conditions, anticipating what might happen, recognising developing hazards and all the rest of it.

Says you. Seriously, why not? Every time this sort of issue comes up with regard to AI, it turns out that there isn't some mystical human-only power that computers are metaphysically incapable of reproducing, rather it's just that we haven't yet properly framed the problem and described the capability that is required to respond to it.

The idea that humans are absolutely optimised for driving is belied by the number of accidents and injuries caused by drivers despite our having largely segregated motor vehicels from squishy pedestrians. Part of robots advantage is that they don't have to be "creative" about driving because they adhere to road laws, which massively reduces the degree to which they have to respond to hazardous situations and the speed with which they have to do it.

http://www.bbc.co.uk/news/entertainment-arts-43469566

This is remarkable

The robots might not win a rally, but they'll definitely kill and injur less people doing normal journeys, and they'll probably waste less energy and shout "fucking cyclists" less whilst doing it.

Part of robots advantage is that they don't have to be "creative" about driving because they adhere to road laws, which massively reduces the degree to which they have to respond to hazardous situations and the speed with which they have to do it.

Nah nobody would use them unless they drove at the limit, all the time, like a target, as in the tragic case above (38mph in a 35mph zone).

Yeah OK legislation

I suppose the human error bit will never be avoided - it will be human error in the programming of these things.

Just out of interest how does/would an AV recognise one of these for example ?

What does it need to recognise? Exactly what it is or just something in the way heading in a certain direction? Another bonus of these is that a lot of the "programming" is done by machine learning, rather than having to have some human try to think of everything and program it, and robots can share knowledge, so if the first few ain't quite sure what the weird giant bike is, they can default to super careful, but as more and more encounter them, and share what they've learnt about them, they can all learn and get better together.

they can all learn and get better together.

There is a thread just for this topic: so far it's claimed 3 lives.
https://www.lfgss.com/conversations/234810/?offset=775#comment14147567

It's only trying to help.