AI Programming to be based on public opinion

[The utility cyclist]

So I was watching on catchup Guy Martin and his 'race' against an automated vehicle. Try as they could they couldn't get the race car to match the inputs of a human so the AI failed/lost the race.
However right at the end GM was asked to do a little test to see what/who should be sacrificed in an incident with an AI vehicle whether that be abimals, pedestrians, peiple on bikes or the occupants.
By far the public are voting that those outside the vehicle ahould be sacrificed, this is already the case with Mercedes Benz technology.
GM viewpoint thankfully is in the camp of those in the killing machine should be sacrificed as those outside the vehicle are the innocent parties (for the most part).
This is rather worrying that just as we see in courts the opinion of the motor centric public could end up making the roads more dangerous to keep the operators of killing machines safe and not be punished.

[thirdcrank]

Preaching to the converted?

[gaz]

Isaac Asimov's "Three Laws of Robotics" come to my mind, tempered by a knowledge that motoring corporations have already created software programmes that cheat emmisions tests. A Robocop style "Directive 4" seems most likely, take no action that will harm the company.

I can't find the article at the moment but I've seen the idea of a driver controlled "moral compass" switch as a proposed solution, where "kill self / kill others" becomes an input before the start of the journey. I don't find that reassuring.

[thirdcrank]

I suspect the manufacturers are in a dilemma over the introduction of this technology.

In terms of commercial vehicles, this has a big potential for deskilling. Why pay HGV drivers if you can use a computer instead?

With private vehicles, things may not be quite so clear cut. eg How much of the sense of being in control are drivers prepared to surrender for the convenience of letting the car do the work?

[reohn2]

........With private vehicles, things may not be quite so clear cut. eg How much of the sense of being in control are drivers prepared to surrender for the convenience of letting the car do the work?

Nail,head,on!
Drivers like the power of control,and like to out do other road users,I see it everytime I drive,the pettiness is unbelievable at times.It makes think drivers will be reluctant to give over the wheel to something that doesnt have such an attitude and at the same time renders them impotant.

The voting for self protection from viewers doesn't surprise me as they are only looking at the issue from a driving POV,the program comes from a position of power not vulnerability GM sees both sides as he probably more than most knows the vulnerability of cycling due to his racing background,there's no rollbar on a 200mph motorcycle.

[The utility cyclist]

I think it's more to do with the fact he's a decent human being and being able to understand the concept of something that inherently does kill millions of people should be sacrificed/sacrifice the operators as opposed to those just going about their business whose actions doesn't kill or maim.
This is a bit like whether a police officer or soldier should shoot a gunman, armed robber, crazed motorist, knife wielder or enemy combatant or allow them to live and put others in danger. For me it's a head/chest shot every time and so it should be for motorvehicles.

The programming should take into account the more vulnerable and their mistakes just as we should make that so currently with motorists.
After all a cyclist was jailed on the back of not taking into account the mistakes made by a pedestrian even though he had very little chance of avoiding collision because of the several unexpected movements unfolding directly in front of him, it certainly wasn't any less action by him than most others as he tried to avoid collision and slowed down substantially.
Yet when a person on a bike or pedestrian makes a small error and pays with their life or is seriously injured motorists are never ever charged, rarely even when they've done nothing wrong.

[Cyril Haearn]

Are there parallels with things that already exist, semi-automatic trains, lathes with cutoff, circuit breakers etc?

It is earnestly to be hoped that vehicles will not be able to exceed maximum speed limits and other laws, that will be very strange for normal motons

[Annoying Twit]

It could be that the 'sacrifice people outside the car' is very rare and there will be other benefits for cyclists that considerably outweigh that.

E.g. if cyclists move out and take up the primary position, will self-driving cars and trucks always wait patiently behind until it's actually safe to overtake, then give good room? If so, then cyclists may quickly learn how to control the cars by our own actions and therefore increase safety as an overall balance.

Disclaimer: I have a doctorate in AI.

[thirdcrank]

The likelihood that the situation will arise where a vulnerable road user can bring everything to a grinding halt seems remote to me. I'd plump for more things like pedestrian railings and onboard cameras to record events. This is little to do with what's technically possible and a lot to do with expediency. As one small but significant example, we've already had mention on another thread about one of the manufacturers lobbying (in the US?) for these vehicles to be allowed to match the speed of general traffic rather than being compelled to obey the law in the form of speed limits.

PS

And the greater the suggestion that it would be a jolly good wheeze to stop traffic like this, the more ammunition for the manufacturers to protest that they have to be allowed to ignore it.

[kwackers]

It could be that the 'sacrifice people outside the car' is very rare and there will be other benefits for cyclists that considerably outweigh that.

E.g. if cyclists move out and take up the primary position, will self-driving cars and trucks always wait patiently behind until it's actually safe to overtake, then give good room? If so, then cyclists may quickly learn how to control the cars by our own actions and therefore increase safety as an overall balance.

Disclaimer: I have a doctorate in AI.

Exactly.

What makes me laugh is how this suddenly becomes an important issue. For decades we've accepted that self survival is the driving force for the monkey behind the wheel and that's OK but as soons as we have a computer that can make an actual decision in the time available to it then it becomes some sort of moral dilemma.

Even if the computer is instructed to protect the occupants it'll be no worse than what's out there now and that's before you factor in the reduction in likelihood of the vehicle being in a position where it has to make that decision anyway.
Not only that but if the computer has to make that decision it's unlikely to be a black and white one. If it has time to evaluate the threat it can also work out the odds of success and the numbers of casualties.
By far the cleverest thing here is the imagination applied to thinking up increasingly bizarre and unlikely scenarios that the computer *needs* to be able to solve.

I've probably also mentioned this before. Long before self driving cars are prevalent you won't be able to buy a car that isn't capable of a fair degree of collision and hazard avoidance. By then you'll have a pretty good idea of just how safe these vehicles are likely to be. In my experience 99.9% of 'incidents' I encounter daily are the result of having a monkey behind the wheel and *all* of them go away as soon as you put something in control that has patience and a basic understanding of vehicle dynamics.

[the snail]

I can see big legal issues for the manufacturers if they design a system that decides to kill a third party -wouldn't they be guilty of murder/manslaughter? This seems to be a major change with autonomous vehicles - the passengers have no control, so there's no blame/negligence in an accident. The blame falls on the designer of the algorithm that controls the vehicle. Especially in the U.S. there are potentially huge punitive damages for the manufacturer to pay.

[thirdcrank]

The thread title refers to "public opinion" and I'm suggesting that there are already quite strong lobbying attempts by the manufacturers to sway opinion. In case anybody has missed it, the govt., has been falling over itself to promote this technology AKA woo the manufacturers.

[Annoying Twit]

I must admit that I'm more concerned about the quality of the computer vision (including sonar and any other electronic senses) than I am about the reasoning of the car. There may be significant risk of 'sorry human, I didn't see you.'

[kwackers]

The thread title refers to "public opinion" and I'm suggesting that there are already quite strong lobbying attempts by the manufacturers to sway opinion. In case anybody has missed it, the govt., has been falling over itself to promote this technology AKA woo the manufacturers.

Public opinion is fickle. The first computer controlled car to run over a child "deliberately" will probably be the last.

Accidents involving self driving cars are likely to be fairly scarce. Imagine how low accidents rates would be if the monkey brains in charge now took a little care? They'd at worse match that rate and due to improved sensors and 360 vision better it.
With low accident rates there'll be a lot of interest in each one and if I were a manufacturer I'd be quite keen to avoid adverse publicity.

Any attempts by manufacturers at the moment are made by the old gits in overall charge who have little concept of what a self driving car is and assume it's like a monkey brain in a jar with all the same problems. If you assume this then you also assume it'll make all the same mistakes but that you might be liable for them so trying to change perception isn't a bad idea.
They're wrong of course.

[kwackers]

I must admit that I'm more concerned about the quality of the computer vision (including sonar and any other electronic senses) than I am about the reasoning of the car. There may be significant risk of 'sorry human, I didn't see you.'

Plenty of improvements happening all the time (and I'm still of the opinion that genuine self driving cars are 10 years away).
But they only have to drive without a phone in front of their faces to be an improvement.

I guess the real improvement here is when the car detects the sensors are below par. A vehicle that actually drives to the conditions! Imagine that...

I do wonder what sort of hybrid system we'll end up with. I doubt it would just be optical - looking at people tells us just how poor that can be, currently they're backed up with lidar and radar systems.
Perhaps optical and IR would be enough? Together they could pick up enough detail from the surroundings especially if they are multiple cameras covering 360 degrees.