[Watsey]

https://www.bbc.co.uk/news/technology-60126014

The law commissions for England, Wales and Scotland are apparently recommending a change in the law such that "Human drivers should not be legally accountable for road safety in the era of autonomous cars".



If this became statute it'd be an absolute minefield.

Two vehicles crash; one is in autonomous mode; the isn't. Who has the liability in the event that the autonomous vehicle/driver is at fault ? The vehicle manufacturer ? Would vehicle manufacturers be required to insure their own vehicles, globally, for inherent/latent defects in their autonomous vehicle software and systems and the plain facts of a crash (i.e. the non-autonomous vehicle was stationary; the autonomous vehicle hit it) ?

Would vehicle insurance for the owner/'driver' be cheaper on the basis that liability rests with the vehicle manufacturer, OR, or more expensive on the basis that proprietary software would be encrypted and not released for analysis by third parties (i.e. insurers won't know what the vehicles are programmed to do, or the AI parameters, i.e. preserve the 'driver' or preserve third parties) ?

As Clarkson said, in a Top Gear episode many years ago...a mother and child steps into the road in front of an approaching car; another car is approaching in the other direction. The computer decides who dies.

This is going to need a huge amount of consideration by the law-makers. At the very least I hope that full autonomous driving will be GPS-restricted to motorways and barrier-separated dual carriageways. At least the vehicles on the same carriageway are travelling in the same direction.

[Nibbles]

I think it's good that they're looking into this now and looking to change the laws. It would be more of a minefield without changing the current laws as it really does need to be defined what is classed as autonomous, what's driver assistance and in all cases what is expected of the driver/user of the car.
The only new tech bar EV's in the latest cars is autonomous/assisted driving and I'm not sure either it, or I am at the level where it's to be trusted. I know if I started using it regularly (and from talking to friends that do) you start to switch off to what is actually happening on the road around you which I think is one of the reasons that these accidents happen as the tech is not quite good enough to cope with every situation that occurs.

[JustChris]

It does sound quite bizarre, also saw this today.

[mikeoz]

Not sure how insurance could work with that?

[Stainy]

Originally Posted by mikeoz

Not sure how insurance could work with that?

Everyone who runs an autonomous car pays Gov a flat fee for 3rd party liability cover. They sort out all claims and there's no need for a NCB as in theory all 'drivers' are the same. (The only fly in the ointment is you can never actually take control of your car as this changes all the risk factors)
Any 'extra' cover like accidental damage, Fire/Theft is still done by an insurer.

For my next trick - world peace and ending famine

[gangzoom]

Originally Posted by Watsey

As Clarkson said, in a Top Gear episode many years ago...a mother and child steps into the road in front of an approaching car; another car is approaching in the other direction. The computer decides who dies.

The reality of this happening is virtually none, in an stress situation like the above our brains go into Type 1 thinking mode - ie its a almost a relfex subconscious decision rather than some complex moral judgment.

The current AI systems are trained on human behaviour so it'll essentially replicate human decision making. But again the reality is computers can react much much faster than humans, its why TC and ABS works so well. The biggest complaint of AI algorithms at present is 'phantom braking', ie the cars are extremely cautious when it detects any sign or hint of vulnerable road users. These AI systems also will NEVER break the speed limit. At 20 mph most accidents in the situation above are 100% avoidable.

I suspect we'll get to a stage where insurance companies will in effect penalises human drives for over riding AI algorithms, for example making the car go faster than 20mph in an urban zone when the above situation may occur. Most accidents on the roads are avoidable at lower speeds or drivers were more cautious- leaving larger gaps to allow for safer braking distances. AI algorithms will have no problems mandating safer/slower driving, the real question is if motorist are willing to do things like obey 20/30mph speed limits.

[Watsey]

Originally Posted by gangzoom

The reality of this happening is virtually none, in an stress situation like the above our brains go into Type 1 thinking mode - ie its a almost a relfex subconscious decision rather than some complex moral judgment.

The current AI systems are trained on human behaviour so it'll essentially replicate human decision making. But again the reality is computers can react much much faster than humans, its why TC and ABS works so well. The biggest complaint of AI algorithms at present is 'phantom braking', ie the cars are extremely cautious when it detects any sign or hint of vulnerable road users. These AI systems also will NEVER break the speed limit. At 20 mph most accidents in the situation above are 100% avoidable.

I suspect we'll get to a stage where insurance companies will in effect penalises human drives for over riding AI algorithms, for example making the car go faster than 20mph in an urban zone when the above situation may occur. Most accidents on the roads are avoidable at lower speeds or drivers were more cautious- leaving larger gaps to allow for safer braking distances. AI algorithms will have no problems mandating safer/slower driving, the real question is if motorist are willing to do things like obey 20/30mph speed limits.

The AI can only compute based on sensors, inputs (data) and programming.

Consider the fact that several people have been killed in Teslas (including trying to drive under an articulated lorry trailer) and a pedestrian who was crossing the road whilst pushing a bicycle being run over by a Volvo which was being co-developed with UBER (from memory), AI is clearly still a long way from being safe enough for me to trust it to make fully autonomous decisions which hold my life in its hands.

Driver assistance can have its advantages but I still maintain that removing the driver from the process (and liability) of driving is a fundamentally bad idea.

[Wills2]

Unless you're in your 20's you won't be alive (if/when) level 5 autonomy is available, turns out we're pretty clever machines after all.

We react on the fly in real time and that's really hard to code.

[Pentland]

Originally Posted by Watsey

Driver assistance can have its advantages but I still maintain that removing the driver from the process (and liability) of driving is a fundamentally bad idea.

Even the simplest of automated transport fails.
Rollercoasters for example.

Current “driver assistance” doesn’t even work properly. High beam assist doesn’t react fast enough, lane departure warnings kick in unnecessarily etc etc.

[gangzoom]

Originally Posted by Wills2

We react on the fly in real time and that's really hard to code.

Well this is the most interesting bit on AI development. The current companies working on a solution, Tesla, Intel, Google all seem to believe the opposite. They believe they can make their code stacks 'super human' through essentially learning from driving experiences across thousands and eventually millions of people. The approach to auto labeling and neural nets that code themselves intheory means these cars can end up 'knowing' pretty much every single situation (and road), simply because computers have pretty much unlimited data storage where as us humans don't.

Tesla AI engineers for example are hardly doing any manual 'coding', they are simply 'supervising' the results of computers writing their code.

I don't disagree with you, as this approach isn't how our brains work. But the people working at these companies are quite literally geniuses. We have to see where it all leads, but if you watch some of the latest FSD beta 10.9 results the outcomes aren't that far away.