Advanced Driving Hub

Straw man argument. If an autonomous car is making decisions about harming humans, it's doing it wrong. Surely that was the whole point of introducing them in the first place?

Surely it's being honest and accepting that, sooner or later, this situation will occur? Your expectation might vary - but realism says that nothing is perfect, so it would be a bit remiss to not consider it.

Horse wrote:Surely it's being honest and accepting that, sooner or later, this situation will occur? Your expectation might vary - but realism says that nothing is perfect, so it would be a bit remiss to not consider it.

If you start considering it, you then start making excuses about shoddy software being acceptable because the car can make ethical decisions.... and what's to say you won't have bugs in the ethics code too?

Sure, it's going to go wrong, but I'd rather it started with the premise that it can't be allowed to and we design to that quality.

jont- wrote:

Horse wrote: but realism says that nothing is perfect, so it would be a bit remiss to not consider it.

If you start considering it, you then start making excuses about shoddy software being acceptable because the car can make ethical decisions.... and what's to say you won't have bugs in the ethics code too?

Sure, it's going to go wrong, but I'd rather it started with the premise that it can't be allowed to and we design to that quality.

Probably more difficult to write software that incorporates those aspects! But why do you think that designers won't be trying to prevent those situations occurring? Another real-world parallel: you develop and practice mental and physicals skills to avoid bad situations - but learn how to do emergency stops for when they do happen.

Horse wrote:

jont- wrote:

Horse wrote: but realism says that nothing is perfect, so it would be a bit remiss to not consider it.

If you start considering it, you then start making excuses about shoddy software being acceptable because the car can make ethical decisions.... and what's to say you won't have bugs in the ethics code too?

Sure, it's going to go wrong, but I'd rather it started with the premise that it can't be allowed to and we design to that quality.

Probably more difficult to write software that incorporates those aspects! But why do you think that designers won't be trying to prevent those situations occurring? Another real-world parallel: you develop and practice mental and physicals skills to avoid bad situations - but learn how to do emergency stops for when they do happen.

Engineers are lazy. If they know there's a fallback, they will start to rely on it.

At some level we all accept that a pedestrian who dives sideways into a road at the wrong moment will get hit, or a car that runs a red into a stream of traffic, or swerves to the other side of the road in the wrong situation. It doesn't mean that an autonomous car is wrong for getting into this situation, so it may need to consider evasive action without there being any planning failure. It's rare, but happens.

jont- wrote:Engineers are lazy. If they know there's a fallback, they will start to rely on it.

Ah, that will be the 'glass half leakey' view of engineering?

Perhaps you should be more concerned about AVs being built down to a price point, with compromise as a result.

We know how fast pedestrians can move sideways (it's not an infinite speed) and have automatic cars moderate their speed when passing humans at that distance, such that any impact will not do significant harm (another calculable limit that we can stay well below). This is likely to mean that vehicles will move at walking speed in close proximity to pedestrians (which is no bad thing) and be able to speed up when their trajectory puts them at significant distance from pedestrians.
Just like good human drivers of course, who slow down when passing through a crowded shopping street and speed up on motorways.

Horse wrote:

jont- wrote:Engineers are lazy. If they know there's a fallback, they will start to rely on it.

Ah, that will be the 'glass half leakey' view of engineering?

Perhaps you should be more concerned about AVs being built down to a price point, with compromise as a result.

That too. Again, why waste expensive compute on solving difficult problems if you can just slam the brakes on and say "it did its best".

Actually, one positive outcome from dieselgate would be looking for cheating on any tests that come in too.

ancient wrote:We know how fast pedestrians can move sideways (it's not an infinite speed) and have automatic cars moderate their speed when passing humans at that distance, such that any impact will not do significant harm (another calculable limit that we can stay well below). This is likely to mean that vehicles will move at walking speed in close proximity to pedestrians (which is no bad thing) and be able to speed up when their trajectory puts them at significant distance from pedestrians.
Just like good human drivers of course, who slow down when passing through a crowded shopping street and speed up on motorways.

awesome - so pedestrians will just walk across the road whenever they like, knowing that cars will stop - at present there is a balance - pedestrians take care and look etc. as they can't guarantee that the cars will stop - knowing that the cars will stop will change pedestrian habits, maybe over time, but it will happen...

Alasdair