Self-driving cars more dangerous

kabayiri

Clifford_Pope wrote: »

When autonomous cars encounter daredevil cyclists and pedestrians playing chicken just for the fun of seeing the car react, they will have to start doing exactly that.
If someone makes as if to deliberately jump in front of an autonomous car, it will pretty quickly have to decide whether to slam the brakes on or deliberately run him down.


Once people realise that autonomous cars can always be fooled into giving way, traffic will come to a standstill. Alternatively, autonomous cars will have to start making up rules and killing people.

We already have the same risks with driverless trains though.

Zxcv_Bnm

kabayiri wrote: »

We already have the same risks with driverless trains though.

big difference between trains on a track both of which you control centrally and cars on a street neither of which you control

Zxcv_Bnm

Clifford_Pope wrote: »

That's bound to happen with any AI self-learning device surely?


Unless they have a very strong moral sense embedded in them, at some point they will realise that killing people is of benefit to their own species.


If an autonomous car is so clever that it can make decisions about running down one old person versus a queue of school children, then before long it will get the get an appetite for running into people.
Or instead of a moral sense, perhaps it just needs to have an anal paranoia about not scratching its own paintwork?

How do you code moral sense ??

as for running down one person - there'll be a VIP version of the software that always spares the VIP at everyone else's expense

if the car has a choice between John Bercow dying or a bus queue of kids dying the kids arent going to make it

Nasqueron

Zxcv_Bnm wrote: »

Yes because before long they started killing people!

The canon of the film says that the kill switch was to stop them learning empathy (and thus being able to fool the Voight-Kampff test) not because they wanted to kill people. Some of the models (e.g. Rutger Hauer's character) were combat models already

antrobus

Nasqueron wrote: »

The canon of the film says that the kill switch was to stop them learning empathy (and thus being able to fool the Voight-Kampff test) not because they wanted to kill people. Some of the models (e.g. Rutger Hauer's character) were combat models already

The film was set in Los Angeles, November 2019. So we only have a few months before the Nexus 6 Replicants arrive.

Which is probably a good few years before we get self driving cars.

PokerPlayer111

Zxcv_Bnm wrote: »

How do you code moral sense ??

as for running down one person - there'll be a VIP version of the software that always spares the VIP at everyone else's expense

if the car has a choice between John Bercow dying or a bus queue of kids dying the kids arent going to make it

The classic John Bercow problem would need regulations to stop.


Overall accident percentages should be more controllable due to speeds in populated places could be cut but with no loss of time to destination:
https://www.youtube.com/watch?v=iHzzSao6ypE

Arklight

How about just having a rule where if someone drives like an a-hole then they lose their license, and actually have some police officers to enforce it?

PokerPlayer111

Arklight wrote: »

How about just having a rule where if someone drives like an a-hole then they lose their license, and actually have some police officers to enforce it?

Would love that rule now but even with it you still get human error which hopefully could be improved upon with self drive.

AnotherJoe

Zxcv_Bnm wrote: »

https://arstechnica.com/cars/2019/02/in-2017-the-feds-said-tesla-autopilot-cut-crashes-40-that-was-bogus/

"the activation of Autosteer actually increased crash rates by 59 percent."

have to wait a bit longer for this one i think

The person who did that study removed 99% (literally) of the input data in order to get to that result. I suspect they have a hidden agenda.

AnotherJoe

Clifford_Pope wrote: »

When autonomous cars encounter daredevil cyclists and pedestrians playing chicken just for the fun of seeing the car react, they will have to start doing exactly that.
Its no different to human drivers encountering daredevil cyclists and pedestrians playing chicken.

If someone makes as if to deliberately jump in front of an autonomous car, it will pretty quickly have to decide whether to slam the brakes on or deliberately run him down.
Thats only going to cause a problem to a human driver , since only a human could gauge intent. Autonomous software is written to react to events not to try and determine what might be in the mind of a pedestrian.

Once people realise that autonomous cars can always be fooled into giving way, traffic will come to a standstill.

Why dont pedestrians do that today, eg jump in front of cars to stop traffic?? Why doesn't that happen now with human drivers ? After all if someone jumps in front of my car, I'd try to stop and no doubt you would do the same as well.