Philosophy of programming a driverless car. Kill or be killed.

Technology in driving is becoming more dominant...
User avatar
ChristianAB
Posts: 270
Joined: Mon Sep 28, 2015 9:51 pm

Re: Philosophy of programming a driverless car. Kill or be killed.

Postby ChristianAB » Sat Nov 14, 2015 12:16 am

Or better: gang of 10 psychos robs banks, then panic and flee, as they run across the road unexpectedly, your autonomous car then has to decide whether to kill them, or stop unexpectedly henceforth killing you since there is a 30 tons lorry furiously tailgating just behind.
So yeah, you end up dead, for the sake of 10 psychos.

Another one: bored kids in suburbia will play at jumping unexpectedly in front of driverless car, just to enjoy seeing them swerve. It will be one-of-the-thing-to-do-before-you-die, if indeed it doesn't kill you instantly. The winner is the one who jumps closest to the approaching car.

I could go on. :twisted: I will enjoy watching the automakers struggle with this one. I'm curious to see how they will justify their offerings.
Last edited by ChristianAB on Mon Nov 23, 2015 11:25 am, edited 1 time in total.

WhoseGeneration
Posts: 374
Joined: Tue Sep 29, 2015 5:02 pm

Re: Philosophy of programming a driverless car. Kill or be killed.

Postby WhoseGeneration » Mon Nov 16, 2015 12:01 am

I'm surprised any car manufacturers want to get involved with this.
Think about it, driverless cars will have to obey speed limits. They won't "think", "Oh lovely series of bends on this well sighted rural B road, let's have some fun".
A driverless Ferrari, Aston, Caterham, McLaren, come on.
There'll be only a few types of car needed, the urban runaround, the motorway cruiser, the family wagon and the inhospitable areas one.
What's the point of driveway prestige when everyone knows the bloody thing drives itself and there's no way of asserting your "superiority" on the road?

CautiousD
Posts: 32
Joined: Wed Oct 07, 2015 2:29 pm

Re: Philosophy of programming a driverless car. Kill or be killed.

Postby CautiousD » Sun Nov 22, 2015 11:37 pm

The thing is, that without the immense variability of the human conscience and factors which influence it, how will a computer make anything close to a sentient decision? It's more likely to be the case that drivers will need to either fill in a lengthy cross-reffed psychological survey, upon which their car will be programmed (don't start to dissect that as you'll go mad trying) or that there will be a variety of driving modes available dependent upon driving experience, knowledge and track record. The result may well be the entirely predictable outcome of still being stuck behind 40mphers on NSLs.
I think/ hope the golden age of modern motoring is going to be unprogrammed electric cars, as I hate rollercoasters.

User avatar
ChristianAB
Posts: 270
Joined: Mon Sep 28, 2015 9:51 pm

Re: Philosophy of programming a driverless car. Kill or be killed.

Postby ChristianAB » Tue Feb 16, 2016 8:58 pm

Except that my car will be 'programmed' to overtake those 40mphers rather aggressively... 8-)

More seriously, a car doesn't have to be sentient to be effective in most situations.In the same way that IPSGA works all the time except when some truly unexpected happens. What's also likely is that road designs might evolve to meet autonomous cars halfway

trashbat
Posts: 21
Joined: Wed Sep 30, 2015 2:56 pm

Re: Philosophy of programming a driverless car. Kill or be killed.

Postby trashbat » Tue Feb 16, 2016 10:19 pm

This is a made up problem.

Noone is ever - in the foreseeable future of autonomous systems - going to program cars to make value judgements. It's always going to be best effort avoidance, which is probably going to be maximum braking for the most part, with possibly a few set piece routines within given parameters, like swerve into an empty lane if something intrudes into the path. If someone dies somehow as a result, then it won't be because the car made any kind of meaningful decision to do it, FFS, and it's nonsense to claim that it will be.

At most it will be a matter of negligence in system design and deployment.

And this is exactly what humans do too. You can trot out the Trolley Problem all you like, but pretty much noone in an ongoing car accident ever decides exactly what to swerve into with a rationale for it. It's a pre-baked response.

Most of the other scaremongering or 'what ifs' about AVs are junk too.

User avatar
Horse
Posts: 3557
Joined: Mon Sep 28, 2015 9:20 am

Re: Philosophy of programming a driverless car. Kill or be killed.

Postby Horse » Tue Feb 16, 2016 11:01 pm

trashbat wrote:This is a made up problem.

Noone is ever - in the foreseeable future of autonomous systems - going to program cars to make value judgements. It's always going to be best effort avoidance, which is probably going to be maximum braking for the most part, with possibly a few set piece routines within given parameters, like swerve into an empty lane if something intrudes into the path.


Guessing, the programming will be based heavily around speed 'choice', with 'slow down' as the default response to any potentially surprising situation.

Not a bad bit of advice for all of us, really.
Your 'standard' is how you drive alone, not how you drive during a test.

trashbat
Posts: 21
Joined: Wed Sep 30, 2015 2:56 pm

Re: Philosophy of programming a driverless car. Kill or be killed.

Postby trashbat » Wed Feb 17, 2016 9:25 am

Horse wrote:Guessing, the programming will be based heavily around speed 'choice', with 'slow down' as the default response to any potentially surprising situation.

Not a bad bit of advice for all of us, really.

Probably - because it is after all a continuation of the trend for things like adaptive cruise and autonomous emergency braking (AEB). The sorts of things a child could do over & over again using minimal rules - but would get bored of and eventually fail at, which is why having a machine do it is attractive.

People talk about autonomous operations & AI and expect something exotic, some sentient self-aware beast that can outperform and outwit humans at every possible turn. Well, no, that's not what you're getting any time soon, well into the days of self-driving cars. You're getting something that's distinctly based on rules and measurements, maybe a sophisticated ruleset with frills on, but that's still what it is. And it doesn't have to outperform an expert driver on a great day after their morning coffee, it just has to outperform the average person. The average person probably doesn't care about driving, might well be on their phone and is very far from an expert.

So if presented with this Trolley Problem emergency scenario, Occam's razor will apply, and the car will brake really hard on the predestined trajectory, but despite that, it might very well kill the pedestrian that's in its way. But if a thousand average human drivers did the same, at least a thousand people would die. A thousand robots might actually only kill a high nine hundred, and that's ultimately the criteria for acceptance, even though that's anathema to experts who think they as individuals can do better.

User avatar
Horse
Posts: 3557
Joined: Mon Sep 28, 2015 9:20 am

Re: Philosophy of programming a driverless car. Kill or be killed.

Postby Horse » Wed Feb 17, 2016 10:55 am

trashbat wrote:
Horse wrote:Guessing, the programming will be based heavily around speed 'choice', with 'slow down' as the default response to any potentially surprising situation.

Not a bad bit of advice for all of us, really.


So if presented with this Trolley Problem emergency scenario, Occam's razor will apply, and the car will brake really hard on the predestined trajectory, but despite that, it might very well kill the pedestrian that's in its way. But if a thousand average human drivers did the same, at least a thousand people would die. A thousand robots might actually only kill a high nine hundred, and that's ultimately the criteria for acceptance, even though that's anathema to experts who think they as individuals can do better.


Yup, I think that sums it up.

Hopefully, the vehicle will be focussed on identifying the opportunity for surprise - e.g. the pedestrian running out, looking for hidden areas from where something might happen - on the basis that 'if it happens, I can stop'. That way the 'kill either/or' scenarios are less likely to occur.

Boring, but [probably] true.
Your 'standard' is how you drive alone, not how you drive during a test.


Return to “Technology”

Who is online

Users browsing this forum: No registered users and 13 guests