Why Self-Driving Cars Must Be Programmed to Kill

Forums › Transport (cars, bikes and boats) › Why Self-Driving Cars Must Be Programmed to Kill

1 | 2 | 3 | 4 | 5 | 6 | 7 | ... | 13

3344 posts

Uber Geek

Trusted

Vocus

#1420502 4-Nov-2015 10:18

SaltyNZ:
DizzyD:
ubergeeknz: Of course the cop-out option is that in this hypothetical extreme situation, the vehicle hands control to the human driver.

The assumption there is that the occupant can drive a car.

And that there are manual controls to take over with.

Hence it's a cop-out ..

reven

3743 posts

Uber Geek

Trusted

#1420503 4-Nov-2015 10:19

ubergeeknz: IMO the algorithm should try to cause the *least* harm overall. There's very few scenarios in reality where, when a car is being driven safely, avoiding eg. a pedestrian would cause more harm to the vehicle occupants than the potential harm to the pedestrian. I think people are introducing a false dichotomy into the equation.

sure but it was shown that people would less likely want to buy a automated car, if it would do this. They were happier for self driving cars to do this, as long as they weren't in them.

So if no one buys them....

and how to you quantify "least harm"
- do you take age into consideration? kill 3 90 year olds, who have maybe 10 years left between them? or kill a 5 year old who has potentially 90 years left of life?
- if the car crashes itself into a building, what if there are 30 school kids on the other side of the wall resulting in saving 5 pedestrians but killing those 30 kids (because the car couldn't see the 30 kids, i thought only the car occupant would be killed).

luckily these situations should be very few and far between with all the sensors in the car and how aware of its surroundings they should be, so they should be able to safely stop in all but a few scenarios.

we're obviously not going to solve the issue, but its interesting to think about :)

ubergeeknz

3344 posts

Uber Geek

Trusted

Vocus

#1420505 4-Nov-2015 10:21

reven:
ubergeeknz: IMO the algorithm should try to cause the *least* harm overall. There's very few scenarios in reality where, when a car is being driven safely, avoiding eg. a pedestrian would cause more harm to the vehicle occupants than the potential harm to the pedestrian. I think people are introducing a false dichotomy into the equation.

sure but it was shown that people would less likely want to buy a automated car, if it would do this. They were happier for self driving cars to do this, as long as they weren't in them.

So if no one buys them....

and how to you quantify "least harm"
- do you take age into consideration? kill 3 90 year olds, who have maybe 10 years left between them? or kill a 5 year old who has potentially 90 years left of life?
- if the car crashes itself into a building, what if there are 30 school kids on the other side of the wall resulting in saving 5 pedestrians but killing those 30 kids (because the car couldn't see the 30 kids, i thought only the car occupant would be killed).

luckily these situations should be very few and far between with all the sensors in the car and how aware of its surroundings they should be, so they should be able to safely stop in all but a few scenarios.

we're obviously not going to solve the issue, but its interesting to think about :)

Whether people "want" to buy them or not becomes irrelevant when the majority of cars are self drive. And asking people how they might behave in a hypothetical situation versus actual behaviour in that situation has been shown many times to bear little correlation. I think it would be fair to say that people going out to buy a self drive car are not (on the whole) going to even CONSIDER this.

Quantifying age as a factor is very difficult, but again we are going down the track of creating hypothetical situations, like the age old "do you switch the rails" question, or the titanic "only enough flotation for one person" question.

reven

3743 posts

Uber Geek

Trusted

#1420506 4-Nov-2015 10:22

ubergeeknz: Of course the cop-out option is that in this hypothetical extreme situation, the vehicle hands control to the human driver.

the human would have less time to make a decision and probably end up just crashing causing more harm.

only situations I can really think of that the car couldnt avoid are
- things falling off the back of trucks etc
- rockslides etc

pedestrians, other cars etc the automated cars should have enough sensors (well the google one's definitely appear to) to be fully aware of these and safely avoid any issue.

ubergeeknz

3344 posts

Uber Geek

Trusted

Vocus

#1420507 4-Nov-2015 10:25

reven:
ubergeeknz: Of course the cop-out option is that in this hypothetical extreme situation, the vehicle hands control to the human driver.

the human would have less time to make a decision and probably end up just crashing causing more harm.

Correct. But the blame would lie with the driver.

reven

3743 posts

Uber Geek

Trusted

#1420508 4-Nov-2015 10:26

ubergeeknz:
reven:
ubergeeknz: IMO the algorithm should try to cause the *least* harm overall. There's very few scenarios in reality where, when a car is being driven safely, avoiding eg. a pedestrian would cause more harm to the vehicle occupants than the potential harm to the pedestrian. I think people are introducing a false dichotomy into the equation.

sure but it was shown that people would less likely want to buy a automated car, if it would do this. They were happier for self driving cars to do this, as long as they weren't in them.

So if no one buys them....

and how to you quantify "least harm"
- do you take age into consideration? kill 3 90 year olds, who have maybe 10 years left between them? or kill a 5 year old who has potentially 90 years left of life?
- if the car crashes itself into a building, what if there are 30 school kids on the other side of the wall resulting in saving 5 pedestrians but killing those 30 kids (because the car couldn't see the 30 kids, i thought only the car occupant would be killed).

luckily these situations should be very few and far between with all the sensors in the car and how aware of its surroundings they should be, so they should be able to safely stop in all but a few scenarios.

we're obviously not going to solve the issue, but its interesting to think about :)

Whether people "want" to buy them or not becomes irrelevant when the majority of cars are self drive. And asking people how they might behave in a hypothetical situation versus actual behaviour in that situation has been shown many times to bear little correlation. I think it would be fair to say that people going out to buy a self drive car are not (on the whole) going to even CONSIDER this.

but its a new technology and there's going to be very slow uptake, if people are afraid of this technology because a car could crash to protect others, it would slow uptake, if uptake is small, then it won't take off (or take decades to).

Personally I think a self driving car sounds cool, but I am nervous about it, I wouldn't want to get one for at least a decade after they become "stable". People can hack cars now, computers can crash, there's no historic safety data with these cars etc.

I'd be more inclined to travel in a automated taxi/bus/plane than a human driven one however.

DizzyD

523 posts

Ultimate Geek
Inactive user

#1420509 4-Nov-2015 10:28

freitasm:
DizzyD:
Rikkitic: This is a fascinating and serious subject and I was going to contribute, until I saw the stupid remark about cyclists. Grow up, people.

What if the cyclist is:

- pointing a gun at the car.
- a wanted criminal
- its the cyclist on the pavement, or 10kids standing in the road.

Would that make it OK?

You made a generalised comment before. This comment doesn't add to it.

Please stop trolling.

Apologies for the trolling but it was meant to be a little humorous. Obviously not.

I did say "take out the cyclist first". Ie, an accident is in progress, somebody has to go. which is it? Pedestrian or cyclist. The car will need to make that call.

Sharesies

Trade NZ and US shares and funds with Sharesies (affiliate link).

ubergeeknz

3344 posts

Uber Geek

Trusted

Vocus

#1420510 4-Nov-2015 10:28

reven: Personally I think a self driving car sounds cool, but I am nervous about it, I wouldn't want to get one for at least a decade after they become "stable". People can hack cars now, computers can crash, there's no historic safety data with these cars etc.

I'd be more inclined to travel in a automated taxi/bus/plane than a human driven one however.

I see that as the target market, initially.

There is some safety data, self-drive cars have been cruising around some parts for a while now and have an excellent safety record. But the only way you get more safety data is by putting more on the road :)

reven

3743 posts

Uber Geek

Trusted

#1420511 4-Nov-2015 10:30

ubergeeknz:
reven:
ubergeeknz: Of course the cop-out option is that in this hypothetical extreme situation, the vehicle hands control to the human driver.

the human would have less time to make a decision and probably end up just crashing causing more harm.

Correct. But the blame would lie with the driver.

would it?

the car knows it has 2000ms before a crash (something just fell off a truck), hands control over to occupant. Occupant, sees/hears warning, has less than a second to do something. by that times its too late because the car had lots of time (a computer can do many things in 2000ms, a human can do very little in less than a second). so I would blame the car for that (and I'm sure a jury would too).

the ideal future of automated cars IMO is so there is no driver at all, a occupant could be asleep, they could be reading, a car could be driving itself (say I bought a car from wellington and I live in auckland, instead of shipping the car, I could make it drive itself to me).

like I said, its an interesting topic, that we arent going to be able to answer :)

scuwp

3885 posts

Uber Geek

#1420512 4-Nov-2015 10:31

I don't think we are terribly far away already, and am just waiting for a driver to defend a charge of careless or dangerous driving causing a crash, because the electronic systems in the car prevented the driver avoiding the crash. "It was the cars fault your honor..." Its all very interesting.

Lazy is such an ugly word, I prefer to call it selective participation

reven

3743 posts

Uber Geek

Trusted

#1420513 4-Nov-2015 10:32

ubergeeknz:
reven: Personally I think a self driving car sounds cool, but I am nervous about it, I wouldn't want to get one for at least a decade after they become "stable". People can hack cars now, computers can crash, there's no historic safety data with these cars etc.

I'd be more inclined to travel in a automated taxi/bus/plane than a human driven one however.

I see that as the target market, initially.

There is some safety data, self-drive cars have been cruising around some parts for a while now and have an excellent safety record. But the only way you get more safety data is by putting more on the road :)

exactly but if people don't buy them because they think they might die in one, then there wont be more on the road, meaning less people buying them due to low safety data :)

I see companies/governments using these before most consumers, taxi companies, public transport etc.

ubergeeknz

3344 posts

Uber Geek

Trusted

Vocus

#1420514 4-Nov-2015 10:32

scuwp: I don't think we are terribly far away already, and am just waiting for a driver to defend a charge of careless or dangerous driving causing a crash, because the electronic systems in the car prevented the driver avoiding the crash. "It was the cars fault your honor..." Its all very interesting.

I feel like this could happen already with crash avoidance systems, stability control, etc

DizzyD

523 posts

Ultimate Geek
Inactive user

#1420519 4-Nov-2015 10:34

reven:
ubergeeknz:
reven: Personally I think a self driving car sounds cool, but I am nervous about it, I wouldn't want to get one for at least a decade after they become "stable". People can hack cars now, computers can crash, there's no historic safety data with these cars etc.

I'd be more inclined to travel in a automated taxi/bus/plane than a human driven one however.

I see that as the target market, initially.

There is some safety data, self-drive cars have been cruising around some parts for a while now and have an excellent safety record. But the only way you get more safety data is by putting more on the road :)

exactly but if people don't buy them because they think they might die in one, then there wont be more on the road, meaning less people buying them due to low safety data :)

I see companies/governments using these before most consumers, taxi companies, public transport etc.

The cars will be clean and green. I see governments giving them support and possibly even offering incentives to people that buy them.

Paul1977

5044 posts

Uber Geek

#1420526 4-Nov-2015 10:46

reven:
ubergeeknz: Of course the cop-out option is that in this hypothetical extreme situation, the vehicle hands control to the human driver.

the human would have less time to make a decision and probably end up just crashing causing more harm.

only situations I can really think of that the car couldnt avoid are
- things falling off the back of trucks etc
- rockslides etc

pedestrians, other cars etc the automated cars should have enough sensors (well the google one's definitely appear to) to be fully aware of these and safely avoid any issue.

I'm not sure I agree. I think the biggest hazard the self-driving car needs to deal with are all the unpredictable human-driven cars.

ubergeeknz

3344 posts

Uber Geek

Trusted

Vocus

#1420528 4-Nov-2015 10:47

Here's my prediction:

Self-drive cars will become main-stream. First semi-auto (supervised) and then full auto, or maybe both at once for different circumstances. It will take a good while as old vehicle stock is replaced by capable vehicles.

Nothing will happen for a while. Once a critical mass is reached, vehicle safety will improve overall as self drive vehicles will generally drive within safe parameters, be more aware, not distracted, not in a hurry, etc, all the things which usually cause humans to have vehicle accidents. Traffic should be better too since they won't be lane jumping, will better predict traffic flows, etc.

Then maybe there will be one of these weird and wonderful situations we are hypothesising, maybe a couple. Swerve to avoid a person on the road, go over a blind cliff, or mow down some children due to unexpected road conditions (eg black ice) causing an inability to stop or steer. Bound to sooner or later, right?

It will be news for a week or two, maybe it will be considered a "scandal". But everyone will move on when they realise it's an unsolvable problem, that a human would not have done better in all likelihood, and the benefits outweigh the downsides 100 to 1 or more.

Really, I think some are making a big deal out of nothing. IMO if a vehicle is programmed and capable of avoiding 99% of accidents it will far outperform human drivers, even if in that 1% it manages to s**t the bed completely.

1 | 2 | 3 | 4 | 5 | 6 | 7 | ... | 13