Hacker News Reader

Comment by nolok | Hacker News Reader

Interesting question I read in another thread on this. Suppose a driverless car is going for a collision scenario, and its computer gets the choice between hurting the driver but no one else, or protecting the driver but hurting several people. What should it take ? Does your car have a responsibility toward you its owner, or should it go for "the greater good" of saving the most lives possible ? And what if it's not your car but one you rent per use "a la Uber", which will probably come to exists rather soon ?

It does not apply to the google car since the way I understand it their maximum emergency level is "full stop", but it's something that would probably come up at some point, and I'm not sure what the "right" answer is, if there is even one.

View on HN

We could ask a human driver the same questions, but no country (that I know of) insists that people be able to answer questions like "You have to choose between killing your spouse sitting next to you and three kids walking across the street. What will you do?" before getting their driver's license.

(In any case, the "correct" answer would be "Why the F--- should I be in such a situation, and what should I have done differently to avoid being in such a situation?")

Honestly, why are people so fascinated with these never-happens-in-a-lifetime ethical thought experiments when it comes to self-driving cars? We people aren't that good with ethical problems either, and we have no problem letting people drive.

It seems like there's a pretty simple solution to this, based on the definition of "ownership" that so many products are trending towards lately: you don't own the car. Even if you've technically purchased it outright, all you really have is a perpetual and transferable license to use a car that's really owned by google. The car has a responsibility toward its true owner - google - to do the least damage to their reputation and their company, so if it has to choose between killing its occupant or killing four other potential google customers, it's going to choose the occupant.

I don't think that solution is simple since humans are INCREDIBLY squicky about making this exact decision (https://en.wikipedia.org/wiki/Trolley_problem).

This just sidesteps the question because Google won't be the only manufacturer. If Toyota's self driving car and Google's self driving car are headed into a collision and Google's AI says "kill Googlecar occupant" and Toyota's AI says "kill Googlecar occupant" then no one's going to buy licenses to drive Googlecars.

But what if it's Larry or Sergei in the car? Should the algorithm take that into account?

Especially when you consider that it may not be certain who the "other people" are. It may have a probabilistic model that says there's a 45% chancce it's hurtling towards a mother pushing a stroller and that it has a 100% that it has a driver sitting in the seat. If it's two actions endanger either the driver or the mother, what kind of calculus should it do?

I was staying on raw numbers without factoring the "some people may deserve saving/protection more than others", because that's yet another can of worms ...

I think too much has been made out of this thought experiment. Practically, I don't think driverless cars would have any strategy other than "full stop". There are too many variables to consider for any other strategy. Of course there will always be Bayesian priors that these cars will use to avoid such situations to ever happen in the first place.

When I was learning to drive my dad told me that if I was going to have to have an accident to hit the softer thing, like instead of hitting a tree hit a car.. Like 2 lane road and an approaching car crosses over and there is no option..

Reply on news.ycombinator.com