Comment by jzwinck | Hacker News Reader

Comment by jzwinck | Hacker News Reader

Parent Full thread

1) Some of the extant cars cede control to the human driver if things get too difficult. This is a problem because people will context switch away from driving and there is no way to get them back in the game in 100ms when the car needs help.

2) Engine failure is not a big deal, it just means coast to a stop on the side of the road with hazard lights on. Blowouts are a little trickier. But what about unusual stuff like hitting a deer? A wheel falling off? Or another car drifting into the side of our car on the expressway? Those are tricky but not truly rare. This is an area where implementors will need to compete yet purchasers will have a very incomplete view of the relative performance of cars on the market (it's not a single number like HP).

3) Hitting a pole instead of a person is an easy choice. Yet one that many humans would not be able to make quickly enough. I think autonomous cars will outperform human drivers in "moral" decision making quite early on.

Thanks.

1) Expecting a human to be on deck within some fraction of a second in challenging times will not fly.

2) Perhaps software in cars can be independently safety tested. Just like cars are collision tested today.

3) Understand about human reaction times. My question is about making moral choices: hit a pedestrian or sacrifice the driver?

Your wording is very interesting on #3, "hit" a pedestrian versus "sacrificing" a driver. I don't think this situation is very likely. A self driving car should be aware of people and would likely be able to compensate without killing the passenger.

If it was a case of hitting a pedestrian versus driving off a cliff I feel like the pedestrian will just be SOL for getting in that situation. Otherwise you could have murder by car relatively easily.

Maybe we could make people register their car settings at the DMV like we do for organ donations. "Would you like to opt-in to the murder-children AI driver, or the martyr driver?"

http://www.technologyreview.com/view/542626/why-self-driving...

That article raises an interesting question that I've seen before. Namely: in the event where loss of life is inevitable, who does the car choose to save, and who does it choose to kill? Pedestrians or occupants?

I expect auto insurance companies will have the final decision on this one. They'll provide lower insurance rates for cars that make the correct decisions minimizing total insurance payouts (whatever those happen to be). With lower insurance costs and therefore a lower effective price, sales of that particular brand of self-driving software should increase. Competitors will then rush to do the same.

Presumably it's better to kill whomever is at fault, which the car will assume is not it?

In the driving class I took when I was 16, we were taught to always prefer crashing into another car instead of a pedestrian if there was no alternative - humans don't have crumple zones designed into them, cars do.

I would assume self-driving cars would be programmed similarly.

Reply on news.ycombinator.com