Free startup idea: Orange clothing line with white stripes as a auto-pilot deterrent.
(This is why the trolley problem for AVs isn't real.)
I don't think I that will be accepted by regulators/society until it can be demonstrated that AV's are undeniably safer-- not just as good as human drivers, much much better. Right now it's newsworthy whenever a Tesla auto pilot drives into a barrier. We'll need to get to a point where not only does that not happen, but also that it becomes newsworthy every time an AV saves a life-- swerving out if the way faster than any human reflexes could have reacted to avoid an accident, or a child chasing a ball into the street, etc.
Example running over an animal to avoid an accident, getting rear ended to avoid hitting a pedestrian, taking a calculated risk of being hit by oncoming traffic on a 2 lane situation when coming across a stalled car without optimal view on the road ahead.
It's going to make a decision between different risk factors and between different known costs.
Imagine on a 2 lane there is a pedestrian and oncoming traffic and insufficient time to stop. Do you kill the pedestrian or risk ending up in the ditch.
If you are going 60 and ending up in the ditch would likely be fatal is it ok to just smear the pedestrian?
What if you can slow to 20 and then ditch?
These situations won't hopefully be prevalent because a massive portion of our current issues are down to inattention or stupidity and we believe ai will eventually be much better at avoiding such problems but they are still important and interesting.
Yeah but not trolley problem style. "Don't swerve" is a valid strategy for a car.
> Example running over an animal to avoid an accident
Depends on the kind of accident you have in mind, so I can't give real feedback here.
> getting rear ended to avoid hitting a pedestrian
Always brake to avoid hitting a pedestrian. You don't choose whether to get rear-ended and it won't be your fault either.
> taking a calculated risk of being hit by oncoming traffic on a 2 lane situation when coming across a stalled car without optimal view on the road ahead.
When one of the choices is "sit there and wait longer", that's definitely not a trolley problem. It either thinks going forward is sufficiently safe or it doesn't. It can wait all day for someone to intervene.
> Imagine on a 2 lane there is a pedestrian and oncoming traffic and insufficient time to stop. Do you kill the pedestrian or risk ending up in the ditch.
If it's programmed not to swerve, it will hit the pedestrian, which is an acceptable outcome of an unacceptable situation.
> If you are going 60 and ending up in the ditch would likely be fatal is it ok to just smear the pedestrian?
It's sufficiently ok when the only alternative is so reckless. Getting into that situation is the not-ok part.
> What if you can slow to 20 and then ditch?
If you can slow to 20 then hit the pedestrian at 20.
> Depends on the kind of accident you have in mind, so I can't give real feedback here.
Typical scenario here is a dog appearing on your lane in a high-traffic, multi-lane road. You don't swerve, you don't even break too hard, because hitting the dog is much better outcome than hitting another car, or landing in a ditch, or getting rear ended.
Emergency breaking can typically stop a car within about 100 yards even from freeway speeds. But a distracted human might travel more than that before they even realize they should be trying to stop.
Humans are also pretty bad at adjusting to conditions, so if weather makes the road slick (wet or icy) or they have limited visibility, they don't often slow down enough to compensate.
Autonomous cars have none of these problems, so aren't likely to ever run in to a "can't stop in time" situation.
I can't actually think of any situation where, in the event of a likely collision, the best solution isn't to just try to stop. Even if, for some reason, you can't stop quickly enough to avoid the collision, any reduction in speed will result in a reduction in the severity of collision.