Re:Self-driving car accidents: Robot drivers are ‘odd, and t...
by mikeash on Monday October 16, 2017 @11:15AM
(#15483749)
I don't see how "They blast through stop signs and red lights without heed to traffic or pedestrians" is in any way unclear.
Of course autonomous cars need to predict and handle sloppy human driving. But that's not what the article is talking about. It's talking about programming the cars to break the rules "in a safe and principled way" the way humans do because humans will expect it. I don't even have a problem with programming the cars to break the rules in a safe and principled way. Where I have a problem is when the article points to humans as a model because that's what they do. It's not.
> Every given decision in driving is a balance of risk and reward....
This is probably the center of our disagreement. Every given decision should be a balance of risk and reward, yes. In reality, for 99% of drivers, every given decision is a balance of reflex, selfishness, and vague remembrance of rules they learned 30 years ago and forgot 29 years ago.
It's true that safely breaking the rules is not always an accident. It's merely an accident 99% of the time.
There's a busy intersection near my house which I cross on foot fairly often. Cars routinely turn right on red without stopping first. Every time I cross that street, I have to carefully watch the oncoming cars until it looks like they're going to stop, because if I just start walking when the walk sign comes on, I will get hit. These drivers aren't risking my life because they're balancing risk and reward. They're doing it because they want to beat traffic from the other road and are too clueless to understand that what they're doing is dangerous.
Again, I have no issue with programming autonomous cars to accommodate flawed human drivers, or even with programming them to deliberately break the rules in cases where it's a net gain. I merely object to the proposal that humans are the model to follow in that respect.
Of course autonomous cars need to predict and handle sloppy human driving. But that's not what the article is talking about. It's talking about programming the cars to break the rules "in a safe and principled way" the way humans do because humans will expect it. I don't even have a problem with programming the cars to break the rules in a safe and principled way. Where I have a problem is when the article points to humans as a model because that's what they do. It's not.
> Every given decision in driving is a balance of risk and reward....
This is probably the center of our disagreement. Every given decision should be a balance of risk and reward, yes. In reality, for 99% of drivers, every given decision is a balance of reflex, selfishness, and vague remembrance of rules they learned 30 years ago and forgot 29 years ago.
It's true that safely breaking the rules is not always an accident. It's merely an accident 99% of the time.
There's a busy intersection near my house which I cross on foot fairly often. Cars routinely turn right on red without stopping first. Every time I cross that street, I have to carefully watch the oncoming cars until it looks like they're going to stop, because if I just start walking when the walk sign comes on, I will get hit. These drivers aren't risking my life because they're balancing risk and reward. They're doing it because they want to beat traffic from the other road and are too clueless to understand that what they're doing is dangerous.
Again, I have no issue with programming autonomous cars to accommodate flawed human drivers, or even with programming them to deliberately break the rules in cases where it's a net gain. I merely object to the proposal that humans are the model to follow in that respect.