3 ms·
Driving drunk is also how a fair number of people drive, especially in remote areas when home is far from the bar. I'm not sure that modeling human behavior is
by Johnny555 5y ago
Driving drunk is also how a fair number of people drive, especially in remote areas when home is far from the bar.
I'm not sure that modeling human behavior is the right way to design a self driving car.
- harambae 5y agoThe human behavior here (slow rolling stop signs) at least in theory can prevent being rear-ended by someone who is used to everyone else in the area doing that (of course it's the rear-ender's fault in a crash... but still an annoyance for everyone involved) There's no benefit to the computer swerving in and out of the lane like a drunk driver, or taking an intentionally inefficient route like a crooked taxi driver, or other pure-downside criminality.
- Johnny555 5y agoThe human behavior can also result in an accident when a car or pedestrian that expects the car to stop walks out in front of the car. In theory, the car should be paying attention in all directions at the same time and won't get into an accident, but car sensors are not infallible, especially when detecting pedestrians, moreso if they are partially obscured by foliage, newspaper stands, etc. Automated cars should stop 100% of the time at intersections where they are required to do so. When the stop signs are changed to "Self-driving yield" signs, then the cars can do a rolling stop. So my point wasn't that cars should act like drunk drivers, but that modeling human behaviors is not the right mindset for self-driving cars.