[kdudhshdjf]

Liability should always lie with the self driving car. In the real world weird shit happens on roads, people adapt and are expected to adapt. What good is a car that I can't trust to be at least as reliable as an overly defensive version of myself?

[shiftpgdn]

If human drivers are getting killed in the same spot the same is it still a vehicle saftey problem?

[keanzu]

How many Teslas on lane keep assist safely drove past that spot vs how many humans?

If every million human drivers have one accident and every 1000 Teslas have a similar accident then yes, it is a vehicle safety problem.

An auto driving system needs to have a higher margin of safety than a human driver not the same or worse.

[jakear]

A much higher percentage of autopilot cars had fatal collisions at that location than non-autopilot cars. That’s cause for concern. And for all we know the previous driver was distracted. Being not-worse than a distracted human isn’t very confidence-inspiring.

[mannykannot]

Yes, because responsibility is not a zero-sum game. If the system cannot cope well with sub-optimal roads, it should not be in use wherever one might encounter sub-optimal roads.

[kdudhshdjf]

Responsibility being a zero sum game is literally what a lot of these conversations and the emerging laws are about. If users are making a mistake often enough to be noticible then the system and its interface are encouraging that behavior. If the autopilot is being used in ill-advised situations then the question becomes why the system wasn't more comprehensible to the user, if it was then they wouldn't have made a bad judgement because they would have understood the limitations of the system. This is just another instance of blaming users for bad design.

[keanzu]

GM's Super Cruise does precisely this. It is only available on roads where GM allows it.