Yep. If autonomous vehicles become widespread before they are really ready and the news starts filling up with accidents a human would have avoided, it will cause a massive political backlash. The actual statistics will be irrelevant to most people.

Very frustrating how people fail to see the full context. However, it is also useful in this case. The engineering, marketing and CEOs of companies doing autonomous vehicles are aware of the huge negative media potential a single accident has. Thus Management and marketing have an incentive to give engineering the time and resources to do things properly.

That is genuinely insightful, and is most certainly a political problem to address.

> That is genuinely insightful, and is most certainly a political problem to address.

Why is that a political problem?

Because if your family member is violently killed by a robot car doing something that seems inexplicable to the human mind, and with a fact pattern that doesn't match up with normal human error accidents you're used to, you're not likely to be consoled with a reminder that four people you've never met supposedly are still alive because they weren't killed in some place you've never been by a human driver that never took the wheel.

People accept that humans are imperfect and cause accidents. Accepting that computers will make mistakes and kill people is another story. Especially if the owner of the car is a 1 percenter or a mega corporation and it kills an innocent bystander.

Also, people aren't that accepting of human imperfection and expect to see people have licences taken away and jailed where they've been deemed to be behaving recklessly (even if they've racked up huge numbers of miles with no previous incidents)

Human drivers kill more than 30000 people a year in the US alone. It's not about saving "even one life", self-driving cars would have to be pretty terrible to reach those numbers.

Yeah. The question for me is: is that "1% better driver" compared to the average? In that case, it's still going to drive worse than 49% of the driving population. On the other hand, if it's 1% better than the 95th percentile, then yeah, that makes sense as a target to me.

Slightly better than average doesn't sound that great to me.

So every decision is a trolley-problem style tradeoff. Remember that your null action in this case is "save four lives but kill five people who otherwise would have lived." Where's your cutoff point?

I don't think this is a form of trolley problem. If the decision were going to be made unilaterally that either everyone or no one would use net-positive self-driving cars, then it would be a trolley problem. Actually, this decision is made on a driver-by-driver basis by the driver themselves. If the driver is educated about when it's riskier and when it's safer than their normal driving, it's easy enough to punt the decision to them. Clearly that's not happening though https://youtu.be/qnZHRupjl5E