Hacker News Reader

Comment by sabertoothed | Hacker News Reader

I disagree with respect to one point: The death is insufficient evidence. What matters is relative safety (compared to non-self-driving cars). You cannot only count the hits but ignore misses. Or in this context: Avoided crashes need to be considered as well.

View on HN

Not really though. Human drivers both crash and avoid crashes all the time. In the instances where a fatal crash happens, there is a party at fault and they are held liable under the law. The same logic applies here. If the software caused the crash by being imperfect, then Tesla is liable. It is doubly bad if Tesla caused more people to rely on the software in a perilous manner by overstating the software's safety/feature-completeness. We could argue all day about an ideal future where no human drivers or accidents exist and how we get there, but today Tesla has blood on its hands.

After another read of your post: You seem to argue a totally different point. I did NOT say Tesla was not liable if their AP causes a crash in a fully self-driving mode. They are. My point in the original post was a different one.

It seems from your original comment and your reply that you're arguing that liability can be discharged if the AP system has fewer crashes per capital than humans do. You can't argue relative liability based on statistics. That's like me murdering someone and then arguing that I'm not liable because it just so happens that I didn't increase the per capita murder rate over last year. Now, whether Tesla has any claim that their liability can be discharged because the operator wasn't following legally binding instructions not to operate in that way is a different issue, and one I'm not privy to.

Yeah, I was not arguing about liability. That's the misunderstanding. I was not concerned with liability. And you're completely right wrt liability.

Yes, really, though.

If autopilot (or similar system) are safer than humans, then this should get reflected in the statistics. Just because people die in a crash with a self-driving car (even if that software was at fault), is NOT sufficient evidence. What matters is the net effect!

Reply on news.ycombinator.com