I don't believe that partially automated driving, as shown by Autopilot is fit for human use. People will rely on it as if it's fully automated and more accidents like this will happen. We see this same problem with face recognition software that says an innocent man's face matches a criminal. The cops just accept what the computer says and arrest people because the computer says it's a match. We can't have the machi…
> I don't believe that partially automated driving, as shown by Autopilot is fit for human use. People will rely on it as if it's fully automated and more accidents like this will happen. Keep in mind 125 people are killed on the roads every single day in the US. If all vehicles magically switched to this system overnight and the system only killed 100 people per day, that would be a huge win. The fact some people wi…
The difference is that machines have no agency. Imagine a scenario where on roads with 100% human driving, the most of the people that are killed are PCP addled people running naked across freeways where they aren't expected to be, but with self driving you have 50% grandmas walking with babies in a stroller across crosswalks. We don't have to say that one life is more valuable than another to come to the conclusion that the latter is worse, all we have to do is acknowledge the pedestrian's portion of the responsibility for the outcome.
With that in mind, your last sentence is false, the aggregate death count is not all that matters, what matters is the situations in which they occur.
Now let's think about the link in this comment. We begin to form the rules of crossing a street around the behaviors of machines. Not unreasonable. But then it's just a small step from there to "you should've known how the machine was going to behave because it's behavior is simple and predictable." We wind up in a situation where we absolve the self driving car of all liability when an accident occurs because the only entity involved with agency was the pedestrian. We do this already with train tracks.
So with that in mind, the machine systems killing less people is not enough, it has to be so good that the only deaths are suicides, unbelievably bad behavior by those killed, or freak accidents so unlikely that they genuinely shock everyone who learns of them. That's the only way to transition from humans to software operating the machines and be able to justify the ruthlessness with which machines operate themselves.