If the miles are not enough, they kill less until they first kill someone. At which point they kill several orders of magnitude more. # of events per something are not always a statistically valid way to measure things.

Well, the miles are enough...

Only if you aggregate them across companies and ignore the fact that the distribution those miles are not comparable to to distribution of human driven miles.

Do you really think there's too little data by now, or are you just making a technical point that isn't actually relevant to this specific situation?

As far as I know Uber has done pretty few miles, compared for example to Alphabet, so it is relevant.

What if self driving cars kill less people, but the type of people they kill are different from the type of people who die in human driving accidents?

For example, what if instead of 100 people per day dying from human-driven car accidents(where 95 of them are car drivers/passengers, and the other 5 are pedestrians/bicyclists), self-driving cars only kill 30 people per day, but 28 out of 30 are pedestrians/bicyclists?

From the comments above, this question is answered. Uber is around 25x worse than humans.

The entire potential of self-driving cars shouldn't be dismissed based on accidents like this - that would be an unfortunate side effect of these headlines. But I also reckon this coverage actually incentivizes better and safer programs. It's such early days, we simply need more data to form sound opinions. In the meantime, journalism is serving an important role here.

So far: more, by a factor of 20-25.

If human behavior on the street changes because of the fear of autonomous cars, life on the streets and the streets themselves will be much shittier.

I can see banning autonomous vehicles in city centers.