I guess It's about liability and reproducibility.

When someone does harm with his/her car, holds the civil (and sometimes penal) responsability of his own acts and usually cannot do harm more than once in a very small timeframe.

In the case of autonomous IA, the company making/coding the vehicle software will be liable, and the problem is very likely to show again in a short time period.

That makes this kind of companies technically on the verge of bankrupcy because they are a good target for class-action lawsuits.

Same reason as we're afraid of flying, I guess.

Part of the reason is probably because we're still given captchas asking us to identify lights, buses and zebra crossings.

It could also be that it's killing people from a specific bug. A car might be blind to say, someone wearing a black and white shirt, or maybe a green car and 95% ory deaths could be from that.

You cannot tell whether a self-driving car kills less people than human drivers because to get the same statistical significance we have on human driving you'd need 84 billion hours of data from real autonomous driving within a year, and unbiased across all of the roads of the US.

Americans drive ~70 billion hours each year. You don't need autonomous cars to drive that anywhere near that volume to determine whether it is safer than humans. To prove with 95% probability (arbitrary but widely agreed upon standard) that autonomous cars are safer than humans could be done with much less driving hours. This is how clinical trials work. We don't need to give a new drug to every patient with heart disease to conclusively determine it's effective.

> to get the same statistical significance we have on human driving

I don't know what this phrase means.

Source?