You will agree that “73 percent of Americans drivers report being afraid to ride in a self-driving vehicle” is not a safety assessment, but a safety perception assessment, which could be only weakly related to the former.

What makes difficult to assess the “real” safety of self-driving cars is that it’s a moving target: safety is getting improving little by little while of course human drivers remain basically the same. At some point the robot cars will no doubt surpass humans in safety, because for starters the AI drivers don’t get drunk every Friday or so.

Other aspect is that the accidents and fatalities of humans and robots are not going to be the same (that is, a robot which avoided one killed person could later cause the death of another one). But of course media puts the attention in the deaths by robot cars because this sells news, while car accidents with fatalities are no longer news, except for the most gruesome ones…