By the same token, " millions of miles in an extremely complex urban environment with zero life-threatening injuries or fatalities" is also not a useful data point. How many times in those million miles did the autopilot* kick out and default to the human driver? That's analogous to a driver falling asleep, yet they don't count that.
For what it's worth, I'm looking forward to a true automated system where I can get in the back seat and sleep on the way home, but we're not there. I think the tipping point will not be when the system is better than an average driver. The average is skewed by drunks, texters, nearly blind, and just plain bad drivers. It needs to be better than a sober, alert, skilled driver. and I wouldn't be surprised if it's better than halfway there.