Is there a standard for "real-world conditions"? I could easily see Waymo driving their cars up and down the highway near their headquarters for 50 million miles and calling that "real world".

Given the incredibly risky maneuvers that drivers take on 101 in the Bay Area, I'd say that Waymo cars are experiencing the "real world."

That depends on the nature of the collisions. Are those 4 fatal accidents vs 51 fender benders? The devil is in the details, not the executive summary.

For purposes of the thought experiment, assume the distribution of severity is roughly the same between the two.

Doesn't change my opinion much because I am not a domain expert in self-driving cars.

All this would do is make me question the nature of the data collected. For instance, is that 50 million miles of diverse roads and weather conditions or the same mile over and over again on a sunny day? Or were the human-driven cars equipped with contemporary collision-avoidance equipment or a motley representative assortment of cars from the past half-century?

What would sway me is auto insurers offering a discount for relying on self-driving car technology. I can wait patiently for that to happen.

Depends on the price. I suspect it will be prohibitively expensive for the long foreseeable future.