@bruces that was an interesting article but I "can't endorse" the approach taken with self-driving.
Abusing math a bit, the denominator he described is not miles driven, it's who you're replacing. Humans vary widely in how awake, attentive, and sober they are. When a self-driving does better than them an aggregate it may just be doing better than the sleepy, inattentive, or impaired driver.
Thus it may be a bad recommendation for the better driver to adopt it.