I think you're making OPs point for them. Real drivers also crash into things. Self-driving cars don't have to be perfect all of the time. Inebriated and distracted drivers already cause disproportionate problems: self-driving, even one that makes a mistake once in a blue moon, is preferable than some dude on their phone who hasn't noticed the traffic in front of them has slowed down. (And 120 km/h sounds like highway cruising: precisely the sort of place where self-driving cars have a relatively easy time and "better than humans" is not very far off from where we are today.)
The dude on their phone is not comparable to the an AV carrying a family of 5 unpredictably veering into a concrete divider. One is negligent, the other has done nothing and could have done nothing, they only trusted the system. How do you reconcile it as merely a question of numbers then? So you trade 100 human-error accidents for 5 "blue moon" AV accidents, and this is good because it's statistically much safer. But that's also 5 accidents that wouldn't have happened to safe, diligent drivers.
