This same concept is why you can’t make a 100% safe self driving car. Driving safety is a function of everyone on the road. You could drive as safely as possible, but you’re still at the mercy of everyone else’s decisions. Introducing a system that people aren’t familiar with will create a disruption, and disruptions cause accidents.

Again, we don’t need a 100% safe self driving car, we just need a self driving car that’s at least as safe as a human driver.

I disagree with the premise that humans are entirely predictable on the road, and I also disagree that self driving cars are less predictable. Computers are pretty much the very definition of predictable: they follow the rules and don’t ever make last minute decisions (unless their programming is faulty), and they can be trained to always err on the side of caution.