This is a nice low-stakes example of how we can't expect automated driving to just be "like people but with better attention spans and reaction times and no risky stuff."
There will be things that are alien to human expectations, novel problems, and differently shaped risk profiles.