It is strange to see people assuming blame already given how vague the article is. Even if it turns out that the bike broke a rule, it is possible that the col

It is strange to see people assuming blame already given how vague the article is. Even if it turns out that the bike broke a rule, it is possible that the collision was easily avoidable by the car. I am suspicious of the car because self-driving cars do not have a great reputation for dealing well with unusual situations. And how unusual is a situation? Anything that is not a car, truck, or bus doing something in the road? Maybe self-driving cars are too dumb. How many specific types of "unusual" interactions with bikes, walkers, kids, dogs, etc., will each self-driving car AI model need to encounter to build up a large enough training set to learn to avoid a collisions for a specific type of interaction. Will there need to a training set of thousands of crashes or near misses of every precise combination of interactions that can occur with non-negligible global frequency in order for any self-driving car model to avoid collisions that a perfectly attentive, smart, and instant-to-react person would avoid? or do the models allow fuzzy predictions of what (for example) a bike might do even in cases where the model does not comprehend that a particular blob is a person on a bike.