[photon]

Yeah I'm kind of running on the assumption that the car would have all the relevant data (which is a big assumption I agree, but in principle there shouldn't be any reason it doesn't and it should in fact have far more information).

I know I've almost caused an accident because I braked to avoid hitting a dog, if I'd had eyes on the back of my head and had seen how close the car was behind me I might have opted to definitely injure or kill the dog instead of risking injuring the people in the cars behind me. Maybe, though I don't know if I could have. Similar thing happened with a mom and baby ducks, though not nearly as close, just got honked at.

So yeah I agree that there's a ton of situations, but I think when it comes down to your decision making all these things get abstracted out and processed quickly in our minds, and I don't see why there couldn't be a similar set of rules for the car. A decision tree is something a computer would be very good at.

On a basic level I could imagine a standard (maybe even legally mandated) set of rules that governs all those decisions (i.e. steer around the puddle to avoid splashing people, but don't steer but brake if there's a car in the neighbouring lane, don't brake and splash if there's a car too close behind, etc). It'd be a lot of work to establish, but it'd also be something that could improve over time with both simulations and real world data.

You could even mount equipment on millions of human driven cars to collect the data for a baseline.

But you raise an interesting question too, should I be able to tweak the priorities of my car based on my desires? That's interesting.