The Modern Trolley Problem - Responsible AI Principles

We live in an age where self-driving cars are no-longer just an interesting philosophical thought experiment, but a reality. When things go wrong, there is an immediate and heated debate around where the blame lies (Programmer, Manufacturer, Owner, Driver), but the more interesting discussion always starts further back than that, before the car is even built. How is the algorithm to be coded? What are the rules that govern the decision making processes of the car. Are those hard-coded or is a machine learning as it drives, taking into account the behaviour of other road users etc?