These sorts of ethical conundrums are interesting, especially as autonomous cars may be the first truly powerful expert system an average consumer will own.
That being said, this becomes less of a dilemma when considering the reality of driving. Drivers have auto insurance policies which cover their car, and it’s unlikely that even autonomous cars will remove the need for the owner of a vehicle to have such an insurance policy. As such, for insurance compliance, all vehicles will be programmed to perform actions in an emergency situation that will minimize harm to themselves. Ethical dilemma solved by the cold facts of financial risk management.
There’s also the element of the network which isn’t covered here, and is a problematic omission. If we have autonomous cars, they would be able to at some level communicate with each other. Security issues aside, network awareness is what gives immense benefits to driverless cars, as that level of communication allows traffic to be managed seamlessly, among other things. So in the masthead scenario from this article, the car wouldn’t choose between hitting another car and going off a cliff…it would tell the other car what’s happening, allowing all other traffic to move out of the way until the vehicle recovers or hits another object at a lower, non-fatal speed. The possibilities to reduce accidents are so much more vast that these bizarre ethical edge cases would be akin to putting shoulder belts on airplanes…unnecessary cost for questionable benefit.