Such a car would not drive so close to a truck that it could not simply stop should the truck lose its cargo. It would not approach a tunnel at a rate of speed where it could not brake to avoid an obstruction. People do that because they ignore risk in favor of some perceived time advantage, but robots don't.
If we imagine the sets of circumstances that require a life and death decision to occur while driving safely, suddenly the number becomes vanishingly small. Things like "an earthquake occurs and the bridge you are on loses a section." or "a power transformer explodes and drops a power line across the road", or "a plane crashes into the ground in front of you." Under these extreme situations how the human in control would perform is practically a random number function. That a robot would minimize the loss of life, seems like the "correct" choice regardless.
People sometimes need the back sides of lane markers to be red to show there driving in the wrong direction at highway speeds. That's the kind of safety issues self driving cars need to address. Highly convoluted but what if a clown and a child jump off an overpass style crap completely misses the point. Self driving cars could chose to randomly kill someone randomly once per 1,000 years on the road and still end up safer than people.
As to cars the major upside is they are all one on one collisions. And insurance companies are really good at handling that already. There are even laws limiting how much pedestrians for example can sue for if they are struck in some areas.
Then again, given some of the stuff there's a market for, I guess there is the worry that 'driverless car aimed at person who wants to be in James Bond/Bullitt/whatever' might turn out to be a selling point for some manufacturers with questionable ethics.
Do people really want to lose their remaining time with their family just to spend it commuting and doing the same stuff they could be at doing at home in the car? How about losing their sleep in the morning because their car takes longer to get them to work than it used to?
Some people might want to roll straight out of bed and into their offices/workplace. A lot of others would rather minimise the time spent going to/leaving work and spend it in their own home doing stuff they want to be doing.
Particularly when this "getting ahead" behavior results in accidents that require road closures that back up highways for hours of bumper-to-bumper traffic.
IMHO, safety will win. Human life is much more valuable than wasted time. 50km/h is consider as safe speed for human driver in city. Computers are faster, so 60-70km/h will be safe for them.
A 30 kph collision with a pedestrian is almost always survivable. Anything faster isn't.
Robots can react faster, but cars still take time to stop. Most streets don't have more than a couple of feet of distance between moving traffic and parked cars. Reaction time is meaningless there.
Robots can use also other things to improve safety, which cannot be used by human driver because (s)he is busy with driving, e.g. "laser" pointers of car direction of drive at road, like "mark my lane" light used by bicyclist; or directed audio/video/vibration alarms to warn of potential crash; or safety hands at front of car (like human hands), which will be released by computer to pickup child right before hit, or to to take hit from larger distance, slowly, or to change direction of car, or remove something from the road.
Humans cannot do that but robots can.
As cars go autonomous, they'll communicate faster, drive more predictably, and generally be safer. At some point - 1%, 5%, 20%? - the network as a whole will benefit with less accidents and less of the resulting delays which means human drivers will be less likely to be late which takes away anxiety and potential distractions, making it even safer.
Therefore, I think these "Do I kill A or B?" scenarios become less frequent over time.
I just hope that "over time" part happens quickly... or the first "bad" choice by an autonomous car will delay the industry by years.
What I'm saying is that there are diminishing returns on eliminating all accidents, unless you are continuously elevating someone's life to an infinite value (your own, for example).
Is our expectation at some point that ZERO people of INFINITE value die shuttling themselves physically around this planet every a year? I'd be willing to accept a small chance of death (by contract even) if I always got to my appointments on time. Think of it as a fancy donor card.
But I think the ethical dilemma will end up solving itself over time, as well. When a car has to make an emergency decision, it will make a 5-10 millisecond call to its attorney, Dr. Watson, Esq., who will review all case history and settlements, run background checks on the potential victims, weigh all the factors against the current political climate, and direct the car to choose the accident with the lowest expected liability.
This does come with quite a few caveats in regards to road surveillance and infrastructure and legal turnabout, but it makes a lot more sense than trying to merge the two completely different styles of driving. Essentially that is the problem that will cause accidents; pretty much everyone who drive should know that traffic patterns that deviate from what is expected and currently existing on the road are common causes for accidents. Trying to merge two completely separate types of driving just seems like a fools errand, though I do understand that the alternative would probably be a bit too much for the current public and governments.
Ultimately though, I think that while the ethical concerns are real, they become greatly diminished along with the natural evolution of where autonomous cars take city infrastructure.
It seems that, as driverless cars become more common, you are more frequently on the other side of the equation.
Only for games with multiple iterations.
In addition, "Me first" strategies often create something fairly close to optimal for a lot of cases--we call them "greedy" algorithms for a reason.
You run into a cascade problem if your car decides to weigh the fates of other cars by trying to run some value comparison. Who sets the values? Can they be spoofed, and so on.
So first rule is that it cannot harm another vehicle to save itself. You aren't really part of the equation here, you cannot be. By the act of using the autonomous automobile you and it become one for sake of the how the rules play out.
However, if the person would only be injured, it should prefer that to someone else's death (as the person himself would be legally obligated to do.)
Of course, this is rather less of an issue that it seems, since the article itself points out that other, more normal safety features are being included that lower the risk from accidental collisions and the cars might not be going fast enough for this sort of issue to be a concern in most instances (read, those not involving a motorway or autobahn). And in those cases, well, anyone who might be in the path of such a car in such a situation might be best classed as 'too dumb to live', since high speed roads are not good places for pedestrian crossings or playing chicken.
But in the 0.00001% of cases which this issue might be a thing, then they should honestly just value the driver and passengers. Like every other tool and item in human society,for good or bad alike.