Page - (000085) - in Autonomes Fahren - Technische, rechtliche und gesellschaftliche Aspekte

734.1 Why ethics matters or at least morally obligated, to absorb the brunt of any harm, at least when squared off against pedestrians, bicycles, and perhaps lighter vehicles. The ethical point here, however, is that no matter which strategy is adopted by an orig- inal equipment manufacturer (OEM), i. e., auto manufacturer, programming a car to choose a collision with any particular kind of object over another very much resembles a targeting algorithm [33]. Somewhat related to the military sense of selecting targets, crash-optimi- zation algorithms may involve the deliberate and systematic discrimination of, say, large vehicles or Volvos to collide into. The owners or operators of these targeted vehicles bear this burden through no fault of their own, other than perhaps that they care about safety or need an SUV to transport a large family. 4.1.3 Beyond harm The problem is starkly highlighted by the following scenario [15, 16, 17, 34]: Again, imagine that an autonomous car is facing an imminent crash, but it could select one of two targets in adjacent lanes to swerve into: either a motorcyclist who is wearing a helmet, or a motorcyclist who is not. It probably doesn’t matter much to the safety of the car itself or its occupants whether the motorcyclist is wearing a helmet; the impact of a helmet into a car window doesn’t introduce that much more risk that the autonomous car should want to avoid it over anything else. But it matters a lot to the motorcyclist whether s/he is wear- ing a helmet: the one without a helmet would probably not survive such a collision. There- fore, in this dreadful scenario, it seems reasonable to program a good autonomous car to swerve into the motorcyclist with the helmet. But how well is justice and public policy served by this crash-optimization design? Motorcyclists who wear helmets are essentially being penalized and discriminated against for their responsible decision to wear a helmet. This may encourage some motorcyclists to not wear helmets, in order to avoid targeting by autonomous cars. Likewise, in the previous scenario, sales may decline for automotive brands known for safety, such as Volvo and Mercedes Benz, insofar as customers want to avoid being the preferred targets of crash- optimization systems. Some readers may want to argue that the motorcyclist without a helmet ought to be targeted, for instance, because he has acted recklessly and therefore is more deserving of harm. Even if that’s the correct design, notice that we are again moving beyond harm in making crash-optimization decisions. We’re still talking about justice and other such ethical considerations, and that’s the point: it’s not just a numbers game. Programmers in such scenarios, as rare as they may be, would need to design cost- functions – algorithms that assign and calculate the expected costs of various possible options, selecting the one with the lowest costs – that potentially determine who gets to live and who gets to die. And this is fundamentally an ethics problem, one that de- mands much more care and transparency in reasoning than seems currently offered. Indeed, it is difficult to imagine a weightier and more profoundly serious decision