Seite - (000091) - in Autonomes Fahren - Technische, rechtliche und gesellschaftliche Aspekte

794.2 Scenarios that implicate ethics from the outside, you find yourself standing next to a switch: if you pull the switch, you can shunt the train to a right-hand set of tracks, thereby saving the five individuals on the track. Unfortunately, there is one person standing on the right-hand set of tracks who would then be killed. What is the right decision? The “correct” decision continues to be a subject of much debate in philosophy. Both answers seem reasonable and defensible. A consequentialist might justify switching the tracks to save five people, even at the regrettable expense of one. But a non-consequential- ist, someone who considers more than just the math or results, might object on the grounds that switching tracks constitutes an act of killing (the one person), while doing nothing is merely allowing someone to die (the five individuals); and that it is morally and legally worse to kill than to let die. Killing implies that you are directly responsible for a person’s death: had you not done what you did, the person would have lived. Letting die, however, involves much less re- sponsibility on your part, if any, since some causal process was already underway that was not initiated or otherwise controlled by you. The question of whether it is worse to kill than to let die is also subject to debate in philosophy. But let us bracket that for the moment, as a final answer is not necessary for our discussion, only that it is reasonable to believe that proposition. Adapting the trolley problem to the technology at hand, let us suppose that you are driving an autonomous car in manual mode; you are in control. Either intentionally or not – you could be homicidal or simply inattentive – you are about to run over and kill five pedestrians. Your car’s crash-avoidance system detects the possible accident and activates, forcibly taking control of the car from your hands. To avoid this disaster, it swerves in the only direction it can, let’s say to the right. But on the right is a single pedestrian who is unfortunately killed. Was this the right decision for your car to make? Again, a consequentialist would say yes: it is better that only one person dies than five. But a non-consequentialist might appeal to a moral distinction between killing and letting die, and this matters to OEMs for liabili- ty reasons. If the car does not wrestle control from the human driver, then it (and the OEM) would perhaps not be responsible for the deaths of the five pedestrians while you were driving the car; it is merely letting those victims die. But if the car does take control and make a decision that results in the death of a person, then it (and the OEM) becomes responsible for killing a person. As with the trolley problem, either choice seems defensible. Results do matter, so it is not ridiculous to think that the car should be programmed to act and save lives, even at the expense of a fewer number of lives. Yet it also seems reasonable to think that killing is worse than letting die, especially in the eyes of the law. What I want to highlight here is not so much the answer but the process of deliberation that points us toward one answer over another. To the extent that there could be many acceptable answers to any given ethical dilemma, how well one answer can be defended is crucial toward supporting that answer over others. Industry again would do well to set expectations by debating and explaining in advance its reasoning behind key algorithms that could result in life or death. Transparency, or showing one’s math, is an important part of doing ethics, not just the answer itself.