Seite - (000088) - in Autonomes Fahren - Technische, rechtliche und gesellschaftliche Aspekte

Why Ethics Matters for Autonomous Cars76 4.2.2 Self-sacrifice As we can see, real-world accidents can be very complicated. In philosophy and ethics, a familiar method is to simplify the issues through hypothetical scenarios, otherwise known as “thought-experiments.” This is similar to everyday science experiments in which re- searchers create unusual conditions to isolate and test desired variables, such as sending spiders into outer space to see how micro-gravity affects their ability to spin webs. It is not a good objection to those experiments to say that no spiders exist naturally in space; that misses the point of the experiment. Likewise, it is no objection to our hypothetical examples that they are outlandish and unlikely to happen in the real world, such as a car that can distinguish an eight-year old from an 80-year old (though with improving biometrics, facial recognition technologies, and linked databases, this doesn’t seem impossible). Our thought-experiments are still useful in drawing out certain ethical intuitions and principles we want to test. With that understanding, we can devise hypothetical scenarios to see that reasonable ethical principles can lead to controversial results in the context of autonomous driving. Digging into a standard philosophical toolbox for help with ethical dilemmas, one of the first principles we might reach for is consequentialism: that the right thing to do is what ever leads to the best results, especially in quantified terms [44]. As it applies here, consequen- tialism suggests that we should strive to minimize harm and maximize whatever it is that matters, such as, the number of happy lives. In this thought-experiment, your future autonomous car is driving you on a narrow road, alongside a cliff. No one and no technology could foresee that a school bus with 28 children would appear around the corner, partially in your lane [29, 36]. Your car calculates that crash is imminent; given the velocities and distance, there is no possible action that can avoid harming you. What should your robot car do? A good, standard-issue consequentialist would want to optimize results, that is, maxi- mize the number of happy lives and minimize harm. Assuming that all lives in this scenario are more or less equally happy – for instance, there’s no super-happy or super-depressed person, and no very important person who has unusual influence over the welfare of others – they would each count for about the same in our moral calculation. As you like, we may either ignore or account for the issue of whether there is extra value in the life of innocent child who has more years of happiness ahead of her than an average adult; that doesn’t matter much for this scenario. The robot car’s two main choices seem to be: (1) to slam on the brakes and crash into the bus, risking everyone’s lives, or (2) to drive off the cliff, sparing the lives of everyone on the bus. Performing a quick expected-utility calculation, if the odds of death to each person (including the adult bus driver) in the accident averaged more than one in 30, then colliding into the bus would yield the expected result of more than one death, up to all 30 persons. (Let’s say the actual odds are one in three, which gives an expected result of 10 deaths.) If driving off a cliff meant certain death, or the odds of one in one, then the expect- ed result of that would be exactly one death (your own) and no more. The right consequen-