Autonomous Systems in Ethical Dilemmas : Attitudes towards Randomization

Ethically, it is debatable whether we may program autonomous systems (AS) to actively impose harm on some to avoid greater harm for others. Responses to surveys on ethical dilemmas in the programming of self-driving cars have shown that people favor to impose harm on some people to save others from suffering and are consequently willing to sacrifice smaller groups to save larger ones in unavoidable accident situations. This is, if people are forced to directly impose harm. Contrary to humans, however, AS feature a salient deontological alternative to minimizing harm by discriminatory factors: randomizing over dilemmatic outcomes. To be applicable in democracies, randomization must correspond to people’s moral intuitions. In three studies (N = 935), we present comprehensive empirical evidence that many people ̶ driven by moral considerations ̶ prefer to randomize between dilemmatic outcomes. Our findings hold in hypothetical and incentivized decision situations and are culturally and contextually robust