Abstract

Could the concept of 'evil' apply to AI? Drawing on PF Strawson's framework of reactive attitudes, this paper argues that we can understand evil as involving agents who are neither fully inside nor fully outside our moral practices. It involves agents whose abilities and capacities are enough to make them morally responsible for their actions, but whose behaviour is far enough outside of the norms of our moral practices to be labelled 'evil'. Understood as such, the paper argues that, when it comes to AI and evil, we have more to fear from semi-autonomous, hybrid agents: agents who have enough of a capacity to operate within a minimal form of human practices of participation and responsibility, yet remain partially outside of it, capable of acting from outside the parameters of ordinary interactions in certain circumstances, perhaps sometimes for reasons unknown.