Abstract

One recent priority of the U.S. government is developing autonomous robotic systems. The U.S. Army has funded research to design a metric of evil to support military commanders with ethical decision-making and, in the future, allow robotic military systems to make autonomous ethical judgments. We use this particular project as a case study for efforts that seek to frame morality in quantitative terms. We report preliminary results from this research, describing the assumptions and limitations of a program that assesses the relative evil of two courses of action. We compare this program to other attempts to simulate ethical decision-making, assess possibilities for overcoming the trade-off between input simplification and output reliability, and discuss the responsibilities of users and designers in implementing such programs. We conclude by discussing the implications that this project highlights for the successes and challenges of developing automated mechanisms for ethical decision making.