An Approach to Computing Ethics

It might seem impossible to "compute" ideas that humans feel most passionately about and have such difficulty codifying: their ethical beliefs. We've been attempting to make ethics computable for three reasons. First, to avert possible harmful behavior from increasingly autonomous machines, we want to determine whether one can add an ethical dimension to them. Second, we want to advance the study of ethical theory by making it more precise. Finally, we want to solve a particular problem in ethical theory, to develop a decision procedure for an ethical theory that involves multiple, potentially competing, and duties. We've adopted the action-based approach to ethical theory, where the theory tells us how we should act in ethical dilemmas. This approach lends itself to machine implementation by giving the agent either a single principle or several principles to guide its actions, unlike other approaches that don't clearly specify the correct action in an ethical dilemma. The approach to computing ethics that we describe in this article is illustrated by MedEthEx, a system that uses machine learning to resolve a biomedical ethical dilemma. This, we believe, lends support for our approach to computing ethics