AI and ethics

Abstract

Since it is of critical importance that autonomous systems, whether software or hardware, that interact with human beings (and perhaps other sentient beings as well) behave in an ethical manner, we consider six possible approaches to effecting this. We argue that the first five approaches are unsatisfactory and defend the last approach, the approach we have taken. It involves discovering ethically relevant features and corresponding prima facie duties present in the various possible actions such a system could take in particular domains and discovering decision principles for when there is a conflict between those duties. We, further, maintain that there are a number of additional benefits to taking this approach that involve becoming clearer about human ethics, in addition to the ethics to which autonomous systems should adhere, and the chance that it might well lead to providing inspiration for humans to behave more ethically.