Moral disagreement and artificial intelligence

Abstract Artificially intelligent systems will be used to make increasingly important decisions about us. Many of these have made without universal agreement the relevant moral facts. For other kinds disagreement, it is at least usually obvious what kind solution called for. What makes disagreement especially challenging that there are three different ways handling it. Moral solutions apply a theory or related principles and largely ignore details disagreement. Compromise method finding compromise taking information as input. Epistemic an evidential rule treats evidence truth. Proposals for all can found in AI ethics value alignment literature, but little has been said justify choosing one over other. I argue choice best framed terms risk .