To what extent can the conceptual framework of applied ethics predict the unexplored problems caused by AI?

Abstract

Today, ethical issues are becoming a major risk in AI businesses. In order to deal with these risks, it is desirable to be able to predict possible problems of AI in advance. However, it is not easy in case of unprecedented AI products and services due to the absence of data. On the other hand, many theories and conceptual frameworks are being proposed in AI ethics. Therefore, we arrive at a simple question: "Can those conceptual frameworks of AI ethics help us predicting the unexplored ethical problems caused by AI?" In this paper, we select two conceptual frameworks in applied ethics discussed by Maeda and Mizukami: "normative theory of discrimination" and "fictitiousness of moral actors". We experimentally apply them to several AI use cases to predict their potential ethical risks, and discuss the effectiveness and difficulties of utilizing such conceptual frameworks of applied ethics to predict ethical risks in unprecedented AI use cases.