Abstract

This paper examines the ethical agency of artificial general intelligence (AGI). In many studies, the ethical agency of AGI is divided into four categories: 1) Ethical-impact agents, 2) Implicit ethical agents, 3) Explicit ethical agents, and 4) Full ethical agents. This paper will deploy a critical-analytical method to examine the fourth category, namely full ethical agents in AGI. If AGI is possible, such intelligence would have many capabilities, and therefore, there would be many ethical concerns. This categorization of ethical agency has been repeatedly used by researchers, and there is a need for re-examination. The results of this study show that if AGI is possible, it should be placed in the category of a full ethical agent. However, such a full ethical agent has two characteristics: first, this agent can learn human ethical principles or infer other ethical principles from them; however, it should not be assumed that it considers human ethical principles to be its ethical principles; and second, this agent can form its own goals and ethical principles, and these principles may be different from or contrary to human ethical principles.