Ethical AI: Conceivable Versions and Projected Issues

The article presents the results of an analysis of the various versions of ethical Artificial Intelligence (AI), presented in modern theoretical discussions and attempts at practical implementation, including in the context of the emergence of GPT. The following generalized models have been identified. First, the Artificial Moral Agent (AMA), which is usually considered in two forms: the real AMA, which has all or some moral qualities, and the ascriptive AMA, to which moral qualities are normatively attributed. Secondly, Ethically aligned AI, which is designed so that its functioning complies with moral norms and values accepted in the community. Thirdly, the moral aspects of creating an Artificial Ethical Assistant (AEA) are considered, which purports to be an all-knowing and impartial Ideal Observer in the field of ethics and is able to advise people to solve moral problems. This option for ethical AI comes in the form of general and personal AEAs. Special attention is paid in the article to the ethical advantages and disadvantages, as well as the ethical risks of each of the options considered. The risks of ethical paternalism are highlighted among the most significant. It imposes certain moral values on people, which creates conditions for the manipulation of people’s freedom, which can lead to immoral behavior. The problem is that the massive use of AI, including the uncontrolled spread of GPT, can strengthen these negative trends, reproducing existing ethical bias and generating new ones, which requires special attention in creating ethical AI.