From Categorical Imperative to Algorithms — A Dialogue Between Kantian Ethics and Artificial Intelligence Ethics

Abstract

As the rapid development of AI (Artificial Intelligence), a series of profound ethical issues are illustrated to individuals. The most significant issues are the spread of algorithmic discrimination, and the encroachment of human dignity caused by instrumental rationality. In order to solve these issues, it is urgent for ethics to provide principled guidance. In this paper, I will display that Kantian Ethics, especially its core concept of categorical imperative, can provide irreplaceable formal foundation and critical lens for AI ethics. Through systematically explaining the key concepts of Kantian ethics, which are autonomy, self-legislation, and the principle that humanity is an end in itself. The current AI cannot be considered as a Kantian moral agent, because of its internal heteronomous nature and lack of free will. Based on this argument, this paper further discusses the programs of reconceptualized responsibility in the age of algorithms. Specifically, on the one hand, although AI is not a moral agent, its distributed feature may correspond to a distributed model of responsibility attribution. On the other hand, the formal features of algorithms provide a possibility for transforming the procedural cores of categorical imperative, such as the test of universalizability, into principles of ethical algorithm design. In conclusion, the dialogue between Kantian Ethics and AI is mutually enriching. Kantian ethics not only offers an essential ethical standard for evaluating and regulating AI systems, but also illustrates enduring vitality while facing contemporary technological challenges.