Can artificial intelligence be a Kantian moral agent? On moral autonomy of AI system

Abstract

Machine Ethics emphasises the importance of collaboration between engineers, philosophers and psychologists to develop artificial intelligence-endowed systems and other ‘smart’ machines as artificial moral agents (AMA). They point out that there are top-down and bottom-up approaches for programming values into artificial autonomous systems. A number of thinkers argue that formalisation of the Kantian categorical imperatives is feasible, and hence, it is possible for smart machines to become Kantian moral agents, through the top-down approach of programming the Kantian categorical imperatives as algorithms into the AI systems. This paper examines some of the arguments put forth by the defendants of the possibility of Kantian AMAs such as Powers to point out that what these thinkers ignore is that in the Kantian schema, a moral agent is a rational being who is capable of ‘universalising’ as the law, the subjective maxims of her actions. Can the AMA be rational in this Kantian sense? The paper argues that though Kantian deontology may be attractive a theory for designing AMAs, the artificial agents cannot be Kantian moral agents in the real sense of the term.