Ethics in designing intelligent systems

Abstract

The idea of Hume’s guillotine contains the argument that one cannot derive values from facts. As intelligent systems operate with facts, Hume’s famous dilemma seems to contradict the very idea of being able to create ethical intelligent systems. In a closer look, ethics is a system of rules guiding actions. Actions always have factual or cognitive aspects, as well as evaluative or emotional aspects. Therefore, Hume’s juxtaposition of facts and norms is not well-founded. Instead of separating the facts and norms it should rather ask what kinds of facts are associated to what kinds of norms. Consequently, Hume’s guillotine sets no limits in processing ethical information, as one can combine facts and values while constructing information processes. However, intelligent machines cannot process ethical information independently of the human mind, as one cannot discuss ‘relevance’ in any formal language. Intelligent systems can be given descriptions of ethical situations and related norms, but machines cannot initiate ethical rules themselves.