Formal ethical reasoning and dilemma identification in a human-artificial agent system

Overview & Synopsis

This thesis comes within the scope of autonomous machines embedding ethical concepts in their reasoning. Overall, this work consists in proposing a formal reasoning allowing an artificial agent to judge decisions and justify its judgement, starting from philosophical concepts of ethics. This has been done through a situation model composed of states of the world, facts, decisions and events. Ethical frameworks of normative ethics have been formalized through rules to respect in order to judge a decision as "acceptable" or "unacceptable". Thanks to this formalism, a formal definition of a moral dilemma has been given in order to allow an artificial agent to identify a situation as such, based on dilemma properties coming from dilemma definitions of the literature formulated in natural language. This work of formalization has allowed the subjectivity sources as well as the factors of sensitivity of the model to be properly identified. A last chapter focuses on an experiment designed to study the influence an artificial agent embedding ethical reasoning may have on a human operator. The experiment reveals that most people are not influenced by the artificial agent’s suggestions. Nevertheless, deontological people tend to be more influenced than consequentialist people.