The Ethics of Deception: Why AI must study selfish behaviour

The study of ethics is concerned with defining the rules governing social interaction. However, the first models of social interaction developed within Artificial Intelligence (AI) research were based on earlier theories concerning the structure and dynamics of conversation. This is not a unusual; conversation has long been held to be a paradigm case of social interaction and AI borrowed theories from both the philosophy of language and socio-linguistics which merged social and linguistic behaviour. Central to such theories is a concept of cooperation. This paper will argue that such a cooperativistic view cannot account, specifically, for certain aspects of conversation and, more generally, essential features of agent hood such as autonomy, conflict and choice. Instead, an account of social interaction based on rational self-interest is described which has been implemented in a AI program to reason about natural language. Such an account is argued to be a more promising foundation for an ethics of intelligent agents.