A schema for specifying computational autonomy

Abstract

A key property associated with computational agency is autonomy, and it is broadly agreed that agents as autonomous entities (or autonomous software in general) have the capacity to become an enabling technology for a variety of complex applications in fields such as telecommunications, e/m-commerce, and pervasive computing. This raises the strong need for techniques that support developers of agent-oriented applications in specifying the kind and level of autonomy they want to ascribe to the individual agents. This paper describes a specification schema called RNS ("Roles, Norms, Sanctions") that has been developed in response to this need. The basic view underlying RNS is that agents act as owners of roles in order to attain their individual and joint goals. As a role owner an agent is exposed to certain norms (permissions, obligations and interdictions), and through behaving in conformity with or in deviation from norms an agent becomes exposed to certain sanctions (reward and punishment). RNS has several desirable features which together make it unique and distinct from other approaches to autonomy specification. In particular, unlike other approaches RNS is strongly expressive and makes it possible to specify autonomy at a very precise level. Moreover, RNS is domain- and application-independent, and is of particular value for agent-oriented requirements elicitation and analysis.