In human society, where different backgrounds, cultures and objectives coexist,udnorms help predict, control and coordinate individual behavior. Similarly, norms haveudbeen used in multi-agent systems to describe ideal behaviour for software agents. Inudspite of this, agents are still expected behave autonomously, due to the leeway allowedudby norms as soft constraints on individual behavior. Existing work dealing with normudidentification in multi-agent systems generally assumes that agents are fully aware ofudall norms, either at design time or as a result of communication with other agents.udSimilarly, work examining the impact of norms in agent decision-making proposesudstrategies that assume agents have complete knowledge of normative states.udThis thesis proposes that agents do not have complete knowledge about normativeudstates; consequently, it is the agents’ duty to identify norms. To this end, weudpropose an agent architecture and algorithms for identifying dynamic permission andudprohibition norms in open multi-agent systems. Using Event Calculus, we proposeuda formal representation of norms and a normative practical reasoning mechanism.udOther studies assume that the normative states that are neither identified prohibitedudnor obliged are permitted. Central to our proposal is that a normative state can beudunknown if it is not explicitly identified as prohibited, obliged or permitted. Thisudallows us to integrate permission norms into our proposed normative practical reasoningudmechanism. Thus, the contribution of this thesis is a set of techniques andudalgorithms that allow agents to join and function in a society regulated by (possiblyudunknown) norms, while minimizing behaviour that violates such norms.
展开▼