Run-Time Support for Norm-Governed Systems Visara Urovi*, Stefano Bromuri*, Kostas Stathis*, Alexander Artikis** *Department of Computer Science, Royal Holloway, University of London, UK. **Institute of Informatics and Telecommunications, NCSR Demokritos, Greece. Computer Science Technical Report: CSD-TR-10-01 Royal Holloway, University of London, UK. Abstract We present a knowledge representation framework with an associated run-time support infrastructure that is able to com- pute, for the benefit of the members within a norm-governed multi-agent system, physically possible and/or permitted ac- tions current at each time, as well as sanctions that should be applied to violations of prohibitions. To offer the envi- sioned run-time support we study selected versions of the Event Calculus to support efficient temporal reasoning with- out compromising the expressive power required to specify the norms. Both the knowledge representation framework and its associated infrastructure are highly configurable in the sense that they can be appropriately distributed in order to support real-time responses to agent requests. To exemplify the ideas, we apply the infrastructure on a benchmark sce- nario for multi-agent systems. Through experimental eval- uation we also show how distributing our infrastructure can provide run-time support to large-scale multi-agent systems regulated by norms. Introduction An open multi-agent system (Pitt, Mamdani, and Charlton 2001), such as an electronic market, is often characterized as a computing system where software agents developed by different parties are deployed within an application domain to achieve specific objectives. An important characteristic of this class of applications is that the various parties develop- ing the agents may have competing goals in the application domain and, as a result, agent developers for a specific party will have every interest to hide their agent’s internal state from the rest of the agents in the system. Although open- ness of this kind may encourage many agents to participate in an application, interactions in the system must be reg- ulated so that to convince skeptical agents that the overall specification of the application domain is respected. Norm-governed multi-agent systems (Artikis, Sergot, and Pitt 2009) are open multi-agent systems that are regulated according to the normative relations that may exist between member agents, such as permission, obligation, and insti- tutional power (Jones and Sergot 1996), including sanc- tioning mechanisms dealing with violations of prohibitions and non-compliance. Although knowledge representation Copyright c  2010, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. frameworks for specifying such relations exist, these frame- works often focus on the expressive power of the formal- ism proposed and often abstract away from the computa- tional aspects and experimental evaluation. Works study- ing executable specifications exist but they normally do not provide experimental evaluations of multi-agent system de- ployment over distributed networks. As a result the com- putational behavior of many representation frameworks for norm-governed multi-agent systems are often studied in iso- lation, at times theoretically only, and in many occasions their experimental evaluation is left unexplored. The aim of this paper is to use a specific knowledge rep- resentation framework to develop an infrastructure for com- puting at run-time initially the possible actions, permissions, and sanctions, but eventually the obligations, and institu- tional powers for an application. The need for such infras- tructure is motivated by the observation that agents will not be expected to be capable of computing these normative re- lations on their own. Practical reasons for this include (a) computational constraints agents may have (e.g. due to lack of CPU cycles, memory, or battery), (b) incomplete knowl- edge agents may have about the application state (e.g. due to partial view of the environment) and (c) agent development focusing on action selection and execution (e.g. due to the expectation that it should be the application responsible for checking violations of norms and not the agents themselves). Our envisaged run-time infrastructure integrates selected versions of the Event Calculus for describing an open multi- agent system as two concurrent and interconnected compos- ite structures that evolve over time: one representing the physical environment of the open multi-agent system and the other representing the social environment required by the norm governed component. The focus of our knowl- edge representation framework and its associated run-time infrastructure is to provide real-time responses to agent re- quests. The novelty of our approach relies on the ability of our framework to distribute the environment of an applica- tion, in order to efficiently compute the social and physical states as they arise in norm-governed applications. The paper is organised as follows. First, we introduce a scenario of a norm-governed multi-agent system. We then use this scenario to describe our run-time infrastructure, the knowledge representation framework and extensions of this framework to support a social state with norms. An exper-