Agent System Communication Theory And Organizational Structure

Agents are active computational entities siuated in complex environmnts, where theyutilize available resources to undertake assigned tasks.In simple cases, individual agents are capable enough to implement very meaningfulapplications. However, to fuly exploit agents' potentiality (especially in a distributed,autonomous, and open environment), we need a group of agents which could coordinate theirbehaviors and be functioning together.Almost all the coordination and cooperation activities amongst agents involve, in one wayor another the agents' ability to communicate in speech-act languages. Speech--actcommunication is playing the role amongst agents what human speech is playing amongst agroup of people.This thesis is a study of multi-agent system from the very perspective of speech-actcommunication. As the communicative aspect of agents is the focus of our work, the otheraspects of agents are necessarily treated as of lesser importance. So, the mulh-agent systemsunder study by us are also called agent communication systems.As a broad outline, we can say our work on agent communication systems is carried on inthe following five respects.l. We introduce a semantics framework based on conversational observer's mental statesfor understanding and designing speech-act languages, and have used the framework togive an operational semantics for a light-weigh exemplar speech-act language, SAIL.2. We make it clear that, just as there are some inherent structural principle underlyinghuman conversation, apent communication should as well abide by some structurizationrules. Among these rules are, the turns-taking system of a conversation, the nestedoccurrences of topics, and the protocols regulating communicative behaviors within atopic. All these and some other things around conversation collectively comprise what wecalled ARC Model for agent communication.3. Agent systems (esp. large-scale multi-agent systems) usually have very complexorganhaonal relationships within. To structure these relationships, we propose a flexibleand scalable construct, scope, as the basic organizational unit for agent communicationsy5tems. ScoPe can not only orgtri a collechon of roles around an agenF it also.'oWze a gr'OuP of agents within an orgtrindon, which may fimchon just lthe an agentby pwticipating in conversations. Scopes. oFgedZations, agents and roles collectively. fcomPnse what we called OrgaItizational scope Theory4. TO suPport the developmen of agent communication systCms based on ARC Model andOrpboZational ScoPe TheOry we need a modeling lmpge tailored specifically fOr thespecification of agent coInInbocation sySt6m. As a result, in this thesis, we have designedthe ASML langUage to fulfill this requirement.5. The development of agent conununication system faces many methodological challenges,especially in the aSPect of dynndc system conUnwhcative behavior. TO alleviate thesitUation, we develOP a suite of methods for analyZing, designing, and veribog.connnumcative behavior within an agent coInmtalcation system. These methOds helP ustO extrat conversation strUbo from examPle conversation sceIhao, helP us tO checkthe seInanic validity of the connntalcative act (i.e. Speech-act) occuring in the scenario,and helP us tO verify tha different imP1ementaion strategies adOPted by different agentsindeed confOrm to the protocols they need to comPly with.As a whole, our work has followed a softwar engineering aPProach tO investigating agentsystems. TherefOre, as a cooclusion of our work, we finally aPPly our langUaes, Inodls andmethOds to the analysis and design of a realishc distributed aPPlication system. The outCom ofthe analysis nd design is evaluate4 and our methods are also comPared to many other methodscoInin hom both hational software engineering field and from the emergent agentbasedsoftwar enginering field, illuswi their relative meritS and drawbacks, and pointin thedirechons for the...