Computational Logic for Social Abilities in Multiagent Systems

在本研究中，我们研究了多智能体系统的计算逻辑。我们关注的是代理的社会能力，如协调、联盟、共识和谈判。然后我们用计算逻辑来表述这些能力。为此，我们首先将具有知识库的智能体视为具有答案集语义的逻辑程序。答案集表示由理性推理者在程序的基础上建立的信念集。一般来说，不同的智能体有不同的答案集。为了协调不同的智能体的信念集，我们引入了程序转换，在不同的智能体之间构造具有中间意义的新程序。我们还开发了几个程序转换：一个是通过组合不同的答案集来实现联盟，另一个是通过从不同的答案集中提取共同的信念来实现共识。这些程序转换产生一系列知识库，这些知识库反映了代理社区中协调、联合和一致的结果。此外，我们还提供了在谈判过程中评估提案和构建新的反提案的方法。为了构建新的建议，我们在人工智能中使用了溯因法和归纳法，在数据库的协作查询应答中使用了松弛技术。所提出的方法在逻辑规划的背景下集成了常识推理和多智能体的社会推理，并在现有答案集求解器的基础上实现。本研究的结果有助于用计算逻辑来理解多智能体的社会行为，也是人工智能和逻辑编程的重要应用。

In this research, we study computational logic for multiagent systems. We focus on social abilities of agents such as coordination, coalition, consensus, and negotiation. We then formulate those abilities in terms of computational logic. To this end, we first consider an agent who has a knowledge base as a logic program with the answer set semantics. Answer sets represent belief sets which are built by a rational reasoner on the basis of a program. Different agents have different answer sets in general. To coordinate different belief sets of agents, we introduce program transformations which construct new programs having intermediate meanings between different agents. We also develop several program transformations: one realizes coalition by combining different answer sets, and another one realizes consensus by extracting common beliefs from different answer sets. Those program transformations produce serial knowledge bases which reflect results of coordination, coalition, and consensus in an agent community. Moreover, we provide methods for evaluating proposals and constructing new counterproposals in the process of negotiation. Tb build new proposals, we use abduction and induction in artificial intelligence, and relaxation techniques in cooperative query answering for databases. The proposed methods integrate commonsense reasoning and social reasoning of multiple agents in the context of logic programming, and are realized on top of the existing answer set solvers. The results of this study contribute to a step on understanding social behavior of multiple agents by means of computational logic, and also serve as an important application of artificial intelligence and logic programming.