Game-Theorectic Foundation for Multi-Agent Systems

在多智能体系统中，功利推理一直是一个新兴的兴趣领域。该奖项支持的研究就在这一领域，尽管它也将利用更经典的符号推理概念。该项目将研究非合作博弈论工具和方法如何有效地应用于模拟基于人工计算机的代理之间的相互作用。由于这些代理必须被编程，研究必须提供明确的模型，说明代理如何推理到一个解决方案，这是传统博弈论中从未涉及的任务。博弈论必须用表征来补充，在表征中，可以定义代理的行为、参考和合理性，以及允许代理计算其决策问题的解决方案的机械程序。该项目旨在将智能体视为简单的推理机器，可以从一组公理和推理规则中推导出解决方案。由于智能体可能会反复交互，因此可能有机会从它们的环境中学习，我们需要为它们提供信念修正工具，并将这些工具嵌入到可行的计算程序中。本研究结果将为研究多智能体系统中如何实现合作与协调提供基础。

Utilitarian reasoning has been an area of emerging interest in multi-agent systems. The research supported by this award is in this area, though it will make use of more classical concepts of symbolic reasoning as well. This project will investigate how non-cooperative game-theoretic tools and methods can be usefully applied to model the interactions of artificial computer-based agents. Since such agents must be programmed, the research must provide explicit models of how agents reason to a solution, a task which has never been addressed in traditional game theory. Game theory must be supplemented with representations in which actions, references, and rationality for agents can be defined, as well as with mechanical procedures that allow agents to compute solutions to their decision problems. This project aims to view agents as simple inferential machines that can derive a solution from a set of axioms and inference rules. Since agents may repeatedly interact, and thus may have a chance to learn from their environments, we need to provide them with belief revision tools and embed such tools into viable computational procedures. The results of this research will form basis for studying how cooperation and coordination can be attained in multi-agent systems.