SGER: Reasoning about Utilities and Incentives in Multi-Agent

由于设计多用户软件系统的挑战和无处不在的电子商务的前景，本研究项目的目标是发展关于多智能体环境中效用的推理的理论基础，并利用这些基础来构建有效地推理其他智能体的偏好和动机的方法。这种方法借鉴了人工智能和经济学的发展。具体地说，该项目定义了效用分布的概念，其作用类似于概率分布，以及效用网络的概念，其允许人们以贝叶斯网络允许人们推理概率的方式来推理效用。除此之外，该项目正在研究混合网络，这种网络允许同时对概率和效用进行推理。该项目的结果将提供一个坚实的理论框架，在其中推理多智能体系统中的信念和偏好，在软件设计和电子商务中具有潜在的应用。

Motivated by the challenges of designing multi-user software systems and by the prospect of ubiquitous electronic commerce, the goal of this research project is to develop the theoretical foundations for reasoning about utilities in a multi-agent setting, and use these foundations to construct methods with which to reason effectively about other agents' preferences and motivations. The approach draws on developments in AI as well as in economics. Specifically, the project defines a notion of utility distribution that plays a role analogous to that of a probability distribution, and the notion of a utility network that allows one to reason about utilities the way Bayesian networks allow one to reason about probabilities. Beyond this, the project is investigating hybrid networks, which allow simultaneous reasoning about probabilities and utilities. The results of this project will provide a firm theoretical framework in which to reason about beliefs and preferences in multi-agent systems, with potential applications in software design and electronic commerce.