ITR - (ECS+ASE) - (dmc+soc): Automated Mechanism Design

本ITR项目将对一种新的机制设计方法（MD）进行深入的理论和实验研究，即，设计机制的科学（例如，投票协议、拍卖、离婚程序或合作评级系统），以便尽管代理人以其自身利益行事，但仍达到期望的结果。 MD在传统上是一种手工奋进。在这项研究中提出的新方法是“自动化机制设计（AMD）”，其中的机制是计算创建的具体问题的实例在手。这种方法有几个潜在的优点：1）它可以产生比迄今为止已知的更好的机制; 2）它适用于迄今为止人工研究的问题类别; 3）它可以避免开创性的经济不可能性结果; 4）它将设计的负担从人类转移到机器。AMD的新应用，如投票，调解和慈善捐款的表达机制，将被研究。 其次，近似最优AMD，当设计者不完全知道代理类型的先验分布，将被研究。第三，可扩展的AMD优化算法，以及AMD软件，将被开发。 其结果具有广泛的影响。更有效地分配社会货物、服务、信息和任务是可能的。例如，电子商务和投票应用程序将允许更公平的结果，其中更精明或更知情的一方不再利用知识较少的一方进行交互。安全性也可以得到加强，因为没有任何一方会有动机或能够操纵系统对其有利。

This ITR project will conduct an in-depth theoretical and experimental study of a new approach to mechanism design (MD), i.e., the science of designing a mechanism (e.g., voting protocol, auction, divorce procedure, or collaborative rating system) so that a desirable outcome is reached despite the fact that agents act in their own self-interest. MD has traditionally been a manual endeavor. The new approach proposed in this research is "automated mechanism design (AMD)," where the mechanism is computationally created for the specific problem instance at hand. There are several potential advantages to this approach: 1) it can yield better mechanisms than the ones known to date; 2) it applies to problem classes beyond what has been studied manually to date; 3) it can circumvent seminal economic impossibility results; and 4) it shifts the burden of design from the human to the machine. New applications of AMD, such as voting, mediation, and expressive mechanisms for charity donations, will be studied. Second, approximately optimal AMD, when the designer does not completely know the prior distribution from which the agents types are drawn, will be studied. Third, scalable optimal algorithms for AMD, as well as an AMD software, will be developed. The results have broad impacts. More efficient allocation of social goods, services, information and tasks is possible. Ecommerce and voting applications, for example, would allow for more fair outcomes, where more savvy or better-informed parties can no longer exploit less knowledgeable parties to the interaction. Security can also be enhanced as no party will be motivated or able to manipulate the system to his or her advantage.