RI: Small: New Directions in Computational Social Choice and Mechanism Design

我们经常需要做出集体决策：谁将代表我们担任总统，我们今晚将在哪里共进晚餐，谁将获得奖项，等等。人工智能中的多智能体系统也面临着类似的问题。作出这种决定的最佳程序是什么？代理人可以投票决定结果，但具体程序应该是什么？社会选择理论和机制设计理论对此进行了研究，后者特别关注那些为了自身利益而采取战略行动的代理人。然而，越来越多的活动正在网上进行，集体决策也不例外。例如，我们对在线内容、产品和人员进行评级或投票。这些新应用的关键方面并不存在于传统的社会选择和机制设计模型中。例如，在一个基于互联网的机制中，谁可以进入，谁将参与？我们怎么知道单个代理没有多次参与？当投票事件的数量可能是压倒性的时，我们如何允许代理有意义地参与？该研究旨在扩展传统模型，以纳入这些方面，并开发新的算法和其他技术，以确保结果是有意义的，提高经济效率和人类福利。社会选择理论关注的是如何做出这样的集体决策。与此密切相关的是，机制设计理论关注的是如何为这些问题设计机制，即使代理人的行为具有战略性，也会产生良好的结果。近年来，在理解社会选择和机制设计的计算方面取得了重大进展。在拟议的研究中，PI和他的团队着手将这些技术应用于新的领域，例如互联网或新数据的可用性。例如，一个关键问题是参与代理人的身份。在一个基于互联网的机制中，谁可以进入，谁将参与？我们怎么知道单个代理没有多次参与？PI和他的团队旨在通过基于社交网络结构的技术以及代理人参与的努力来解决这些问题。另一个关键问题是代理商战略性地提供不准确数据的可能性。同样，对代理人这样做的努力成本的明确建模将发挥关键作用。

We often need to make collective decisions: who will represent us as president, where will we all go out for dinner tonight, who will receive the award, and so on. Similar problems are faced in multiagent systems in artificial intelligence. What are the best procedures for reaching such decisions? The agents could vote over the outcome, but what should the exact procedure be? This is studied in the theories of social choice and mechanism design, with the latter focusing particularly on agents that act strategically in their own self-interest. However, an ever-increasing amount of activity is moving online, and collective decision making is no exception. For example, we rate or vote on content, products, and people online. Key aspects of these novel applications are not present in the traditional models of social choice and mechanism design. For example, in an Internet-based mechanism, who gets to and who will participate? How can we know that a single agent is not participating multiple times? How can we allow agents to meaningfully participate when the number of voting events is potentially overwhelming? The proposed research aims to extend the traditional models to incorporate these aspects and to develop new algorithmic and other techniques to ensure outcomes are meaningful and increase economic efficiency and human welfare.In many domains, multiple self-interested agents need to make a collective decision. The theory of social choice concerns how such collective decisions should be made. Closely related, the theory of mechanism design concerns how to design mechanisms for such problems that result in good outcomes even when agents behave strategically. In recent years, major progress has been made on understanding the computational aspects of both social choice and mechanism design. In the proposed research, the PI and his team set out to adapt these techniques to novel domains such as those enabled by the Internet or the availability of new data. For example, one key issue is the identity of the participating agents. In an Internet-based mechanism, who gets to and who will participate? How can we know that a single agent is not participating multiple times? The PI and his team aim to address such issues with techniques based on social network structure, as well as on the effort that agents expend participating. Another key issue is the possibility of agents strategically providing inaccurate data. Again, explicit modeling of the agents' effort costs in doing so will play a key role.