COAL: Coalition Organization with Agent Leadership

这个项目是由救灾小组的需求推动的，其中包括有效的废物清除、人道主义援助、电子团购、航空货物运输和电力传输规划。然而，该项目探索的策略适用于许多合作工作环境。研究任务是动态的，agent被提供不准确和不可靠的信息，agent信任被用来评估出价，领导和激励被用来提高联盟形成和运行效率的环境下的联盟形成问题。该项目是(1)研究导致理想群体行为的简单局部策略，(2)评估这些策略的可扩展性和适应性，(3)通过机制设计表明联盟框架可以在不限制自利主体使用的算法的情况下缓和偏见并增加全局利益，(4)表明领导可以提高效率。该项目正在开发算法和自适应策略，以帮助动态和未知环境中的人工（虚拟、机器）代理，但该研究对人类社会也有影响，包括应用于应急响应和其他合作工作，以及理论化个人和团体对晋升、招聘和大学录取等领域的多种形式的偏见的反应方式。

This project is motivated by the needs of disaster response teams, which include efficient waste removal, humanitarian aid, electronic group buying, air cargo transport, and power transmission planning. However, the strategies that the project explores apply to many cooperative work environments. The research focuses on the problem of forming coalitions in an environment in which tasks are dynamic, agents are presented with inaccurate and unreliable information, agent trust is used to evaluate offers, and a combination of leadership and incentives are used to improve the efficiency of coalition formation and operation. The project is (1) investigating simple local strategies that lead to desirable group behavior, (2) evaluating the scalability and adaptability of these strategies, (3) showing that through mechanism design the coalition framework can temper bias and increase the global good without restricting the algorithms used by self interested agents, and (4) showing that leadership increases efficiency.The project is developing of algorithms and adaptive strategies to aid artificial (virtual, machine) agents in dynamic and unknown environments, but the research has implications for human society, which include applications to emergency response and other cooperative work, and also for theorizing about the way individuals and groups respond to the many forms of bias in areas such as promotions, hiring, and college admissions.