ICES: Small: Decision Making with Bounded Categorization

古典经济理论将人类的理性行为建模为期望效用的最优化。有限理性范式通过对代理确定最佳决策的能力施加计算限制，向更大的现实主义迈出了一步。行为和认知研究表明，人类也是有类别限制的，这意味着他们对可能提出的决策问题集使用有限的分类，只有少量的类别。本项目将著重于运用量化理论与资讯理论，为分类与决策之间的相互作用建立基础。本研究旨在了解分类对个人决策、多数投票团队决策和顺序决策的影响。在项目中发展的理论将包括对分类固定的情况的分析和分类的优化设计。除了研究分类的行为理由外，学习上的信息限制表明，将决策问题分类对工程设计有影响。这个项目有可能影响经济理论和对某些社会和组织现象的理解。具体来说，该项目提供了一种理解团队决策制定绩效的方法，其中包含了某些人为限制和团队成员之间不同偏好（例如，在类型I和类型II错误之间）的可能性。偏好的差异会导致对团队不和谐的量化惩罚，即使团队拥有做出正确决策的共同目标。当团队成员采用不同的分类时，团队多样性也有一个可量化的优势，即获得更好的绩效。这些新概念有助于形成团队的原则，以及如何优化数据收集政策，以实现公平决策的目标。

Classical economic theory models humans as acting rationally through the optimization of their expected utilities. The paradigm of bounded rationality takes a step toward greater realism by placing computational limitations on agents' abilities to determine optimal decisions. Behavioral and cognitive studies reveal that humans are also categorically bounded, meaning that they use a finite categorization of the set of decision problems that may be posed, with a small number of categories. This project will focus on the use of quantization theory and information theory to establish foundations for the interplay between categorization and decision making. The researchers aim to understand the impact of categorization on individual decision making, team decision making through majority voting, and sequential decision making. The theory developed in the project will include both analysis of situations in which the categorization is fixed and optimal design of categorizations. In addition to the behavioral justification for the study of categorization, informational limitations on learning suggest that categorization into classes of decision problems has ramifications for engineering design.This project has the potential to influence economic theory and the understanding of certain social and organizational phenomena. Specifically, the project offers a way to understand the decision making performance of teams, with the incorporation of certain human limitations and the potential for differing preferences (e.g., between Type I and Type II errors) among teammates. Differences in preferences lead to a quantifiable penalty of team discord, even when the team shares the common goal of making correct decisions. There is also a quantifiable advantage from team diversity in the sense of obtaining better performance when teammates apply different categorizations. These new concepts could contribute to principles for team formation and for how data gathering policies can be optimized with the goal of fair decision making.