AF: Medium: Behavioral design for online environments

在线系统在越来越大的程度上是为大量用户设计的经济系统，每个用户都有自己的动机。当经济设计基于其目标代理人群体的准确模型时，它更有可能是有效的。然而，越来越多的文献表明，在各种环境下，无论是线上还是线下，人们的行为都不太像标准的经济代理人。这种行为差异可能会对这些环境的优化设计产生什么后果？这个项目将为行为机制设计理论奠定基础，目的是引入新的理论问题，并在实践中影响在线系统的设计。这项研究所利用的文献的多样性，以及可能产生的潜在影响，使其特别适合于将理论研究的意识带给原本在很大程度上与数学方法脱节的社区。PIs希望通过为很少或根本不接触数学的研究生开设新课程，以及为中学生提供实践学习经验来实现这一点。这项研究将基于实验工作中详细记录的三个行为现象家族来解决问题，并将调查它们如何影响一系列在线系统的设计，如社会计算和众包平台、匹配市场、排名和推荐系统。该项目将考虑为那些对其参与决策而不是他们对系统的输入(以参与为条件)制定战略的“简单”代理进行机制设计；为其决策可能遭受“选择过载”的代理设计定价和菜单设计；以及设计其输入来自行为偏向代理而不是来自平稳随机抽样过程的学习算法。

Online systems are, to a growing degree, economic systems designed for a large population of users, each with their own motivations. Economic design is more likely to be effective when it is based on accurate models of the agent population toward which it is targeted. A growing literature, however, suggests that people do not quite behave like standard economic agents in a variety of settings, both online and offline. What consequences might such differences in behavior have for the optimal design of these environments?This project will lay the groundwork for a theory of behavioral mechanism design, with the aim of both introducing new theoretical problems and impacting the design of online systems in practice. The diversity of the literature that this research draws upon, and can potentially impact, makes it particularly suitable for bringing awareness of theoretical research to communities that are otherwise largely dissociated from mathematical methods. The PIs hope to achieve this via new courses for graduate students with little or no exposure to mathematics, as well as hands-on learning experiences for middle-school students.This research will address questions based on three families of behavioral phenomena well-documented in experimental work, and will investigate how they impact the design of a range of online systems, such as social computing and crowdsourcing platforms, matching markets, and ranking and recommendation systems. The project will consider mechanism design for "simple" agents who strategize about their participation decision but not about their input (conditional on participating) to a system; pricing and menu design for agents whose decisions may suffer from "choice overload;" and designing learning algorithms whose input comes from behaviorally-biased agents rather than from stationary random sampling processes.