ICES: Small: Impartial decision making in distributed systems

利益冲突普遍存在于同伴之间的集体决策中。这种冲突可以被建模为代理对最终结果评估的两个不同组成部分之间的紧张关系：她的自私利益和她的无私意见。当这两种命令在逻辑上分开时，有时可以设计一个规则来汇总个人意见，同时防止代理人的信息对其自私的福利产生任何影响。这样的规则被称为公正的。这个项目关注两个简单的问题，一个是不可分割的私人物品（奖品）的奖励，另一个是同伴排名，当代理人自私地只关心赢得奖品或他们自己的排名时。有相当简单的投票规则，你的投票不能决定你赢得或失去奖品，但如果你不这样做，它会影响谁赢。也可以设计一个同伴排名系统，其中agent对自己的排名没有影响，但对同伴的排名有贡献。该项目致力于的一些公正决策规则类似于我们熟悉的投票规则，如限定多数或一条线上的中位数投票。这项研究的一项成果包括与任何其他投票规则一样普遍适用的实际机制。另一个方向是问，自私利益和无私意见的一般模式，哪些允许公正的规则，哪些不允许。该项目为关注信息披露的战略性扭曲的机制设计的广泛经济学研究做出了贡献。同行评估是商业伙伴关系、学术团体和几乎任何共享专业知识的团体的中心制度。评估专业工作需要的知识只能在这些专家中找到；它不能委托给一个公正但无知的外部观察者。学者们为诺贝尔奖挑选自己的一些人，电影专业人士为奥斯卡颁奖。同行评估容易受到利益冲突的影响：参与者可能会为了自己的私利而破坏自己无私的意见（决定正确决策的宝贵信息），而这对团队来说往往是无用的。一些腐败的例子是很容易避免的：体育比赛的裁判不应该给自己国家的运动员排名，配偶之间不能互相作证，等等。更微妙的问题，如互联网上的虚假（好评或批评）评论的倍增，只能通过巧妙设计评估过程来避免。该项目旨在构建少数不受利益冲突影响的决策机制，直接适用于商业伙伴之间的资金分配，同行之间的奖励，以及学生和校友对大学的排名。这一基础研究在互联网环境中具有更广泛的影响潜力，在互联网中，同行评估无处不在——从电子交易商（在EBay或亚马逊上）的声誉评分，到社交网络（Epinions, Twitter）的用户，以及搜索引擎对网页的评分和排名系统，完全依赖于页面之间相互链接的模式。

Conflicts of interest pervade collective decision-making among peers. Such conflicts can be modeled as the tension between two distinct components of an agent's evaluation of the final outcome: her selfish interest and her disinterested opinion. When those two orderings are logically separated, sometimes a rule can be designed that aggregates individual opinions while preventing an agent's message having any impact on her selfish welfare. Such a rule is called impartial. This project focuses on two simple problems, the award of an indivisible private good (a prize) and peer ranking, when agents care selfishly only about winning the prize, or their own ranking. There exist fairly simple voting rules where your vote cannot make you win or lose the prize, yet it influences who wins if you do not. It is also possible to design a peer ranking system where an agent has no influence on his own ranking, but contributes to that of his peers. Some of the impartial decision rules to which the project is devoted resemble familiar voting rules such as qualified majorities, or the median vote on a line. One output of the research consists of practical mechanisms as universally applicable as any other voting rule. Another direction is to ask which general patterns of selfish interest versus selfless opinions allow for impartial rules, and which do not. The project contributes to the broad stream of economics research on mechanism design concentrating on the strategic distortions of information revelation.Peer evaluation is a central institution in business partnerships, academic communities, and virtually any community sharing specialized knowledge. Evaluating specialized work requires knowledge that can only be found among those experts; it cannot be entrusted to an impartial but clueless outside observer. Academics pick some of their own for Nobel prizes, and movie professionals award the Oscars. Peer evaluations are vulnerable to conflicts of interest: a participant may be tempted to corrupt her disinterested opinion (a valuable piece of information to determine the correct decision) to serve her selfish interests, which by contrast are often useless to the group. Some examples of such corruption are easy to avoid: judges in sports competitions should not rank athletes of their own countries, spouses cannot testify for or against each other, etc. More subtle ones, such as the multiplication of fake (favorable or critical) reviews on the Internet can only be avoided by the clever design of the evaluation process. This project aims to construct a handful of decision mechanisms immune to conflicts of interest, applicable directly to the division of money between business partners, the award of a prize among peers, and the ranking of Universities by their own students and alumni. This foundational research has potential for much broader impact in the setting of the Internet, where peer evaluations are ubiquitous- from reputation scores for electronic traders (on EBay or Amazon), to the users of social networks (Epinions, Twitter), and the scoring and ranking systems of web pages by search engines, relying exclusively on the pattern of mutual links between pages.