RI: Small: Computational Social Choice: For the People

社会选择理论领域涉及将个人的偏好或意见汇总到集体决策中；投票就是一个典型的例子。长期以来，经济学和数学一直在研究这个丰富的问题空间，并得出了一系列引人注目的理论结果。但实际应用很少。从人工智能的角度来看，计算社会选择的研究被许多研究人员视为构建多智能体系统基础的核心组成部分。本研究扩展了计算社会选择的理论工作，使人们能够共同决策。该项目旨在将人类决策置于计算社会选择的中心，同时利用多智能体系统中为投票开发的方法和技术。这项研究计划是由非营利网站RoboVote.org直接推动的，该网站使人们能够根据分析和经验证据实施任何看起来最好的投票方法。该项目将充分发挥RoboVote在教育、推广和社会影响方面的潜力，旨在改变人们在广泛应用中做出群体决策的方式。将要解决的具体挑战分为两个子集，对应于目前在RoboVote上模拟的两种根本不同的民意调查类型。1. 主观偏好：当每个选择的可取性是一个品味问题时，偏好是主观的。RoboVote通过假设选民对备选方案有潜在效用来汇总主观排名，并通过使用报告的排名作为这些效用的代理来选择一个最大化效用总和的结果。必须解决的一个直接差距是，这种方法不适用于结果是排名的情况。第二个影响深远的挑战是重新思考选民表达他们偏好的方式，以便在保持较低认知负担的同时，获得有关其实际效用函数的更多有用信息。最后，该项目包括在上述框架下的最优聚合方法的公理化性质的研究。2. 客观意见：在这种情况下，一些选择客观上比其他选择更好，但选民不知道这种客观比较。部署在RoboVote上的解决方案旨在处理最坏情况下的噪音。该研究旨在创建和测试改进naïve方法的算法，特别是通过建立固定参数可处理算法设计的协同进步。

The field of social choice theory deals with aggregating the preferences or opinions of individuals towards a collective decision; voting is a paradigmatic example. This rich space of problems has long been studied in economics and mathematics, leading to a slew of striking theoretical results. But real-world applications have been sparse. From the AI viewpoint, the study of computational social choice is seen by many researchers as a central component in the effort to build the foundations of multiagent systems. This research extends this theoretical work in computational social choice to enable people to make joint decisions. This project aims to put human decision making at the center of computational social choice, while leveraging the very approaches and techniques developed for voting in multiagent systems. The research plan is directly motivated by the not-for-profit website RoboVote.org, which enables people to implement whichever voting methods appear to be best based on analysis and empirical evidence. This project will realize the full potential of RoboVote for education, outreach, and societal impact with the aim of transforming the way people make group decisions in a wide range of applications.The specific challenges that will be tackled are divided into two subsets, corresponding to the two fundamentally different types of polls currently modeled on RoboVote. 1. Subjective preferences: Preferences are subjective when the desirability of each alternative is a matter of taste. RoboVote aggregates subjective rankings by assuming that voters have latent utilities for the alternatives, and selecting an outcome that maximizes the sum of utilities by using the reported rankings as a proxy for those utilities. An immediate gap that must be addressed is that the approach does not extend to the case where the outcome is a ranking. A second, far-reaching challenge is to rethink the way voters express their preferences, in order to obtain more useful information regarding their actual utility functions while keeping the cognitive burden low. Finally, the project includes a study of the axiomatic properties of optimal aggregation methods in the foregoing framework. 2. Objective opinions: In this scenario, some alternatives are objectively better than others, but this objective comparison is not known to voters. The solutions deployed on RoboVote aim to handle worst-case noise. The research aims to create and test algorithms that improve upon naïve approaches, especially by building on synergistic advances in the design of fixed-parameter tractable algorithms.