CAREER: A New Theory of Social Choice for More than Two Alternatives: Combining Economics, Statistics, and Computation

Rensselaer Polytechnic Institute, Troy NY

Investigators

Abstract

The proposal aims at generating computational mechanisms that will enable individuals to contribute towards making better collective decisions (e.g., news ranking) including crowdsourcing where aggregation of online noise answers can occur. The proposal brings together ideas from economics, statistics, and computation to expand the capabilities of social choice mechanisms to handle large numbers of alternative choices, to extract ground truth from aggregated preferences, and to address problems where individual agents might not be able to compare some alternatives. In contrast to classical social choice theory, which is limited to the selection between two alternatives, the project proposes a rigorous study of a model for computational choice that will be robust enough for discerning between thousands or even millions of alternatives. The proposal could have a profound impact in the way we build multi-agent systems, search engines and recommender systems. The proposed effort can serve as a catalyst in the growing area of computational social choice, including: (1) Rank aggregation has been used in many fields, involving some high impact applications like ranking of news. However, this problem is far from solved using traditional computational social choice methods because they either only work for two alternatives, require full rankings, does poorly in revealing the ground truth, or are hard to compute. The proposed research will develop new methodologies to overcome these deficiencies by designing objective, robust, and computable social choice mechanisms for rich preferences. (2) Crowdsourcing, whereby online workers' noisy answers are aggregated to produce a better overall answer to some question. This cannot be solved by existing computational social choice techniques as the online workers' answers are often partial orders, workers may manipulate the outcome by providing false answers, and the objective of aggregation is to reveal the true answer. The proposed research will directly tackle these challenges by designing new mechanisms, which are directly applicable to existing systems.