CAREER: Structure and Incentives for Eliciting and Aggregating Information

People, organizations, and automated systems all need to base their decisions on information. But good information is often broadly distributed, expensive to obtain, and difficult to piece together. This project lays out the theoretical foundations for this problem and how to solve it. First, it will study how to evaluate quality, correctness, and economic value of information. This part also studies the design of loss functions in machine learning and includes writing an educational text. Second, the project will study how different pieces of information interact, both economically and algorithmically. Third, it will design systems that incentivize people to provide good information and aggregate it into useful outputs. These systems will be based on the above principles: valuing information economically and also processing it algorithmically. The high-level goal of the project is for these system designs to be useful for private and public organizations. Potential applications include data marketplaces, gathering public health information, and forecasting. The first part of the project studies the design of proper scoring rules and proper loss functions in decision making contexts. It includes writing a monograph on the field of information elicitation, given a loss function, what types of predictions and information can be provided by minimizing it, and vice versa. In a decision-making context, scoring rules and loss functions each define a value of information. The first part also investigates the relationship between value of information, good decision making, and accurate predictions, focusing on properties such as strong convexity of the utility and/or loss. The second part investigates the complexity of sequentially communicating and aggregating multiple pieces of information for decision making. This problem is related to classical communication complexity but explored in a fully Bayesian context with success measured by expected utility. Here, the project uses a classification of information structures into substitutes and complements based on value of information to a decisionmaker. The goal is to relate the economic structure of information, such as its substitutability, to the communication complexity of aggregating that information. Finally, the third part investigates the design of incentive-compatible mechanisms in which strategic agents provide private information and aggregate it into accurate predictions. This part's goal is to solve several open problems relating to information markets, including the computational complexity of finding a Nash equilibrium in such markets and the (im)possibility of designs with good properties when signals are complements. The project will consider how all of these principles can be used to design data marketplaces, prediction markets, public information dashboards, and other applications. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2542666 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT,01003031DB NSF RESEARCH & RELATED ACTIVIT,01002930DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Bo Waggoner | Institution: University of Colorado at Boulder, Boulder, CO | Award Amount: $432,921 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2542666 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2542666.html