A dream for better group decision-making

My dream is to realize a new and ambitious paradigm by combining principles and techniques of economics, statistics, and AI: let the decision makers choose the axiomatic, statistical, and computational properties they deem important, and then an automated mechanism designer come up with an application-specific mechanism to explore the best tradeoffs.

Pursuing this dream, we have been working on three frameworks based on (1) statistical decision theory, and have analyzed axiomatic and computational properties of Bayesian estimators; (2) social choice theory to reverse-engineer existing manually-designed mechanisms; and (3) machine learning, where we want to efficiently learn mechanisms that satisfy a customizable set of axioms. We are exploring the idea of converting satisfaction of an axiom into multiple training data and then using the framework in (2) to learn.

In the second half of the talk I will discuss our recent progress on equilibrium analysis, PoA, and PoS of the common interest Bayesian voting game for more than three alternatives.