Monday, November 16, 2009
5130 Upson Hall
We present a simple game-theoretic model for the ESP game, a computer game devised by von Ahn and Dabbish to label images on the web. We model the current implementation of the game as a two-stage game of incomplete information. We characterize the equilibrium behavior under a model of preferences in which agents prefer to match earlier rather than later. Using an appropriate definition of stochastic dominance, we make minimal assumptions on an agent's type and impose minimal structure on the valuation function. We present an alternate model of preferences under which agents prefer to match on infrequent words rather than frequent words and show how the equilibrium behavior changes. Our results suggest the possibility of formal incentive design in achieving desirable system-wide outcomes in the area of "human computation". We will also discuss how this work fits into the broader agenda of "incentives for crowdsourcing" and provide directions for future work.
Joint work with David C. Parkes