Incentive Design in Distributed-Work Systems: Image Labeling Games and Question and Answer Forums

Опубликовано 17 августа 2016, 21:22

In recent years, there has been a great deal of progress in "Games with a Purpose," interactive games that users play because they are fun, with the added benefit that they are doing useful work in the process. The ESP game, developed by von Ahn and Dabbish, is an example of such a game devised to label images on the web. Our interest is in understanding the effect that qualitative changes in preference structure can have on the generation of useful work. To this end, we model the ESP game as a two-stage game of incomplete information and characterize the equilibrium behavior of the model. Under a "match-early" preference model, capturing current game dynamics, we show that focusing on low effort, higher frequency words is a Bayesian-Nash equilibrium. We establish more desirable outcomes of focusing on high effort, lower frequency words under an alternate model of "rare-words" preferences, consistent with a game structure in which players receive more points for infrequent words. The technical results require the use of a stochastic dominance argument and impose minimal requirements on word distributions. More generally, I am interested in the possibility of formal incentive design for problems of human computation and peer production systems. As a second example, I will discuss results on user behavior in question-and-answer forums, where the challenge that we consider is to promote early reporting of information by participants. We offer alternatives to the "best answer" scoring rule that promotes early participation, even in the presence of a complementary information space, where the value of each successive item provided to the asker is greater than the previous. Joint work with Yiling Chen and David Parkes.