Abstract

One way or another, users gain intrinsic and extrinsic rewards for particpating in online communities. Stackoverflow rewards users with reputation points and badges, Yahoo Q&A with points and a leaderboard, while Amazon Mechanical Turk, Epinions and Top Prospect pay users for contributions. But how should extrinsic rewards be designed, and what else can be done in promoting effective user behavior? Professor Parkes refers to this general problem as that of “computational environment design.” A first problem he discusses is that of routing a prediction task on a network, where users can contribute information and also find other well-informed users. He presents a family of local routing rules that are effective in isolating simple, truthful behavior in this setting. A second problem considers how to elicit information from users when there is no observable ground truth, for example in the context of a reputation system, or when voting on good answers in crowdsourcing systems. He also present a robust “Bayesian truth serum” that aligns incentives by eliciting both an information report and a prediction report from users. In closing, he will illustrate the role of simple “to do” prompts in coordinating a crowd in solving a complex task, in this case when planning custom trip itineraries, and reflect on what little we still know about user motivations.