A Sinha and MP Wellman

18th International Conference on Autonomous Agents and Multiagent Systems (AAMAS) May 2019.


Research and design competitions aim to promote innovation or creative production, which are often best achieved through collaboration. The nature of a competition, however, typically necessitates sorting by individual performance. This presents tradeoffs for the competition designer, between incentivizing global performance and distinguishing individual capability. We model this situation in terms of an abstract collaboration game, where individual effort also benefits neighboring agents. We propose a scoring mechanism called LSWM that rewards agents based on localized social welfare. We show that LSWM promotes global performance, in that social optima are equilibria of the mechanism. Moreover, we establish conditions under which the mechanism leads to increased collaboration, and under which it ensures a formally defined distinguishability property. Through experiments, we evaluate the degree of distinguishability achieved whether or not the theoretical conditions identified hold.