Human Computation and Multiagent Systems: An Algorithmic Perspective

Human computation systems such as online task markets provide platforms that allow us to harness the abilities of many human problem solvers to accomplish a variety of tasks. We argue that such platforms share some of the features of multiagent systems, and ask whether algorithms originally designed for multiagent systems can be leveraged for coordinating the problem solving process among human computers. We study this question in the context of the graph coloring problem where each human controls the color of one vertex, and compare the performance of humans using adaptations of two well-known distributed constraint satisfaction algorithms to the performance of humans that were asked to simply color the graph. We find that people provided with algorithmic instructions achieve significantly better performance than people who are left to their own devices.