Impact of information in a simple multiagent collaborative task

In this paper we study the impact of information in a simple multiagent collaborative task - graph coloring. Inspired by the experimental study in [4], we study distributed algorithms for graph coloring where individual nodes are periodically given the opportunity to adjust their color in response to information regarding the color choice of neighboring nodes. When granted such an opportunity, each node chooses an admissible color (if available) that is more prevalent than its current color in its neighborhood. Focusing on the family of ring graphs, our findings demonstrate that there is an inherent trade-off between efficiency and convergence rates for such an algorithm. While increasing the information to the nodes improves the efficiency of the emergent coloring profile, it is also degrades the underlying convergence rates. The degradation in convergence rates provides one possible explanation for the findings in [4], which demonstrate that providing additional information to the nodes, which were controlled by human participants, can actually lead to losses in the efficiency of the emergent coloring profile. These losses could be a byproduct of the human participants not having the desire or time to stay engaged long enough in the revision process.