Forming Effective Teams from Agents with Diverse Skill Sets

Many complex problems can be solved through an effective organization of human experts connected by a human computation network where each node contributes a unique skill set needed to enable a higher order problem solving capability of the group. Multi-Agent Systems (MAS) architecture is one of the examples in this group problem solving space, having demonstrated successful applications for well-defined domains such as military planning and trading auctions. Recent work in crowd sourcing applications based on enterprise social networks (e.g. People Cloud) showed that the group problem solving approach can be extended to enterprise and potentially Internet-wide scales. However, systems operating at such scales assume that candidate group participants make decisions about which groups to join based on limited connectivity and local information. This paper focuses on the relationship between network adaptation for candidate group participants and performance of problem solving groups. We demonstrate that systems that expect to form groups (e.g. crowd sourcing) by engaging participants equipped with diverse skill sets require more sophisticated network adaptation strategies than what can be expected based on previous research. To address this need, we evaluate a set of network adaptation algorithms for crowd sourcing and present some empirical results from a simulation based study.