We are currently working to enrich our results along two fronts: adding to the complexity of the network model and increasing the sophistication of agent collaboration.
In real-world situations network graphs do not stay constant. Nodes fail temporarily or are taken down for maintenance. New nodes are added to the network periodically. And in radio-frequency systems, nodes often physically move so that links in the graph alter as transceiver proximities change. We believe that our mobile agent approach will be particularly well suited for dynamic networks.
Adding dynamic connectivity to the simulation requires that agents treat system data with greater subtlety. As nodes move and edges change, old information about network connectivity becomes incorrect. Similarly, we plan to also consider the effects of byzantine and pathological failures within the system. The possibility of these failure modes implies that unverified information from other agents is somewhat less reliable than information obtained directly from the environment. Next-generation agents will employ three strategies to gauge the validity of data: discarding old information, discounting rumors from potentially-unreliable sources, and making logical inferences about plausibility of data given other available topological information.
Further afield, we would like to explore the effects of agent specialization on overall system effectiveness. As the results of the superconscientious agent trials suggest, division of labor can lead to remarkable changes in overall performance. We are considering two potential strategies to support diversity: geographic focus and resource specialization. As an example of geographic focus, agents could specialize in maintaining accurate knowledge of only specific parts of the network. As an example of resource specialization, some agents might decide to only track information about underutilized paths while others maintain broader network maps. Both approaches could improve the performance of the system as a whole by allowing individual agents to work on different parts of the problem and cooperate to build a complete map.
