Agent Based Models of Trust in Networks
Agent Based Models of Trust in Networks
In our BRIMS 2013 paper that won the best paper award, we examine how to study the impact of trust in networks. We show that an agent model that incorporates decisions on two types of beliefs: competence and willingness, can operationalize the impact of trust in achieving tasks in networks. The implementation of this framework and further documentation is available at rpitrust.github.io/agentsimulation/.
In our model, the agents have various characteristics that impact their behavior in the network. For example, competence is the ability to distinguish between signal and noise; willingness is the level of attention they can dedicate to a specific individual in the network, and selfishness is the degree to which agent’s actions is self-serving. These impact what and how frequently the agents are willing to share within the information.
Within this framework, agents observe the actions of other agents that they are connected to and update their beliefs regarding the others’ competence and willingness as a function of their own experience in the network. This means, agents beliefs are relevant to the given problem, network location, connectivity in the network and the information available to their peers.
We have also shown that our network can easily model different organizational structures in which rules overwrite trust decisions, such as need to know and need to report. These impact the willingness of the agents while in some cases overwriting their competence as the decisions are moved to individual nodes in the network.
We are currently expanding this framework and investigating the trade offs involved in networked problem solving.
March 2014