Action-Based Environment Modeling for Maintaining Trust

In open multiagent systems, agents need to select whom to trust. Traditionally, this selection is done based on the models that are built for the agents in the system. After each interaction with others, agents update their models of others based on the outcome. However, building and maintaining accurate models is difficult especially before many interactions take place. Contrary to traditional modeling approaches, we propose to model the environment in terms of agents' actions and their effects, rather than building individual models for each agent. Based on the effects of its actions, each agent can modify its behavior appropriately. We evaluate our proposed approach in comparison to a traditional approach in the Agent Reputation and Trust (ART) Testbed simulation environment. The simulations compare the two approaches in terms of the accuracy of models, the effectiveness in finding trustworthy agents as well as the effort needed to build accurate models.