Bottom-Up Norm Adjustment in Open, Heterogeneous Agent Societies

Open, distributed multi-agent systems with heterogeneous agent societies face the challenge to maintain an appropriate performance level in the presence of bad-behaving agents. An approach to counter the exploiting behaviour of such agents is the use of trust mechanisms. Yet, it is still possible for agents with a more complex behaviour to manipulate and exploit the trust mechanism. In this paper, we show that the use of norms increases the performance, as well as the robustness of the system against colluding attacks of malevolent agents. We assume a fully self-organised agent society, i.e. no centralised control authority. However, this also demands a self-organised rule establishing and configuration process. This paper introduces a concept of norm adaption at runtime that adjusts dynamically to changing environmental conditions. We evaluate our approach within simulations of a trust-based desktop computing grid, where different classes of agent stereotype behaviour are considered.