Modelling of Agents' Behavior with Semi-collaborative Meta-agents

An autonomous agent may largely benefit from its ability to reconstruct another agent's reasoning principles from records of past events and general knowledge about the world. In our approach, the meta-agent maintains a first-order logic theory, called the community model, yielding predictions about other agents' decisions. In this contribution we introduce a query-based collective reasoning process where the semi-collaborative meta-agents use active learning technique to improve their models. We provide empirical results that demonstrate the viability of the concept and show the benefits of collective meta-reasoning.