Exploiting Domain Knowledge in Making Delegation Decisions

In multi-agent systems, agents often depend on others to act on their behalf. However, delegation decisions are complicated in normgoverned environments, where agents’ activities are regulated by policies. Especially when such policies are not public, learning these policies become critical to estimate the outcome of delegation decisions. In this paper, we propose the use of domain knowledge in aiding the learning of policies. Our approach combines ontological reasoning, machine learning and argumentation in a novel way for identifying, learning, and modeling policies. Using our approach, software agents can autonomously reason about the policies that others are operating with, and make informed decisions about to whom to delegate a task. In a set of experiments, we demonstrate the utility of this novel combination of techniques through empirical evaluation. Our evaluation shows that more accurate models of others’ policies can be developed more rapidly using various forms of domain knowledge.