On the Benefits of Argumentation-Derived Evidence in Learning Policies

Abstract

An important and non-trivial factor for effectively developing and resourcing plans in a collaborative context is an understanding of the policy and resource availability constraints under which others operate. We present an efficient approach for identifying, learning and modeling the policies of others during collaborative problem solving activities. The mechanisms presented in this paper will enable agents to build more effective argumentation strategies by keeping track of who might have, and be willing to provide the resources required for the enactment of a plan. We argue that agents can improve their argumentation strategies by building accurate models of others' policies regarding resource use, information provision, etc. In a set of experiments, we demonstrate the utility of this novel combination of techniques through empirical evaluation, in which we demonstrate that more accurate models of others' policies (or norms) can be developed more rapidly using various forms of evidence from argumentation-based dialogue.

Learning strategies for task delegation in norm-governed environments

“…To achieve this, we need to identify agents whose policy constraints will most likely enable the execution of the delegated task. In our framework, whenever there is a task to be delegated, policy predictions are generated alongside the confidence of those predictions from the policy models that have been learned over time [6]. Confidence values of favourable policy predictions are compared to determine which candidate to approach for a resource, or to delegate a task.…”