The Benefits of Surprise in Dynamic Environments: From Theory to Practice

摘要

Artificial agents engaged in real world applications require accurate resource allocation strategies. For instance, open systems may require artificial agents with the capability to filter out all information which are irrelevant with respect to the actual intentions and goals. In this work we develop a model of surprise-driven belief update. We formally define a strategy for epistemic reasoning of a BDI-inspired agent, where surprise is the causal precursor of a belief update process. According to this strategy, an agent should update his beliefs only with inputs which are surprising and relevant with respect to his current intentions. We also compare in practice the performances of agents using a surprise-driven strategy of belief update and agents using traditional reasoning processes.