New Challenges in Multi-Agent Intention Recognition: Extended Abstract

As multi-agent systems grow in popularity in research and in actual real-world applications, the role for multi-agent modeling techniques becomes increasingly critical. However,the transition from traditional intention-recognition (agent-modeling) settings, where a single agent is modeling another, to settings where an agent is modeling many other agents, is raising many difficult challenges to current techniques. In particular, three of these challenges are the monitoring selectivity problem, the plan-library incompleteness/incorrectness challenge, and the model challenge. Weare actively investigating ways to address these challenges, within the framework of socially-attentive monitoring, which focuses on the social structures and procedures within the monitored multi-agent system. Wehave provided a summaryof current preliminary results demonstrating the usefulness of this approach.