Towards Social Comparison for Failure Detection: Extended Abstract

Social comparison, the process in which individuals compare their behavior and beliefs to those of other agents, is an important process in human societies. Our aim is to utilize theories of this process for synthetic agents, for the purposes of enabling social skills, teamcoordination, and greater individual agent performance. Our current focus is on individual failure detection and recovery in multi-agent settings. We present a novel approach, SOCFAD, inspired by Social Comparison Theory from social psychology. SOCFAD includes the following key novel concepts: (a) utilizing other agents in the environment as information sources for failure detection, and (b) a detection and recovery method for previously undetectable failures using abductive inference based on other agents’ beliefs1.