Component-Based Transparency to Comprehend Intelligent Agent Behaviour for Human-Autonomy Teaming

Abstract

The advancement of artificial situation awareness (ASA) technologies promotes intelligent agents to establish human-autonomy teaming (HAT) performing collaborative works with human as a teammate. However, such agent’s abilities trigger human’s over-reliance mental model. Lack of transparency as a mechanism to comprehend agent’s behaviour, has been pointed as one of the causal factors. Understanding the physical systems behaviours is critical as they affect the agent’s functionalities and, therefore, it becomes a part of human comprehension on agent. By far, the common observation on hardware components is to extract information related to their faulty. Yet, to support transparency, the observation is necessary to be extended to the faulty impacts on agent’s functionalities. Hence, this study aims to propose an observer which can reveal the dependency of agents’s functionalities on physical systems as a part of transparency information. The proposed observer exploits the capabilities of Bayesian Network to model such dependencies and, also, it adopts model-based reasoning concept to define the normal/abnormal components behaviours. The results of implementation case indicate that the proposed observer is applicable and significant to support overall human situation awareness (SA) in HAT.