The impacts of multi-agent quantity, type and transparency on mental workload, situation awareness and human out-of-the-loop

Abstract: With the development of artificial intelligence, more autonomous systems, robots, unmanned aerial vehicles, and unmanned ground vehicles that can adapt and learn, independently determine goals, and allocate resources to perform specific tasks independently are changing human life. These devices are collectively referred to as autonomous systems. Although these autonomous systems can reduce labor demand, expand human capabilities, and improve human security, they will not be able to complete tasks independently in the foreseeable future and need a human to monitor and cooperate with them. Work organization between autonomous systems and humans is essential in the collaboration process. The ideal way is to maximize the agents' autonomy to expand human ability without causing humans mental overload. This study puts forward three objectives to explore a suitable cooperation mode between human and autonomous systems. First, to examine the impact of the multi-agent quantity on human mental workload, situation awareness (SA), and human out-of-the-loop (OOTL) degree. The second is to study the influence of multi-agent type on the above dependent variable. Thirdly, based on the completion of the first two objectives, a new variable, transparency, is introduced to explore ways to improve the cooperation between humans and autonomous systems. We designed a task scenario that simulated humans working with multiple autonomous agents. Two types of agents were adopted and redesigned: an intelligent assistant that helps operators assign unmanned vehicles, and a semi-autonomous dynamic positioning system for vessels. This research conducted two experiments. The first conducted a 2 (multi-agent quantity) × 2 (multi-agent type) within subjects experiment. The second investigated the effect of multi-agent type and transparency. The subjective mental workload, SA, OOTL degree, and performance were measured. According to the experimental results, the following conclusions can be drawn. The more the quantity of agents, the higher the cognitive mental workload, the lower the SA, and the more serious the OOTL degree. The more complex the types of autonomous systems are, the higher the mental workload. When humans interact with multiple autonomous systems, the heterogeneous agents reduce SA compared to homogeneous agents. The higher the transparency of the autonomous systems, the lower the mental workload, the higher the SA, the lower the degree of OOTL, and the better the experimental performance.The results of this study have significant theoretical and applied value. It can improve the existing theories of work organization between autonomous systems and humans, provide a new perspective from the perspective of mental workload, and improve the current research on multiple agent transparency. In terms of practical application, it can help enterprises and individuals decide which work organization mode to adopt in the human and multi-agent teams. It also has important reference value in the design of multiple autonomous systems. When only one aspect can be improved, priority can be given to the tradeoff on the quantity of agents. When it is inevitable to make humans interact with multiple agents, the transparency of the autonomous systems can be improved to increase the explainability and enhance the mental workload and task performance.

Keywords: multiple agents, autonomous system, mental workload, transparency

DOI: 10.54941/ahfe1003964

Cite this paper:

​