Investigating Required Transparency Information and Display Features through an Empirical Study using a Dual-tasks HAT Simulation

Project started in 2023.

The ARC and GVSC have conducted a substantial number of HAT (human-autonomy teaming) studies in terms of fundamental human factors and applicable models and frameworks. The studies include, in general, strategical task allocations in human-autonomy teams, communication methods in manned-unmanned teaming including sharing situation awareness, measuring and modeling of trust and situation awareness in autonomous vehicles, etc. Based on the successful previous studies, successive studies are expected to expand and fulfill the HAT research. This may include: 1) a situation where an AI agent supports or cooperates with a human operator in a multi-tasking situation to improve the task performance in an adaptive manner. This would be a unique task situation different from a single task transfer situation such as in autonomous vehicles or multi-agent tasks such as manned-unmanned teaming, and 2) investigation of guidelines to design effective AI’s ‘transparency’ features, which are the basis of building ‘trust’ and ‘situation awareness’ to improve HAT performance.

The study includes two general objectives. First, the “HAT dual-task simulation” will be developed and validated as a test platform or a tool. An AI agent will be implemented in the dual-task simulation to support a human operator in completing the tasks, by learning, monitoring, and intervening human operator by sharing one of the tasks in a dynamic manner. The current study also will validate the utility of the HAT simulation, which is based on historical simulation validated in previous studies. The simulation will be used to assess AI transparency features in the study and is also expected to use for ARC/GVSC’s further studies through technology transfer. That is, it is expected that the simulation can be used to investigate HAT performance in various realistic complex, and stressful multi-tasks situations (e.g., warfighter), due to its flexibility in manipulating the information display contents and methods, AI behaviors, and task complexity.

The second objective is to investigate effective AI transparency features affecting HAT performance through an experimental study using the AI-cooperating dual-task simulation, in terms of human-AI interaction. Since effective communication methods including AI transparency and appropriate information visualization are critical in HAT performance by building trust and situation awareness, it is vital to understand what useful AI transparency information is and when/how to present them, and how a human operator can use the information in the dynamic multi-task situation. It is also expected that the study will reveal the structural relation of AI transparency with trust, situation awareness, and HAT performance, along with the operator’s behavior in strategic decision-making and task allocation.

#2.18