Exploring Trustful AI Augmentation with Virtual Evacuation Study

Ruohan Wang, Aidong Lu

Abstract: While Artificial Intelligence (AI) has quickly proliferated, the understanding of trust towards AI is still under-researched and it is a primary factor for the success of AI -augmentation. This project created a virtual evacuation study to explore people's trust of AI in dangerous situations. Our study adopts a virtual maze-like scenario with automatic AI -augmentation and three levels of fire simulations, where participants need to make the final decision. We have collected performance data and revised the metrics of KUSIV3 for participants to report their trust of AI in the post-questionnaires. Our results show that the levels of trust towards AI are generally positive, and can be affected by dangerous conditions. The participant-reported and mea-sured trust levels also demonstrate a loose but consistent correlation. We plan to expand this study to design effective AI -augmentation interfaces to support decision-making procedures.