Machine learning-based intrusion detection systems (IDS) have shown strong performance in securing unmanned aerial vehicle (UAV) networks, but their black-box nature can make it hard for operators to know when to trust them. This paper addresses that problem by empirically studying how different explainable AI (XAI) formats influence operator behaviour.
The study contrasts a conversational XAI interface with a traditional dashboard-style interface, measuring both trust and reliance. The setting is UAV intrusion detection, a domain where incorrect trust decisions could have serious operational consequences.
Because the source abstract is truncated, specific results and effect sizes are not available in the provided text. The paper is listed on arXiv under reference 2608.10434v3, and the full findings would need to be read there.