TLDR:
Researchers have developed a new method to improve the accuracy and transparency of automated anomaly detection systems deployed in critical infrastructure. AI ‘hallucinations’ can lead to catastrophic mistakes, but a new approach makes automated decisions more reliable.
Key Elements:
- AI systems designed to identify anomalies in critical infrastructure can make catastrophic mistakes due to “hallucinations”.
- Researchers have proposed a multi-stage method to improve AI reliability, transparency, and accountability in anomaly detection.
- The method includes using Empirical Cumulative Distribution-based Outlier Detection (ECOD) and Deep Support Vector Data Description (DeepSVDD) systems, combining them with eXplainable AI (XAI) tools, incorporating human oversight, and implementing a scoring system to measure AI explanation accuracy.
- This approach aims to provide plant operators with the ability to verify the correctness of AI recommendations and make informed decisions, ultimately improving the reliability and transparency of automated anomaly detection systems.
AI ‘hallucinations’ can have serious consequences in critical infrastructure, but with the proposed method, researchers are taking steps to make automated decisions more reliable and trustworthy.