Review of eXplainable artificial intelligence for cybersecurity systems
Stéphane Reynaud, Ana Roxin
Abstract
Abstract This article reviews approaches based on artificial intelligence (AI), which contributes to the security of cyber environments. We examine existing techniques using several indicators: explainability, performance and robustness. These indicators have been chosen based on their importance for user acceptance and interpretability of the approach. Indeed, the AI field is vast and is divided into several sub-domains. The two most well-known sub-domains are symbolic AI (representation of knowledge, rules and operations based on symbols) and numeric AI (calculations and algorithms using numeric information, focusing on the result, not the representation of knowledge). While most approaches investigated come from numeric AI, we conclude on the need for hybrid AI systems, combining the advantages of both AI sub-fields while maximising the protection provided against cyberattacks.