An Overview on the Explainability of Cyber-Physical Systems

The prevalence of automating complex physical processes through learning and interactions among heterogeneous components adds to the increasing complexity of Cyber-Physical Systems (CPS) and their behavior. Popular Explainable Artificial Intelligence (XAI) methodologies usually overlook the impact of physical and virtual context when explaining the outputs of decision-making software models, which are essential factors in explaining CPS' behavior to stakeholders. Hence in this article, we survey the most relevant XAI methods used for explaining CPS' behavior to identify their shortcomings and applicability in explaining the behavior of CPS. Our main findings are (i) several papers point out the importance of context for explaining CPS, but the explanation methods lack context-awareness; (ii) the explanation delivery mechanisms using low-level visualization tools make the explanations unintelligible. Finally (iii), added information about the system's working and context may increase the actionability of the explanations. Therefore, we propose to enrich the explanations further with contextual information using Semantic Technologies, user feedback, and enhanced explanation visualization techniques to improve their understandability. To that end, context-aware explanation and better explanation presentation based on knowledge graphs and counterfactual explanations might be a promising future research direction for explainable CPS.