Self-Explainability in Self-Adaptive and Self-Organising Systems: Status and Research Directions

Most self-explaining AI systems remain theoretical rather than practical: The paper finds that "most SX approaches remain conceptual, with few practical implementations," meaning the technology to make complex AI systems explain themselves in real-world applications is still largely underdeveloped.

There is no agreed-upon standard for measuring whether systems actually explain themselves well: The review identifies "no formal or de facto standard for evaluating SX" as a "major research gap," making it difficult for developers and regulators to assess whether a self-explaining system is truly doing its job effectively.

Self-Explainability goes beyond current explainable AI efforts: While existing Explainable AI (XAI) tools help humans understand AI decisions after the fact, Self-Explainability represents "a more advanced goal" where systems proactively explain their own reasoning and behavior without external interpretation.

Complex AI systems are becoming harder to understand and trust: The paper emphasizes that "growing complexity of self-adaptive and self-organising systems, fuelled by advances in Artificial Intelligence, has made them increasingly difficult to understand and trust," highlighting why self-explanation capabilities are increasingly important for responsible AI deployment.

A new framework is needed to guide future development: The paper introduces "Levels of Self-Explainability" and a unified taxonomy to help researchers and practitioners position and develop self-explaining systems, establishing "a foundation and roadmap for advancing Self-Explainability in complex systems."