X-SYS: A Reference Architecture for Interactive Explanation Systems

Explainability requires systems thinking, not just algorithms: The paper argues that deploying explainable AI effectively is fundamentally a systems engineering challenge, not merely a technical one. Simply having good explanation methods isn't enough—organizations need to build robust systems that maintain explanation quality across repeated user queries, changing models, and evolving data.

Four key system qualities matter for interactive explanations (STAR): The X-SYS architecture identifies scalability, traceability, responsiveness, and adaptability as critical attributes. This means explanation systems must handle growing demand, maintain clear records of how explanations were generated, respond quickly to user queries, and adjust as models and data change—requirements that go beyond what individual algorithms can provide.

Separating user interfaces from backend systems enables better governance: By decoupling explanation interfaces from the underlying computation through "contract-based service boundaries," organizations can update how explanations are presented to users without disrupting the technical infrastructure. This separation also makes it easier to implement governance constraints and maintain accountability.

Persistent record-keeping supports responsible AI deployment: The paper demonstrates that maintaining detailed state management enables traceability—meaning organizations can track and audit how specific explanations were generated. This capability is essential for responsible AI systems that need to demonstrate compliance and accountability to stakeholders.

A practical blueprint exists for building explanation systems: X-SYS provides a five-component reference architecture that guides practitioners in connecting user-facing explanation interfaces with necessary backend capabilities, offering a reusable template rather than requiring organizations to build explanation systems from scratch.