Composable Building Blocks for Controllable and Transparent Interactive AI Systems

System transparency requires more than explaining individual AI models: The paper demonstrates that making single AI models interpretable through explainable AI (XAI) techniques is insufficient—the overall architecture of interactive systems remains opaque. By representing systems as sequences of structural building blocks with explicit data flows and APIs, the authors show how to make the entire system architecture understandable, not just its individual components.

Visual building blocks can bridge human and machine understanding: The research proposes pairing structural components (AI models, control mechanisms) with visual explanations that serve as a "communication basis for both humans and automated agents like LLMs." This dual approach to interpretability helps align how humans and AI systems themselves understand what's happening in an interactive system.

Modular design enables better accountability and control: By breaking interactive AI systems into composable, documented building blocks rather than treating them as monolithic black boxes, organizations can more clearly identify where decisions are made, who controls what, and how different components interact—essential for responsible deployment and auditing of AI systems.

Explicit system flow documentation improves governance: The paper's emphasis on making data flows and component connections explicit through visual architecture diagrams creates an auditable record of how an AI system operates, which supports better oversight, debugging, and communication with stakeholders about system behavior and limitations.