MATCH: Engineering Transparent and Controllable Conversational XAI Systems through Composable Building Blocks

System transparency requires looking beyond individual AI models: The paper demonstrates that explaining single AI models isn't enough—entire interactive systems remain opaque. MATCH addresses this by mapping out all components (models, controls, and explanations) as interconnected building blocks, giving users a complete picture of how decisions are made end-to-end.

Conversational AI systems need access to model internals for genuine explainability: The research shows that conversational XAI approaches—where users ask questions to understand AI decisions—require direct access to how models work internally. Without this, even interactive explanations can be incomplete or misleading.

Standardized building blocks enable better communication between humans and AI systems: By representing systems as sequences of structural and explanatory components with clear connections (APIs), MATCH creates a common language that both people and automated agents can understand, reducing misunderstandings about how AI systems actually function.

Composable frameworks make AI systems more controllable: The building-block approach allows organizations to swap in different control mechanisms and explanation techniques (like LIME and SHAP) as needed, giving developers and users more agency over how AI systems behave and explain themselves rather than being locked into rigid architectures.