Interplay between Federated Learning and Explainable Artificial Intelligence: a Scoping Review

Critical research gap exists: Only one of the 37 studies reviewed actually measured how federated learning affects the quality of AI explanations, indicating that practitioners are implementing these technologies together without fully understanding how they interact.

Privacy-explainability tradeoff discovered: When interpretability metrics are combined across multiple federated learning nodes to create overall insights, individual patterns specific to certain nodes get lost or diluted—meaning the privacy benefits of federated learning may come at the cost of detailed, localized explanations.

FL+XAI can improve security: Several studies found that combining federated learning with explanation methods helps detect when individual nodes are malfunctioning or behaving maliciously, offering a practical security benefit for distributed AI systems.

Lack of standardization hampers progress: Most studies don't use established federated learning libraries or follow transparent reporting guidelines, making it difficult to compare results across research and slowing down the field's ability to develop best practices for responsible AI deployment.

More rigorous testing needed: The field requires substantially more quantitative research to determine exactly when and how federated learning and explainability techniques work well together, and under what specific conditions their combination provides genuine benefits for responsible AI.