Blockchain-based Trustworthy Federated Learning Architecture

Blockchain creates accountability in AI systems: The paper demonstrates that smart contracts can track data and model changes throughout the training process, creating a permanent record of how AI decisions were made—addressing a key challenge in responsible AI where organizations often cannot explain their models' decisions.

Fairness requires active intervention, not just good intentions: The researchers found that federated learning systems don't automatically treat all data fairly; they developed a "weighted fair data sampler algorithm" that improved both model accuracy and fairness, suggesting that responsible AI requires deliberate technical design choices rather than passive approaches.

Multi-stakeholder AI systems need new governance tools: Because federated learning involves many organizations training models together, traditional accountability methods fail; blockchain-based approaches offer a potential solution for tracking contributions and ensuring no single party unfairly influences outcomes.

Responsible AI improvements can enhance performance: Contrary to assumptions that fairness and accountability add burden, the evaluation showed that the fairness-focused approach "achieved better performance than the default federated learning setting in terms of the model's generalisation and accuracy," suggesting responsible AI practices and effectiveness aren't in conflict.

Real-world testing matters: The researchers validated their approach using actual COVID-19 X-ray detection rather than theoretical examples, demonstrating that trustworthy AI architecture can work in high-stakes healthcare applications where accountability and fairness are critical.