| Summary: | The article focuses on the ethical and epistemic challenges posed by "federation opacity" in federated learning (FL), a machine learning approach where data remain decentralized and inaccessible for direct inspection by stakeholders. It discusses how FL’s distributed architecture complicates accountability, data quality assessment, and vulnerability to attacks, despite its appeal for privacy in healthcare. The article then examines Swarm Learning (SL), a more decentralized alternative that removes the central aggregation server by using peer-to-peer coordination and blockchain technology, potentially enhancing auditability and reducing certain risks associated with FL. However, it emphasizes that both FL and SL require robust governance frameworks—covering data quality standards, responsibility, and fairness—to address ethical concerns, as technical decentralization alone does not resolve issues of transparency or accountability in medical machine learning.
|
|---|