| Sumario: | The article focuses on the concept of "federation opacity" in federated learning (FL) networks, particularly in healthcare, where the full training dataset remains inaccessible due to distributed data ownership. It argues that federation opacity is not merely an epistemic issue but a jurisdictional challenge arising from divided institutional authority, which cannot be resolved solely through transparency or explainable AI. The author proposes establishing an independent Federated Learning Trust (FLT), a nonprofit fiduciary body with authority to represent patient-contributors, audit federations under controlled conditions, and certify ethically permissible FL use cases. This governance model aims to balance privacy preservation with accountability, address consent complexities, cybersecurity risks, regulatory disparities, and fairness concerns, thereby enhancing the legitimacy and ethical oversight of FL systems in medical contexts.
|
|---|