A Question of Benchmarking. Insights from Decentralized Networks for the Governance of Federated Machine Learning

‍The team from the Chair for Philosophy and Blochchain have published a new commentary in The American Journal of Bioethics, responding to concerns about "federation opacity" in federated machine learning (FL) for healthcare.

The commentary makes two central points. First, concerns about FL should be benchmarked against realistic alternatives: centralized data sharing and local-only training carry significant limitations and risks of their own. Second, FL is itself a form of decentralization, and lessons from decentralized networks may help address some of the governance concerns that have been raised. The authors point to Zero-Knowledge Proofs as a form of cryptographic attestation that could strengthen accountability and data quality while preserving FL's privacy benefits, and discuss how Decentralized Federated Learning combined with Distributed Ledger Technology could federate not only model training but also model aggregation.