Federated learning keeps data private but models with organizers.

Big tech often trains artificial intelligence using federated learning. In this setup, your data stays on your personal device, and only mathematical updates are sent out to train an AI. Companies claim this protects privacy. However, the final AI model still ends up owned entirely by whoever organized the training. The researchers examined how creative communities try to control their work across three layers: storage, circulation, and learning. Looking at artist trusts, cooperatives, and consent tools, they discovered a major gap: creators often control where their art is stored and shared, but lose all power once training begins. Artists might agree to let an AI learn, but they cannot control the resulting model. To fix this, the researchers propose four design principles so communities can govern the AI models themselves, not just their raw data.

Why it matters

Artists and creators whose work trains AI have no control over the resulting models. This affects who benefits from creative AI and how fairness is ensured.

Questions answered