Govern the Model, Not Only the Data: Storage, Circulation, and Learning in Creative AI
The episode discusses the paper "Govern the Model, Not Only the Data," which argues that models require new management as materials. Hosts detail proposals for dynamic governance, including a provenance ledger, conditional training loss functions to penalize rule violations during training, and explainable inference modules. The conclusion is that governance should be embedded into the computational pipeline for active resistance to bad patterns.