As this theme encourages a somewhat different frame for ML as a whole, its direct applications to alignment are not immediately apparent. One application would be as a language for assembling collections of models into self-regulating systems trained end-to-end to avoid imbalances (e.g. mode collapse in GAN training). Another application would be an approach to preserving internal features of an ecology (e.g. human values) during a shift to a different computational niche (e.g. new capabilities). In this, parametric ecologies tackle concerns related to unbounded optimization and objective robustness.