This reminds me of the current research in ML of "simplifying" models for inference (pruning, distillation, etc.). Our model of reality is so imprecise, can we be confident that all the local details are insignificant to the global context?