The universe is mostly computationally irreducible, which means it can only be explained by itself. ML models are good at explaining and predicting very limited parts of reality; language has been shown to be one of them, but not human cognition.