Learn more
Join Dexa

Causal Models Explained

Johann discusses a pivotal theorem regarding latent causal models, emphasizing how two models that produce the same dataset must share the same underlying causal structure. This insight highlights the connection between generative processes—like physical laws or human psychology—and neural implementations, suggesting that with sufficient data and a good optimizer, learned models can recover the true causal variables. However, he notes that assumptions about nature's causal modeling can introduce complexities that must be considered.