The first few months of a baby's life are spent learning about the world through observation, without any specific tasks or reinforcement, which forms the basis of common sense. Reproducing this kind of learning in machines is possible through self-supervised learning which involves observing the world and building world models.