Causal representation learning by far (e.g. learning the things that can influence each other, given an image)! This is the topic that Yoshua Bengio gave a guest talk on in the course. Here's the big paper he and others wrote on the topic (after the course):
https://arxiv.org/abs/2102.11107