There are certainly challenges, but "incorporating a world model" has been going well recently: "Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model"
To quote the article, "Model-based reinforcement learning aims to address this issue by first learning a model of the environment’s dynamics, and then planning with respect to the learned model."
So they've sped up learning from examples by learning a model first, but it's still learning from examples.