[bbq]

Frame your problem as a game, let AlphaZero take a shot. Now we just need the AlphaZero part.

[EGreg]

That kind of deep learning (MCTS) only works if it can play against itself.

Many things require a human in the loop -- judging whether a joke is funny, for instance, or clinical trials or educational apps.

What are some good AI learning things for that?