Risk seems like a game that an alpha-zero-like AI could be trained on. The major differences between Go and Risk seem to be an element of randomness, more than

Risk seems like a game that an alpha-zero-like AI could be trained on. The major differences between Go and Risk seem to be an element of randomness, more than 2 players, and a small amount of hidden state (cards). Perhaps the hidden state would interfere with the ability to train too much?