“The self-play paradigm that was successful for AlphaGo is not easily transferable to domains outside of two-player, zero-sum games because the objective function is much harder to define.”