I'm a machine learning researcher and would be interested in exploring reinforcement learning in a setting like DW2.

To be feasible it'd need to be possible to obtain the observable game state, perform arbitrary logic, and return chosen actions.

As far as I've seen no 4x games really offer this ...