So, I've decided that I can't possibly write a good algorithm for this until I've become at least 初段 (shodan), because a lot of complex concepts would help in drastically reducing the size of the decision tree just by the shape of the game. By generating and separating territory sizes and determining which ones are in danger, the computer can react a lot faster, doing a short search and singling out a few moves. I've risen from 30級 to about 24級 now, so I've still got a long way to go. In the meantime, it's time to start writing the paper... =(
