Improving Search in Go Using Bounded Static Safety.

Finding the winner of a game of Go is difficult even on small boards due to the game’s complexity. Static safety detection algorithms can find a winner long before the end of a game, thereby reducing the number of board positions that must be searched. These static algorithms find safe points for each player and so help prove the outcome for positions that would otherwise require a deep search. Our board evaluation algorithm Bounded Static Safety (BSS) introduces two new methods for finding valid lower bounds on the number of safe points: extending liberties and intersecting play . These methods define statically evaluated greedy playing strategies to raise the lower bound of a player’s guaranteed score. BSS can solve positions from a test set of 6 × 6 games at an average of 27.29 plies, a significant improvement over the previous best static safety method of locally alternating play (31.56 plies), and far surpassing Benson’s unconditional safety (42.67 plies).