Controlled Conspiracy-2 Search

Abstract

When playing board games like chess, checkers, othello etc., computers use game tree search algorithms to evaluate a position. The greatest success of game tree search so far, has been the victory of the chess machine ‘Deep Blue’ vs. G. Kasparov, the best human chess player in the world.

When a game tree is too large to be examined exhaustively, the standard method for computers to play games is as follows. A partial game tree (envelope) is chosen for examination. This partial game tree may be any subtree of the complete game tree, rooted at the starting position. It is explored by the help of the αβ-algorithm, or any of its variants. All αβ-variants have in common that a single faulty leaf evaluation may cause a wrong decision at the root.

To overcome this insecurity, we propose Cc2s, a new algorithm, which selects an envelope in a way that the decision at the root is stable against a single faulty evaluation. At the same time, it examines this envelope efficiently. We describe the algorithm and analyze its time behavior and correctness. Moreover, we are presenting some experimental results from the domain of chess.

Cc2s is used in the parallel chess program P.ConNerS, which won the 8^th International Paderborn Computer Chess Championship 1999.

This work was supported by the DFG research project “Selektive Suchverfahren” under grant Mo 285/12-3.