2002

SOLVING THE GAME OF AWARI

Researchers studying artificial intelligence have invested great effort developing game-playing programs—both as a test of AI strategies and to push the limits of software and hardware. One interesting example in the history of game playing involves Awari, a 3,500-year-old African board game. Classified as a count-and-capture game, Awari is a member of a set of strategy games called mancala and has various different names in different countries.

The Awari board consists of two rows of six cup-like hollows—with four markers (beans, seeds, or pebbles) in each hollow—where one row belongs to each player. On a player’s turn, the player chooses one of his or her six cups, withdraws all seeds from that cup, and drops one seed in each cup counterclockwise from this cup. The second player then takes the seeds from one of the six cups on his or her side and does the same. When a player drops his last seed into a cup on the opponent’s side containing only one or two seeds (making a total of two or three seeds), that player captures all the seeds in this cup. The same player also takes any seeds in cups immediately before the emptied cup if they now also total two or three. Players can take seeds only from their opponent’s side of the board, and the game ends when one player has no seeds left in the cups on his or her side. Whoever captures the majority of seeds wins.

Awari has been of immense attraction to researchers in the field of artificial intelligence; but until 2002, no one knew if the game was like tic-tac-toe in which perfect plays from the beginning always ended a game in a draw. Finally, computer scientists John W. Romein (b. 1970) and Henri E. Bal (b. 1958) of the Free University in Amsterdam wrote a computer program that calculated the outcome for all 889,063,398,406 positions that can occur in the game, proving that Awari must end in a draw with perfect play by both players. The massive computation required about 51 hours on a computer cluster with 144 processors.

“Did we ruin a perfectly fine game?” ask Romein and Bal. “We do not think so. Connect-4 was solved as well, and people still play the game. The same holds for other solved games.”

SEE ALSO TicTac-Toe (c. 1300 BCE), Mastermind (1970), Backgammon Champion Defeated (1979), Connect Four (1988), Checkers and AI (1994), Othello (1997)