Using AI to discover new insights

A neural network in chess demonstrates the power of taking a fresh look at a subject and ignoring dogma or conventional wisdom.

From David Hill at The Ringer:

Setting aside IBM’s Deep Blue, which was designed in the mid-1990s specifically to play and defeat one person (Kasparov), the best publicly available chess engines were still only playing top players to a draw by the mid-2000s. By the mid-2010s, however, chess computers were unbeatable. In 2017, a neural network chess engine called AlphaZero took things to an otherworldly level. Programmed with nothing more than the rules of chess, AlphaZero played itself 44 million times. In 24 hours, it was strong enough to defeat the best chess computer on earth with relative ease. After getting a chance to review games played by AlphaZero in 2019, Carlsen declared himself changed. “I have become a very different player in terms of style than I was a bit earlier, and it has been a great ride.”

According to Peter Heine Nielsen in the magazine New in Chess, these changes in Carlsen’s game came from new ideas that AlphaZero uncovered about the value of sacrificing valuable pieces to gain an advantage, aggressively pushing the h-pawn up the board, and using the king as a more active piece.

This last idea is illustrated in a game Carlsen played against Levon Aronian during the 2019 Croatia Grand Chess Tour.

Carlsen was playing white, and in this position, many chess players would consider castling kingside, meaning the king on e1 would move to g1, and the rook on h1 would move to f1. This move secures the king, the most important piece in the game, behind three pawns and a rook. It’s a fortress for the king to hide in while other pieces continue to fight. 

AlphaZero, however, didn’t make moves like that. Despite chess theory’s centuries-long notion that a king should be secreted away from the action until the endgame, AlphaZero preferred to have empty squares surrounding its king; to have the king out in open space meant the piece could be more mobile and would even be able to assist in the fight as needed. So in this case, Carlsen chose to castle queenside, meaning he moved his king to c1 and his rook on a1 to d1, putting his king on the more exposed side of the board. “A spectacular move, at least by our traditional understanding of chess,” Nielsen wrote. “But AlphaZero might approve!” Like AlphaZero, Carlsen believed the king would be more powerful when it had room to breathe.