Alphazero vs Stockfish
December 9th, 2017
The game of chess took a rather depressing turn in 1997 when IBM's deep thought beat Garry Kasparov in a man vs. machine sort of duel. The match was depressing not because a machine had beaten the greatest chess mind, possibly in history, thereby overtaking humans in what was always considered a very human endeavor. It was depressing because the way deep thought operated versus how humans play the game. Deep thought was the first successful brute force chess engine and its only tactic included doing a comprehensive search of future moves and choosing the best move amongst them. In this it was further abetted by the sum total of human knowledge in the form of opening theory developed over many centuries, coupled with a human sort of understanding of the value of the various pieces. It was a purely algorithmic answer to the human way of playing chess. A robotic approach versus a more creative, joyful, and intuitive approach which humans take. That was 1997. In the last 20 years, chess engines have massively improved and they can now easily calculate hundreds of millions of future moves. They have not been defeated by any human since 2005, as far as I know. The advent of the chess engines, however, has lead to an absolute deadening of the spirit of the game as the grandmasters try to imitate the robotic way that the engines use to play their games. So what was so beautiful and creative in such early geniuses as Tal, Morphy, Capablanca, and Fischer, has degenerated into the the grueling and grinding styles of the modern grandmasters. Their games are efficient and brutal and lack all elements of joy and beauty. They are like the chess engines themselves.
Well, we have a new kid on the block and this new kid might very well take chess back to its early roots. Google's Deepmind created a deep learning chess engine called Alphazero and only gave it the rules of the game of chess. Alphazero then played the game against itself for four straight hours with absolutely no input from any human. No opening theory and no hardcoding of piece values. No history of chess games played. At the end of the four hours, Alphazero played 100 games against Stockfish, the strongest chess engine available (playing against humans would be a waste of time and a foregone conclusion.) It did not lose a single game and won 28 out of 100. That's an astonishing feat for any entity, especially one that has basically learned to play chess all by itself in just four hours! There are no words to describe my amazement. What I do know is that this is a historic moment in the game of chess and compared to this the 1997 victory looks like a parlor trick. It's a historic moment not only because Alphazero so convincingly beat the greatest chess entity on the planet but also in the manner of its victory. I don't think I could ever have imagined associating adjectives like sublime, beautiful, imaginative and creative to a product of an algorithm but some of the games that I have seen can only be described as such. Alphazero plays like no other chess engine and it plays like no other current human grandmaster either. It seems to have the kind of strategic understanding of the chess board which was always considered possible only within the human consciousness. Only difference is that its strategic understanding seems deeper than any other human in history and it allows it produce moves of such surprise and delicacy with effects so long lasting and devastating that the games feel like a product of a superior and alien intelligence. In a small way Alphazero feels like a strange throwback to the swashbuckling chess masters of the past such as Morphy and Tal. At the very least, Alphazero and its successors will rewrite much of chess theory. They may even take humans back to a past style of play which was more joyful. They also may very well spell an end for competitive chess. After all, why would anybody want to watch a form of chess which is inferior to what Alphazero can play in every single way?