Modify Tradition to Outlast Chess Computers
Modify Tradition to Outlast Chess Computers
Sathya Gnanakumar Thomas Jefferson High School for Science and Technology
This article was originally published in the 2021 print edition of Teknos Science Journal.
I was planning a grand attack on my opponent’s king. As I frantically tried to swing my queen across the long diagonal, I saw my clock turn red. Time had run out. Disappointed that I lost one of the many bullet chess games I play in a week, I decided to check the analysis board. The Stockfish cloud engine showed I missed a checkmate in the final position.
When we play chess, we are uncertain about future moves and are confident in our evaluations only when we are up or down a lot of material. However, machines can quickly discern their weaknesses and chances in the future. We know machines have far surpassed our capabilities, but can we model our thinking after these computers? For example, the Stockfish engine instantly showed I had checkmate in two moves. Was it possible to see this game-ending sequence earlier? This is the primary aim of chess enthusiasts who attempt to sustain the creative nature of chess and improve human cognition simultaneously.
Just like games such as Tic-Tac-Toe, Checkers, or Connect-Four, the player’s aim in chess is to outsmart their opponent and put them in uncomfortable situations. Different schools of thought have prevailed throughout chess history. In the late 19th century, practitioners such as Paul Morphy emphasized quick, dynamic play in contrast to deep planning and long-term play. Using sacrifices early on in the opening stages such as the King’s Gambit, the Romantic Era led to many spectacular combinations in games as both sides raced to checkmate their opponent’s king. Eventually, sharp, attacking play was replaced by more positional, slow-moving games which resulted in longer and more complex matches. These techniques were passed down through chess coaches and documented in chess books still revered to this day.
Top grandmasters can recall move-by-move hundreds of famous games played by past champions because of chess literature. Not only do chess books provide the notation from the game, but they also provide personal analysis by illustrating the player’s emotions during the tournament. For example, one chess book I read was written by David Bronstein, which described his and fellow players’ experiences in the Zurich 1953 tournament. I was astonished by the psychological demons players faced during their games. Chess is warfare both on and off the board. The book was packed with opening content a chess aficionado could spend years analyzing. Players were so engrossed in the games they couldn’t leave the playing hall. This led to the popularity of post-mortem analysis, in which players spend hours after the completion of the game to uncover hidden possibilities and missed chances. As a result, young players like me were enthusiastic about playing in tournaments and learning from their games.
Analyzing your game after battling your opponent’s ideas for several hours is the beauty of chess. The man you were waging war on is now a curious student searching for answers and new ideas to test in their future games. When I first began playing tournaments, kids always gravitated towards an instructor with a chessboard after finishing their game. It was a sense of excitement when the master illustrated ideas you missed and congratulated you on strong moves. Students gathered around the board anxiously waiting for their turn but also giddy to point out moves to the seated player. This interaction was seen in top-level chess as well as players aimed to tease out minuscule weaknesses in their opponent’s preparation to improve their chances for the next game. Unfortunately, the era of over-the-board analysis started to dwindle as a few Carnegie Mellon students figured out a way to simplify the process: creating a chess computer capable of determining the strongest move within minutes.
Until 1997, top chess players believed no chess computer could surpass their intelligence. Unfortunately, following the creation of Deep Blue, a chess computer that defeated the world champion, Garry Kasparov, the game would never be the same. Initially, Garry Kasparov was able to provide substantial resistance. Kasparov successfully defeated the first version of Deep Blue 4-2 in a six-game match. As described in the official Deep Blue paper, the first version employed a single-chip chess search engine with numerous flaws. Once these flaws were fixed, the developers created a second version which included a new evaluation function and a chip with a faster search speed. After successfully winning the training matches, Deep Blue 2 was ready for the rematch. It was a close match, but ultimately Deep Blue pulled away winning with a final score of 3.5 - 2.5. This result shocked the world as the power of artificial intelligence was revealed to a population getting used to the wave of technology upon them. However, this was only the beginning, as multiple chess engines created after Deep Blue could easily defeat the World Champion. These chess engines could be downloaded on your handheld device and blitz out moves within seconds.
AlphaZero - a computer program designed using artificial intelligence - is now the strongest chess machine in the world after defeating Stockfish (a cloud engine program). Designed by the DeepMind team, AlphaZero uses a deep neural network but remarkably trained itself using constraints, or the rules of chess. AlphaZero played thousands of games against itself and determined the best possible continuations. According to Paul Grünke, who compared the Stockfish and Alpha Zero chess machines in his paper, Stockfish uses an evaluation function where pawns measure the position for both sides. On the other hand, AlphaZero uses a neural network that assigns win probabilities to the current position. AlphaZero’s groundbreaking ability to assess its win probability at any moment is both astonishing and frightening. AlphaZero and Stockfish have simplified the game of chess into mere probabilities, but humans are still searching for new ideas using their logic.
Artificial intelligence has solved the game of chess, but it does not mean all value is lost for humans. Kasparov, who lost to Deep Blue, admits in an interview that he was extremely frustrated towards computer chess initially but now recommends humans to work together with them. Christopher Chabris in an interview with WIRED agrees with Kasparov, elaborating upon the inevitability of artificial intelligence taking over the game of chess. On a positive note, he points to a variety of reasons why chess will not be affected by this development. He explains that since outside assistance is illegal in tournaments, humans will never be able to harness the power of a chess computer. Steps are being taken by researchers and chess enthusiasts alike to improve human cognition while also upholding the integrity of the game.
Vladimir Kramnik, a former World Chess champion, admits top players memorize computer calculations, making the game less creative. However, by working with Google’s DeepMind team, he is looking to spark new life into chess by training AlphaZero to learn nine new variants of the game. For example, one of Kramnik’s personal favorites is no-castling chess. Castling is a maneuver enabling your king to be tucked in the corner next to a rook and behind a stack of three pawns. This variation would presumably make the game more intense as both kings would be at the center of the action. Another modification proposed included pawns moving two spaces on any turn instead of only on the first advance. The researchers then used the power of AlphaZero to determine the validity of these variations. They were shocked to learn the computer began to shift its evaluations in the pawn variation. The queen was valued at 7.1 pawns instead of 9.5 pawns in a standard game. Moreover, they noticed a higher win rate, which could help solve the issue of drawish games in top-level chess and engine games. Researchers are also looking to create chess engines that utilize human decision-making. Reid Mcllroy-Young and others are working on a chess engine called Maia, a variation of AlphaZero that predicts human moves with greater accuracy. We can utilize technology such as Maia to improve our game, as the engine provides a more realistic perspective of the potential moves. It is clear modifications to the rules of chess and current technology can yield promising results for the game. This is why I am a proponent of Crazyhouse, a variant allowing players to think outside the box.
Crazyhouse chess, or bughouse, is a variation of chess played on two boards. There are four players and each team has two players. Normal chess rules apply except whenever a piece is captured by your teammate on his board, you are then given the power to place that piece anywhere on your board. For example, if your partner captures the opponent’s queen, you have the option at any moment to place the queen on any unoccupied square on your board. The game is played using a fast time control, leading to exciting time scrambles with pieces flying across the board. Players are forced to think on the fly and come up with innovative checkmates and traps to lure their opponent into. Unlike regular chess, the board evaluation can flip almost every move and you have no idea when you are winning, as your opponent can turn the table rapidly with a swift “forcing” sequence.
Through innovative modifications to the standard game, theoreticians and researchers alike aim to preserve the creativity of chess. To combat the technological revolution in chess, we need to implement new variants to keep the game enjoyable. It is unfair to chess players when they lose simply due to the superior computing power of their opponent. I have experienced this first-hand and felt the game was worthless to my chess progress. Therefore, we must implement the necessary changes so the future generation can continue to enjoy this mind-game.
References
Campbell, Murray, et al. "Deep Blue." Artificial Intelligence, vol. 134, nos. 1-2, Jan. 2002, pp. 1-27. Science Direct, doi:10.1016/S0004-3702(01)00129-1. Accessed 10 Nov. 2020.
Grünke, Paul. "Chess, Artificial Intelligence, and Epistemic Opacity." Információs Társadalom, vol. 19, no. 4, 17 May 2020, p. 7. PhilPapers, doi:10.22503/inftars.XIX.2019.4.1. Accessed 10 Nov. 2020.
Knemeyer, Dirk, and Jonathan Follett. "How 22 Years of AI Superiority Changed Chess." Towards Data Science, 5 Mar. 2019, towardsdatascience.com/how-22-years-of-ai-superiority-changed-chess-76eddd061cb0. Accessed 10 Nov. 2020.
Knight, Will. "Defeated Chess Champ Garry Kasparov Has Made Peace With AI." WIRED, 21 Feb. 2020, www.wired.com/story/defeated-chess-champ-garry-kasparov-made-peace-ai/. Accessed 10 Nov. 2020.
Mcllroy-Young, Reid, et al. "Aligning Superhuman AI with Human Behavior: Chess as a Model System." arXiv, 14 July 2019, arxiv.org/pdf/2006.01855.pdf. Accessed 10 Nov. 2020.
Simonite, Tom. "AI Ruined Chess. Now, It's Making the Game Beautiful Again." WIRED, 9 Sept. 2020, www.wired.com/story/ai-ruined-chess-now-making-game-beautiful/. Accessed 10 Nov. 2020.
