Chess is one of the oldest games in history, and people often relate this game to having a higher IQ or more computational power. But even the best chess players in the world are prone to making mistakes and bad emotional decisions. So, are these weaknesses enough to make AI better than humans at chess?
AI is better than humans at chess, and it has been for a long time. Ever since IBM’s Deep Blue (a chess program) defeated Garry Kasparov in 1997, we’ve seen a surge in artificial intelligence beating chess grandmasters. We can no longer measure an AI’s strength in chess just by playing with humans but now by playing with other AI’s.
This article will go through the biggest reasons why AI is better than humans at chess. Several factors affect the strength of AI in chess, and we’ll cover all of them in great detail. Plus, we’ll also talk about the history of chess computer programs and their continued acceleration.
What Do You Need To Master Chess?
Before we talk about AI, let’s find out how someone can become a chess master — not based on ELO rating, but based on what most people would consider a “great” chess player.
Most chess masters spend their lifetime mastering the game, and it takes a long time to excel in chess. In fact, it’s one of the oldest games in history, and yet, nobody has solved it — no one has ever played the perfect chess game.
It’s understandable if you consider that every pair of moves (white and black) has 103 possible outcomes according to the Shannon number. And in an average chess game with at least 40 pairs of moves, there will be approximately 10120 possible unique games. That’s a lot of possibilities and calculations for someone to master this game truly.
Computational Power for Chess
Although being a good chess player doesn’t necessarily equate to a high IQ, being a chess master would require higher computational power. It means having the ability to calculate every possible move and pick the best according to any given situation. The more calculations a person can make, the fewer mistakes there will be, and the better they play the game.
Playing chess based on computational power is ideal — the player with the higher computational power will always win the game. But it’s not the only factor that we need to consider when playing chess because emotions and oversight are two other factors that can significantly affect the outcome of any chess game.
The combination of both these factors is why some experienced players lose to less experienced players. Despite all the knowledge and experience in chess, humans are prone to making mistakes — and when you’re playing chess, one mistake could cost you the entire game.
What Makes AI Good at Chess?
As mentioned, three factors can significantly affect any chess game: computational power, emotional decisions, and oversight. Having a higher computational power without emotions and administration can make anyone a great chess player.
Without emotions and oversight, a chess player can always make the best moves, according to the situation, and devise a plan to beat the other player. And these traits are what AI has — it can make optimal decisions without emotions involved and without committing mistakes that humans would have missed. The only thing that AI needs is to have higher computational power than the human player, and it’ll win every game.
Aside from this, AI can learn the game by playing against itself and studying past games to figure out the best move for any given situation. Chess is a very slow game; it’s why chess masters spend years learning to be better chess players. AI can do it faster because it doesn’t need to rest, and it’s not prone to the weaknesses of the human brain when studying, making it more efficient in improving its game.
Please some of our other interesting articles about AI and games, such as “Smartphone AI: Reinforcement Learning for Mobile Games” and “Smartphone AI: What Can Artificial Narrow Intelligence (ANI) Do?”
The History of Chess AI
After World War II, Claude Shannon, an American electrical engineer, mathematician, and researcher from MIT who in 1950 published a groundbreaking paper on computer chess entitled “Programming a Computer for Playing Chess.” This paper describes how a machine could be made to play a game of chess.
Interest in electronic computer chess was stimulated not only by Shannon’s paper, but in 1953 Alan Turing at Manchester wrote a program for playing chess called the “paper machine.”
In the 1980s, world chess champion, Garry Kasparov, was at the height of his career. It was also when computer programmers, particularly in IBM, were developing chess programs that would demonstrate the computational power of computers.
Deep Thought vs. Garry Kasparov
In 1988, Deep Thought — a chess program developed by IBM — became the first computer program to beat a grandmaster after beating Bent Larsen. After that, Garry Kasparov said in an interview that a computer program could never beat him in chess.
In 1989, Garry Kasparov proved it to be true by beating Deep Thought in a two-game match. According to the U.S. Chess Federation (USCF), Deep Thought had an ELO rating of 2551. It was an easy match for Kasparov, but Deep Thought, although a World Computer Chess Champion at that time, was only the start of developing even more powerful chess programs.
Deep Blue vs. Garry Kasparov
In 1996, IBM had a successor to Deep Thought — Deep Blue, and they arranged a match between Garry Kasparov and Deep Blue with tournament conditions. Still, Kasparov won the match 4:2, but he became the first world chess champion to lose to a chess program.
In 1997, Deep Blue and Garry Kasparov had a rematch, and this time, Kasparov lost 2.5:3.5. It also marked the start of the tipping point in chess, where chess programs performed better than humans. During that time, computer computational power was far more limited than what we have today, of course.
Magnus Carlsen vs. Stockfish 9
The world chess champion today, Magnus Carlsen, has a FIDE rating of 2847. Stockfish 9, one of the best chess computer programs, has a rating of 3438. In 100 games, Magnus Carlsen has never won a single match against Stockfish 9, proving that AI is better than humans.
AlphaZero vs. Stockfish
But the improvement in chess programs doesn’t end there. DeepMind — an AI company and a subsidiary of Google Alphabet Inc. — released AlphaZero 2017. It’s an AI capable of “self-training” even without access to chess books and chess games. After 4 hours of training, DeepMind estimates that AlphaZero is already better than Stockfish.
In a 100-game match between AlphaZero and Stockfish, AlphaZero won 28:0:72 (win/loss/draw). The same program that the best chess player in the world couldn’t beat in a hundred games gets crushed in a 100-game match.
The Future of AI Chess
In 2019, DeepMind released a preliminary paper that describes the successor to AlphaZero — MuZero. It’s an artificial intelligence programmed to learn the game from the start and understand winning moves without any input from humans. It uses less computational power than Stockfish and AlphaZero, which means that MuZero may have the ability to understand its position better.
Right now, we’re at a stage where we can no longer use human chess players to measure the strength of a chess computer program. What AIs can do in chess has already exceeded the skills of the best chess player in the world. AI companies are already developing AI to play against each other to help us better understand how we can create even better AIs in the future.
Conclusion of Humans Against Chess AI
Although many people consider chess to be one of the best games, it is also one that exploits the great weaknesses of people’s emotions and oversight. A chess player also needs to have a higher computational power than the other player to win at the game.
You can improve your skills in chess by playing and studying other games and lots of practice. It is not always easy to control your emotions while playing, making us prone to committing mistakes. Add all of these variables with the advancement in artificial intelligence, and it won’t be hard to see why AI is better than humans at chess.
- IBM100 – Deep Blue
- (PDF) What Computer Chess Still Has to Teach Us: The Game That Will Not Go
- Wikipedia: AlphaZero
- Wikipedia: Deep Blue (chess computer)
- Wikipedia: Elo rating system
- Wikipedia: Shannon number
- XXII. Programming a Computer for Playing Chess1
- The Historical Development of Computer Chess and its Impact on Artificial Intelligence
- Turing Plays the First Programmed Chess Game on Paper as a “Human Computer”
- Wikipedia: Deep Thought (chess computer)
- Wikipedia: Bent Larsen
- IBM Archives: Deep Blue
- MuZero: Mastering Go, chess, shogi and Atari without rules
- Wikipedia: United States Chess Federation