Unsolved: The Story of Chess
“Chess is the touchstone of intelligence.”
— Johann Wolfgang von Goethe
Chess, a game played by many, but mastered by few, has been played for centuries. Strategies have changed over time but the rules remain the same. Achieve a checkmate in any possible way. The game was mainly played as a display of art over theory when it was first created in the 1700’s. Wilhelm Steinitz began to take a scientific approach in 1873. He showed the importance of your own positioning on the board as well as learning to exploit your opponents. Short-term tactics and long-term strategy have evolved with each new world champion from attacking, defending and positions, but in 1997 IBM’s “Deep Blue” beat the world champion, Garry Kasparov, and set off a wave of intrigue.
Computer scientists and computer engineers have tried to build computers which played chess since the development of computers in the 1950’s. Creating a chess related machine or computer program is interesting because:
- Problems are easily defined within a set of allowed moves
- An ultimate goal is achievable
- Requires thinking for skillful play
- Solutions are not simple or difficult to find and understand
Due to these factors chess is limited to a finite amount of moves and is susceptible to using a brute force approach. Built with all of the accumulated human knowledge on the game, computers would use their processing power to analyze all of the possible positions in the game, which is far beyond the realm of human understanding. As a reference Deep Blue was able to look at 200 million moves a second while Kasparov is estimated to look at 2.
Although the match in 1997 was controversial and Kasparov lost(Kasparov felt he was at an unfair advantage). Another computer program didn’t beat another champion until 2006, when Vladimir Kramnik lost to Deep Blue in convincing fashion. Human-Computer matches have largely become a thing of the past because computers can now consistently beat the strongest chess players in the world. Since then computers have become more powerful and human updates to algorithms to find positions effectively.
AlphaZero which is the repurposed AI, AlphaGo, was made by Google DeepMind. It beat the worlds best chess playing computer program after having taught itself for 4 hours. Winning or drawing in all 100 matchups with Swordfish-8. The difference in programs are that AlphaZero is given no human input except for the game rules.
“It’s more powerful than previous approaches because by not using human data, or human expertise in any fashion, we’ve removed the constraints of human knowledge and it is able to create knowledge itself,”
— David Silver, AlphaGo’s lead researcher.
AlphaZero like AlphaGo plays itself over and over starting from random placements until it learns the best moves. A good move allows you to win the game. While a bad move allows you to lose the game. AlphaZero is learning the same way a human would and finding new tactics along the way. The program runs on software neurons which forms an artificial neural network. For each turn it looks at the board and sees the moves. Then chooses a move based on the probability of leading to a potential win. After each game it stores the data in its network and becomes stronger for the next game it plays. As this happens who knows what interesting discoveries it will make.
“You can see it rediscovering thousands of years of human knowledge.”
— Demis Hassbis
As humans at least we have one thing over AlphaZero, and that is the ability to say “No”. We are the creators that allow our creations to be amazing in their sphere of influence. The more common we become the more our differences will define who we truly are.