(via The Conversation)
Let’s set the scene. It is May 11, 1997 and Garry Kasparov is the greatest chess player in the world. He is in New York City, facing an unusual opponent – IBM supercomputer, Deep Blue. People all around the globe tune in to their televisions to watch. Unbeknownst to the world, however, this match would change the course of technology and chess alike. After six hard-fought games by both parties, the match ended with two wins for Deep Blue, one for Kasparov, and three draws, cementing the first victory for artificial intelligence.
(Kasparov vs. Deep Blue, via WIRED)
How Deep Blue Works
Chess is a difficult game. It involves a skill called “calculation”, in which during a game, players go through several different variations of moves in their mind and evaluate the resulting position. Top players can calculate upwards of twenty moves for both players. Deep Blue and many modern chess engines work similarly. It calculates the possible moves for itself and for its opponent, but faster, deeper, and more accurately than humans. The only problem with computers is that they have no common sense. When grandmasters play chess, they intuitively select a few moves that are sensible, but a computer doesn’t have this ability, so they rely on brute force. The problem with brute force is that there are a lot of possible chess moves in a position (upwards of 121 million!) which makes it a slow, unreliable method.
The solution to this was to use a searching method called Alpha-Beta Pruning. Essentially, it is a smarter method of searching every possible move. Once Deep Blue finds a move that is worse than a previously evaluated move, it stops evaluating it and only looks for moves that are better. To improve its speed further, Deep Blue used multiple processors at the same time which allowed it to analyze 100,000,000 moves per second.
(Alpha-Beta Pruning visualized, via GeeksforGeeks)
Modern Chess Computers
Chess engines have improved drastically over the past twenty four years, most notably AlphaZero, Stockfish, and Leela. These modern computers use neural networks and artificial intelligence instead of the brute force tree search that characterized Deep Blue. AlphaZero specifically took a different approach to chess – it was only taught the basic rules and left to figure out the rest on its own through deep learning. The results were incredible with AlphaZero beating Stockfish, the strongest chess engine at the time, in its first four hours of playing the game. It later went on to beat Stockfish 155 to 6 in a 1,000 game match. AlphaZero’s games are closely studied by grandmasters today because of its new ideas, and its dynamic playing style is seen in top level chess today. Deep Blue was certainly not the best, but it was the first, which paved the way for technology, artificial intelligence, and modern chess.
