A chessboard only has 64 squares and 32 pieces. Although the possibilities feel endless for a human chess player, it is ultimately a finite, solvable game. It may only be a matter of time until we end up solving chess once and for all.
Engines have already solved tic-tac-toe and Connect 4. In 2007, checkers was put to rest as well. Let’s look at the history of computer engines and try to determine if and when solving chess will be a reality.
The first real chess engine arrives: The MacHack
In 1967, the MacHack VI was the first computer to defeat a human opponent in tournament play. Despite competing in a few state championships and breaking ground for all chess engines to follow, it was not really that powerful of a chess engine. The runner-up to your local chess champ could probably give it a run for its money.
Solving chess with Deep Thought
Deep Thought was the first chess engine to defeat a grandmaster. In 1988, it defeated Bent Larson, an excellent player who even defeated Bobby Fischer multiple times over his career.
The same designer of Deep Thought, Feng-hsiung Hsu created what is maybe the most well-known chess engine – Deep Blue – while working together with IBM. In 1996, in a six-game series against world chess champion Garry Kasparov, it shocked the world when it stole the first game. However, Kasparov regrouped, winning three and drawing two of the remaining games, ultimately taking the series 4-2.
Feng-hsiung Hsu regrouped as well.
He revamped Deep Blue and it defeated Kasparov in the highly publicized rematch. This was the first time that a chess engine legitimately defeated a reigning world champion. Solving chess inched closer to reality.
Also read: More miscellaneous stuff from The Touchback
The engines become unbeatable
Moore’s law maintained every bit of its momentum and Stockfish emerged in 2004. As an open source project, it is constantly upgraded and improved over time. Today, its ELO rating is estimated at around 3400 (For reference, no human chess player has ever eclipsed 2900. The current chess world champion Magnus Carlson is rated at 2880).
Stockfish has been the top chess engine in the world since 2013. Year after year it wins the Top Chess Engine Championship (TCEC). No human can keep pace.
Assuming that Stockfish’s 3400 and Carlson’s 2880 ELO were to hold steady over time, and they faced each other head-to-head, we expect that Stockfish would win approximately 96 of every 100 matches. Is Stockfish beatable? Yes. Is it incoherently super-dominant against the best chess players in the world? Also yes.
The game of chess truly belongs to the machines.
…and then AlphaZero changed everything
In 2017, Google’s AlphaZero entered the chess scene. It went head-to-head (CPU-to-CPU?) against Stockfish for 100 games and incredibly didn’t lose once. It amassed 28 Wins, 0 Losses and 72 draws.
Moore’s Law dispensed something entirely new. AlphaZero was human-like.
AlphaZero only had 4 hours of “training.” It never accessed a single opening book, piece of endgame data or game logs. It studied that game entirely on its own and, unlike Stockfish, its play feels fractionally sentient.
Chess intelligence is no longer reliant on the messy, trite computations of man.
In 2019, a second matchup between a heavily upgraded Stockfish and AlphaZero was arranged. In a contest of 1000 games, AlphaZero claimed 155 wins, lost 6 and drew the rest.
Are we close to solving chess?
Jonathan Schaeffer is an expert in artificial intelligence and the man who helped solve checkers in 2007. He claims that it will require quantum computing to fully map out all the permutations of a chess game.
It is estimated that a 50-move game of chess entails 1*10^120 possibilities.
For comparison, there are 1*10^81 atoms in the observable universe.
That’s pretty impressive for a mere 64 squares and 32 pieces.
