Google headquarters in London from inside, with the DeepMind section on the eighth floor. | Photo: Maria Emelianova/Chess.com

 

In this article what’ll be covered may very well be a defining factor of future of human kind. But before jumping into it let’s quickly browse through human history on chess playing programs and its advancement.

One of many exciting applications of computer based intelligence is used in chess. Dietrich G. Prinz, a German computer scientist and pioneer wrote the first actual, automated chess playing program in 1951 in England. Alex Bernstein, an IBM researcher, wrote the first complete, fully automated chess playing program, running on an IBM mainframe, seven years later.

In 1956, a research program on Artificial Intelligence was officially launched and two years later it had its major success -  electronic computers were made to play chess.

By the 1960s, AI had produced a program that could play real, actual, automated chess against a human opponent. In the next decade chess playing programs consistently improved taking advantage of increasingly powerful hardware and better software along with new innovative techniques allowing chess programs to optimize their search strategies. 

1988 — Deep Thought – a chess playing computer developed by Carnegie Mellon University and later at IBM shares first place in the Software Toolworks Championship, ahead of former world champion Mikhail Tal.

• 1989 — Deep Thought loses two exhibition games to Garry Kasparov, the reigning world champion.
• 1996 — Deep Blue, previously known as Deep Thought, lost to Kasparov, some considers him the greatest chess player ever lived, 4 to 2.
• 1997 — Deep Blue, revamped, could evaluate 200 million chess positions per second, unprecedented as far as computing power went at that time, beat Kasparov 3.5 – 2.5.
• 2006 — the undisputed world champion, Vladimir Kramnik, is defeated 4–2 by Deep Fritz.

 

Stockfish, a free and open-source chess engine is considered one of the top most chess-engine. Stockfish can use up to 512 CPU cores in multiprocessor systems. The maximal size of its transposition table is 1 TB. Stockfish implements an advanced alpha–beta search and uses bitboards. Compared to other engines, it is characterized by its great search depth, due in part to more aggressive pruning and late move reductions. It won the 2016 TCEC Championship and the 2017 Chess.com Computer Chess Championship. +3400 ELO (Chess rating system)

 

AlphaZero   is an artificial-intelligence program developed by Google researchers at DeepMind AI lab.

On December 6, 2017 AlphaZero competed with Stockfish in a series of 100 games and simply obliterated it by winning 25 games while playing as white (with first mover advantage), and picked up three games playing as black. The rest of the contests were draws, with Stockfish recording no wins and AlphaZero no losses. 100-game match with 28 wins, 72 draws, and zero losses.

 

Now, comes the most intriguing and interesting part. One computer chess program beating another one, no matter how badly, is not that important unless there’s something really innovative at the core of it all. A fully autonomous AI system – it only learns by playing itself, never facing humans.  By contrast, other chess programs basically learn how to play the game by watching moves made by human players. After being programmed with only the rules of chess (no strategies), in just four hours AlphaZero had mastered the game. It had totaled about 20 million games at the time it beat Stockfish. 

AlphaZero does not build a database of the games it has played so that it can simply look up what to do like traditional chess engines do. Instead, it uses reinforcement learning to "learn" a policy that tells it what to do on any given state. This policy is derived from the outcome of training via self-play and getting positive or negative rewards based on the outcome of each game played, from which it eventually derives which moves were good or bad. As a result, much like humans, relying on reasoning (which is done by a combination of neural nets in your brain), AlphaZero searches fewer positions than its predecessors. According to the researchers it looks at "only" 80,000 positions per second, compared to Stockfish's (traditional chess engine) 70 million per second. After the Stockfish match, AlphaZero then "trained" for only two hours and then beat the best Shogi-playing computer program "Elmo." 

Let’s wrap it up with what one of the best chess players humans have ever produced, Kasparov, had to say about it. "Of course, I’ll be fascinated to see what we can learn about chess from AlphaZero,” he commented, “since that is the great promise of machine learning in general—machines figuring out rules that humans cannot detect. But obviously the implications are wonderful far beyond chess and other games. The ability of a machine to replicate and surpass centuries of human knowledge in complex closed systems is a world-changing tool."

Primary Source: Chess.com