What chess shows about AI?

What is chess?

Before telling you how fast AlphaZero learned to play chess, I will quickly tell you what chess is. Chess is almost 1500 year old and has an 8x8 playing field and each player has 16 pieces. And what makes the game hard for computer is that it can not be solved by computers or brute forced because just after 3 moves there are 121 million different possibilities and top level games usually take over 50 moves.


AlphaZero

AlphaZero was developed by Google's Deep DeepMind in late 2017. It used reinforcement learning and self-play in order to train it’s neural networks. It was trained to play shogi, go and chess and succeeded in all of these games, but I will only go in depth for chess.


AlphaZero vs Stockfish

When AlphaZero first played Stockfish(strongest chess engine at the time) in December 2017, AlphaZero scored 28 wins, 72 draws and zero losses. It was impressive but there were people arguing that AlphaZero was using Google supercomputer and Stockfish was run on a “laptop”.

So after a year AlphaZero team made a rematch, giving Stockfish a large time advantage and this time Stockfish was run on the same type of hardware that was used in Top Chess Engine Championships. Despite that AlphaZero won again with an impressive score of 155 wins, 839 draws and 6 losses.

Since then the development of AlphaZero has been stopped in 


How strong are chess engines in 2021?

To give some context an average chess player has a elo of 1500, while casual players have 600-800 elo points.

Magnus Carlsen (Best chess player alive) has 2847 elo points.

Currently the top chess engine (Stockfish 13) has 3546.

That basically means that Magnus can’t beat Stockfish even when given an advantage.


Why is an engine beating a human impressive?

It is impressive because chess is’t just a game that can be brute forced to victory. Chess has been played by humans for almost 1500 years and AlphaZero outperforming humans in such a short period of training shows the potential of self-learning AI.



Used articles: 

https://deepmind.com/blog/article/alphazero-shedding-new-light-grand-games-chess-shogi-and-go

https://www.chess.com/terms/alphazero-chess-engine

https://www.quora.com/What-is-considered-an-average-chess-rating

https://deepmind.com/research/case-studies/alphago-the-story-so-far

https://www.livescience.com/32431-who-invented-chess.html

https://www.popsci.com/science/article/2010-12/fyi-how-many-different-ways-can-chess-game-unfold/

 

Comments

Popular posts from this blog

Eesti areng e- riigina

Bitcoin

free to play, non pay to win