The beginning months of 2018 have been rather exciting for the world of chess. To name a few of the spectacles that took place, the month of March was dominated by the Candidates Tournament, in which eight players (each of which who competed and won several other tournaments to gain the opportunity to play in this one) faced off against each other for an opportunity to challenge World Champion Magnus Carlsen for his title this upcoming November. Fabiano Caruana (USA) won this tournament rather decisively, with an impressive record of five wins, eight draws, and only one loss to one of the two runner ups, Sergey Karjakin (RUS). Soon after, the chess world was also treated to the Grenke Chess Classic, another unrelated tournament in which Fabiano Caruana and Magnus Carlsen faced off against each other before their Championship match this November (Caruana won this tournament as well, with a draw against Carlsen).
However, despite the rather strong tournaments that have taken place in chess this year, the really exciting news for many casual players is the emergence of a new Neural Network for Chess: Leela Chess Zero (in which the zero corresponds to the fact that the neural network received no prior instruction in chess before it started the machine learning process).
Neural networks are certainly not new to chess; just last year, Google’s DeepMind (one of Google’s artificial intelligence subunits) developed a neural network (Alpha Zero), who famously trained for four hours and was able to decisively beat the world’s best Chess Engine (perhaps, prior to this encounter) under limited time controls. While it is contentious as to whether Alpha Zero is actually more capable than Stockfish (as there was apparently a steep difference in computing power), the games played between the two engines continue to fascinate professionals.
What a lot of people like about Alpha (and now, Leela) Zero’s style of play is that it has all the creativity and intuition involved in human chess, with the precision of an engine. Peter Heine Nelsen, a Danish Grandmaster, had this to say about Alpha Zero’s games: “I always wondered how it would be if a superior species landed on earth and showed us how they play chess. I feel now I know.”
What is really interesting about Leela Zero is that it is an open source project. As such, it runs on limited resources, which is in clear contrast to Alpha Zero, who has the tech giant that is Google to provide its processing needs. What’s more is that a large portion of its training is open to the public, and people playing games against Leela Zero allows the neural network to learn and improve. Right now, Leela is accepting challenges on lichess.org, under the name LeelaChessOfficial (but beware: a challenge such as this is not for the faint-hearted).
The speed of her growth has also shocked her followers. While she was rated at 1800 on the Elo scale just four months ago, she is now rated above 3000, and is now the first neural network to take part in the Top Chess Engine Championship (which is happening right now). While chess Grandmaster Andrew Tang (USA) was able to clinch a win this weekend after Leela Chess Zero blundered away its queen (interesting how neural networks aren’t infallible to mistakes) in a format where each player had twenty seconds, it is doubtful as to how long such a feat will be realizable; one can only imagine how strong such a neural network will be in another month.
Image Source: Pixabay
Originally Published in Bandersnatch Vol. 47 Issue 13 on April 25, 2018