Russian Chess Grandmaster Uses Zhores Supercomputer to Train for Tournament

30 апреля 2021 г.

Russian chess grandmaster Ian Nepomniachtchi and his team used Zhores Supercomputer located at Skoltech to prepare for FIDE Candidates Tournament in Yekaterinburg.


Chess is one of the oldest known games, and with its age comes complexity. Its earliest forms are said to originate from India and Persia, and legend has it that the king of one of the realms in ancient India was so impressed with the new game that he decided to grant the inventor whatever he wished. As it happens, the inventor was also by some accounts a keen-minded mathematician and so he requested that he be given a certain amount of rice for each square on the board. That is to say, one grain of rice for the first square, two grains for the second, and so on with the number of grains doubling for every square right up to the final sixty-fourth square. The king, who didn't understand what he was about to agree to, granted the wish, but would soon find out that he had agreed to the impossible. If you were to double the number of rice grains for every square on a chessboard, you would have enough rice to cover the whole of modern-day London! Needless to say, this was an impossible request for the king to meet.

Whether this anecdote is true or not is irrelevant; it is merely an interesting numerical fact about a chessboard. However, that is just the beginning of the complexity that surrounds chess. An article in Scientific American states that a typical chess game lasts about 80 moves, yet the number of possible moves is vast. So vast that it would be completely impractical to write the full number down or even say it; it is 10123, or a ten with a hundred and twenty-three zeros following it. To put this into perspective, the number of possible moves in the average chess game exceeds the number of atoms estimated to exist in our universe, which is around 1080 (ten with eighty zeros following it). The number referring to the number of chess moves is known as the "Shannon number," named so after Claude Shannon who wrote the first paper on how to program a machine to play chess back in 1950.

Machines have since come a long way with chess and have long outsmarted humans in their ability; with the blessing of hindsight, it seems almost inevitable that they would become better at the game than their creators. Such is their influence in modern chess that supercomputers are now even being adopted as a training tool in competitive chess. A recent example of this is the Zhores Supercomputer located at the Skolkovo Institute of Science and Technology (Skoltech), which was used to help the Russian chess grandmaster Ian Nepomniachtchi prepare for a tournament. By using this AI-based technology at the Skioltech cluster, Nepomniachtchi and his team were able to assess tens of millions of positions per second during their training for a tournament in Yekaterinburg -- the FIDE Candidates Tournament.

“Chess tournament preparation these days is very dependent on software. Skoltech provided the full range of computational power for me and my team. I’d also like to personally thank Yuri Shkandybin, who set up and supported this work,” Mr. Nepomniachtchi said.

It was the researchers at the Skoltech Center for Data-Intensive Science and Engineering (CDISE) that Nepomniachtchi’s team reached out to initially; the aim was to practice with human-machine interaction as a form of preparation for the tournament.

“Ever since Deep Blue won the game against Garry Kasparov, there is no doubt that computers are better at chess than people. But people keep playing against each other, and to prepare for those games, all leading chess players use machines: they prepare their debut and evaluate various positions to know them better than their competition. Every top chess player has a lot of computational power at their side,” Yuri Shkandybin, “Zhores” systems architect, said.

CDISE researchers, along with the chess grandmaster’s team, adapted a number of existing chess engines for a supercomputer cluster. “Ian’s team had 24/7 access to these resources and used them during preparation for the games,” said Oleg Panarin, the manager of data & information services and technical head at CDISE.

“AI is an important part of modern chess," said Skoltech's Professor Sergey Rykovanov, who leads the high-performance computing group at the CDISE. "Among other things, it is used to evaluate the current position in a game and to predict promising moves. AI can offer previously unknown strategies that later become part of the grandmasters’ repertoires. In the future, you may be able to train a neural network to imitate someone’s style based on their game history."

Magnus Carlsen, the current World Chess Champion who Nepomniachtchi will soon challenge, has said in interviews that one of his heroes is AlphaZero, a Google DeepMind neural network and currently the best chess software in the world. Nepomniachtchi says his team hopes to continue working with Skoltech to prepare for the ultimate game. “We have some ideas, but it would be premature to talk about them,” Nepomniachtchi said.

“We in the Skolkovo ecosystem are very happy our computational resources are applied in all sorts of spheres and have played a small but significant part in Ian’s victory,” said chairman of the Skolkovo Foundation and FIDE president Arkady Dvorkovich.

The game between Nepomniachtchi and Carlsen will take place from November 24 till December 16 in Dubai.