The Chess Federation relies on AI and the cloud for international competitions


French champions can now rely on a super AI in cloud mode. A technology that allows them to prepare their opening tactics a few hours before the battles, once the opponents know.

In 1997, world chess champion, Russian Garry Kasparov, was beaten by a machine: Deep Blue. A first. With a memory of 600,000 games, IBM’s supercomputer analyzes 200 million moves per second. Developed in 2014 by the company Deepmind (since being handed over to the Google chest), AlphaGo has marked a new turning point. Designed for Go, a game that counts to 10600 possible game against 10120 for chess, it is embedded by AI articulated around a network of artificial neurons. At the end of 2017, it gave birth to AlphaZero, which swept Stockfish, one of the most popular training machines among chess players, then world champion in the discipline in the machine category. Since its version 12 was released at the end of 2020, Stockfish has also integrated a neural network. Latest chapter: The French Chess Federation (FEE) took Stockfish to install it on Microsoft’s Azure cloud. The goal? Benefit from a smart supercomputer to prepare for international conferences. FEE notably plans to use it at the next World Team Chess Championship (the Olympiad) which will take place at the end of July-beginning of August. A tool that can really be a game-changer for French women’s and men’s teams.

“Since the mid -1990s, chess players have been using expert systems based on rule engines that, from one position, suggest moves based on memorization of previous games,” explains Eloi Relange. , President of the French Chess Federation. “With AlphaZero and the advent of AI, the machine began to dig up new techniques, to block King for example, but also new moves.” This is the surprise. We thought we knew all about this game that was born in Asia between the third and sixth centuries of our era. Before incorporating an AI, Stockfish followed the brute force principle. For each position, a tree of possible moves was identified and analyzed. The amount of data to be analyzed therefore increased sharply. A side effect compensated by the growth of computing power, but also by the elimination of branches deemed unrelated to players.

200 million positions are checked every second

Available open source, Stockfish now puts AI within the reach of all players. All that is left is to have the necessary computer resources. “Having Stockfish ported to Microsoft’s cloud allows us to free ourselves from limitations in terms of computing power”, argues Eloi Relange. The French Chess Federation benefits from the American group’s HB120rs machine, the largest in their category (CPU). It is estimated that one infrastructure that can check 200 million positions per second, against 20 million for a conventional high-end laptop, the FFE. “Instead of 3-4 minutes to get a reliable positional diagnosis, the waiting time is reduced to 15 seconds, resulting in an evaluation depth of 40 movements. Which is the time required for the human brain. to understand the position ”, shows Eloi Relange. The result: fast, efficient and accurate work. At the same time, FEE continues to use the historic version of Stockfish. A brick that is certainly less creative in terms of strategy, but makes it possible to analyze a larger number of positions and possibilities due to a simpler and better optimized (and therefore faster) method of calculation. “We will opt for this solution, for example, in the case of an opening near the final to explore as many ways as possible”, specified Eloi Relange.

At this summer’s Chess Olympiad, preparers will have approximately fifteen hours to carry out their review work between the moment when the opponents are revealed and the meetings themselves. Hence the interest of having AI in cloud mode. Thanks to the combination of Stockfish’s neural networks and Microsoft’s Azure computing platform, this work will be carried out in time with reliability that France has not yet achieved in the context of a world chess championship. “First, the coach determines the opponent’s playing style, then selects an opening game to take advantage of his mistakes. Then, starting with the games mentioned for this opponent, he deepens the key -related solutions. of positions by exploiting Stockfish in Azure. He then drew up a summary sheet for the athletes “explains Eloi Relange. The main advantage of the machine? This will avoid moves that seem tactically interesting in the short term, but are likely to lead to defeat, in favor of moves that are considered less relevant at first glance but can lead to success in the long run.

“In France, we are among the first to offer this possibility”

“In France, we were among the first to offer this possibility to women’s and men’s teams. Until now, the calculation in cloud mode was usually limited to the top 10 players in the world”, estimates the president of FEE. While waiting for the 2022 Chess Olympiad, Alireza Firouzja was the first Frenchman to benefit from the solution. Goal: prepare for the qualifying tournament to compete for the title against the current world champion (or Candidates Tournament). A tournament that will feature the eight best players in the world and will be held from June 16 to July 5 in Madrid. Here, the technological challenge of benefiting from significant computing power in a limited time disappears as long as the opponents are known in advance.

A competitive advantage

“From now on, we will have the de facto advantage in managing openings that represent an important stage in the game. But we must not forget that chess is more than that. Other factors come into play: the mental, the physically, not to mention the speed of reflection and the mid-game stages that are so impossible to model mechanically ”, insisted Eloi Relange.

For the future, FFE sees significant room for improvement. First track considered: to face the AI-oriented chess engine against each other. A technique that is still little or not used in France that makes it possible to dig out strategies and unexpected positions due to the different playing styles of different machines. Example: pitting Stockfish against Leela chess Zero. Another well -known neural chess engine, the Leela is inspired by AlphaZero and its reinforcement learning technique. On the cloud side, FEE aims to improve Stockfish’s execution performance in Azure to get the maximum out of each compute instance used. Another planned project: make the Stockfish-Azure couple available to preparatory players under 12 years old. An age group that is very good with the best players in the world in their category. Something to prepare for the future.

Leave a Comment