AlphaGo and AlphaZero - artificial intelligence (AI) neural network

The level of a chess player is rouletted quite objectively by the Elo rating.

The highest result among people belongs to Magnus Carlsen. His rating reached an astronomical value - 2889.2 points. This is a lot. An international master is 2400-2499 points. And the rating of the strongest computer program Stockfish is more than 3300. There is already space between it and a person.

So, the artificial intelligence AlphaZero defeated the Stockfish program. Out of a hundred games, Stockfish did not win a single one and made many draws, but only when playing white. And it lost 28 times. But the most interesting thing is that AlphaZero spent only four hours "studying" chess. AlphaZero did not study chess in the traditional sense. It has no opening book, no endgame tables, no complex algorithms for assessing the strength of central and flank pawns.

Its work can be compared to a robot that can use thousands of spare parts, but does not know how an internal combustion engine works - it tries out possible combinations until it builds a Ferrari, and it does this in less time than it takes to watch the Lord of the Rings trilogy. In four hours, the program played many games against itself, becoming its own teacher. So far, the team of programmers has remained silent.

They did not give Chess.com a comment, citing the fact that the report is "still under review." The most interesting thing is that the issue is not the speed of calculating variations or the speed of the new champion. AlphaZero estimates "only" 80 thousand positions per second, compared to 70 million per second for Stockfish. Peter-Heine Nielsen, a long-time second of world champion Magnus Carlsen, said "I have always been curious about what would happen if a more intelligent species landed on our planet and showed us their art of chess. I think now I know what it is like."

After a computer won the game of Go - and AlphaGo did it - in chess, this is to be expected. In case you're wondering, this is the work of neural networks, not any hard-coded algorithm.

I wish we could beat adaptive roulette with a neural network... find the "weak link" in the casino program, but we have to use the "trial and error" method

Boris! Here is the whole web of the idea and it starts from the peak of thought. i.e. simple things as a rule no one takes into account, in fact it is necessary to start with something simple, which is what I wanted to tell you about by writing you a letter to the mail. Have you read it?

Chess was played in a position from the Kama Sutra, copying the logo of the world championship The founders of the chess.com portal, Eric Allebest and Jerome Severson, played chess in a position from the Kama Sutra. This was their reaction to the official logo of the world championship in this sport.

The video was published on the Chess.com YouTube channel.



Allebest and Severson first shared their impressions of the logo, and then took the position depicted on it and began the game. The founders of the chess portal said that they were inspired by the logo and expect it to become world famous. It depicts two people holding a chessboard, with their legs and arms intertwined.

After the end of the game, they concluded that holding tournaments in a similar position will bring a touch of freshness and make the new generation of chess more humane. While people are now separated by screens, such competitions will help athletes to be closer to each other. The logo was developed by the Moscow design bureau "Shchuka". The world championship will be held in November 2018 in London.



On December 20, this logo was criticized by grandmasters. In particular, the Briton Nigel Short suggested that the World Chess Federation (FIDE) "decided to teach the chess world a lesson in sex" )))

.... What? An interesting idea! Especially in the He + She variant)

I will definitely play like that)

Alatissa wrote:
I somehow spent a lot of time trying to find out how alphaGo was created.
So, at the very beginning, a large database of games was collected from tournaments.

Just don't forget to post your best methods of play in tournaments)
And most importantly - after the tournament, play in real life)

I don't know how it was in the beginning, but the main idea and the superiority of the program on neural networks is that in such games (in games with full information) it can improve itself without any databases. And as the creators say after putting the rules into the program - it was enough to let the machine play with itself for a day so that it could beat the champion. I am talking about the next modification - Alpha Zero, which beat chess engines and Go and Sega in one go...
It is a little more difficult to use in games with incomplete information, such as poker, but a couple of years ago a method was discovered for such games too...
For some time I watched how, using Alpha Zero as an example, Leela Zero learned to play...and how the network was slightly improved as the results progressed.

Are you, Alatissa, a master of Go or chess?