AlphaZero; The Future of Chess?

AlphaZero; The Future of Chess?

By BoboTheFlyingSheep67

Hi guys! I have decided to postpone the 4th article in my "How to Improve Your Chess" series until the spring, and instead will be providing you with other interesting articles  As usual, if you have any suggestions, please post them in the comments below ↓ I will also be analyzing games and posting them in my blog. At the end of every week I'll analyze ONE game from the comments. Ok, without further ago, let's start! I hope you enjoy! 

About AlphaZero

I am sure that, by now, you have all heard of the A.I. Chess Engine, AlphaZero. AlphaZero is a computer program developed by Deep Mind, a Google company. AlphaZero's primary goal was to prove whether creating a self-learning artificial intelligence was possible. It was tested in 3 different areas of games, Chess, Shogi and Go. Within 24 hours of learning, the self-learning AlphaZero had already achieved such a high level of play and it defeated the "world champion" computer programs in each field. AlphaZero was trained solely with itself and with no access to opening books or endgame tablebases. AlphaZero did this all using custom 5,000 first-generation TPUs (tensor processing units) to generate the games and 64 second-generation TPUs to train the neural networks, all in parallel. This was an extraordinary feat and signals that the field of technology is advancing ever-more quickly. However, this could also be a potential problem. What will happen to our old, historic games of chess? As Artur Yusupov noted, "No more theory means more creativity". The advancement of technology in chess could signal a huge era of change, in which computers will have a more influential role on the game. Read on to find out more about A.I. as well as expert's opinions on A.I. and my opinion on A.I.. 

AlphaZero vs Stockfish

Many of us know Stockfish to be the greatest engine (available to the public) in the world right now. It won the computer chess championship earlier this year, and has many accomplishments to date. But, Stockfish, in fact, is pretty weak compared to AlphaZero. In 2017, AlphaZero and Stockfish played an 100 game match. The results clearly showed which engine was most dominant, and were pretty shocking. AlphaZero won an astonishing 28 games, and drew all the rest (Check out the Chess.com article here). Yes, you read that right. AlphaZero didn't lose a single game against the world's leading chess engine! Incredible, right? 

How AlphaZero works

Now that you know this, I'm sure all of you want to know how AlphaZero works. AlphaZero started out as a clean slate, with no information about chess except for how to play. It had been engineered to learn how to play chess, but it knew absolutely nothing about tactics or strategies, much as we are born with the vast capacity to play an instrument, even though we don't actually know how to play it.

Of course, most people would give AlphaZero a number of master games to study and break apart, so that it can "learn" strategy. However this would have resulted in AlphaZero only learning how we play chess, with all our mistakes. This forced DeepMind to take a more ambitious and risky approach known as reinforcement learning. This meant that AlphaZero was left to play billions of games against itself. After each game it would tweak some of its play and try to determine what went well and what didn't - just like what we humans do! When it initially started the learning process, AlphaZero could only play random moves and all it knew was that checkmate is the objective of the game. I am sure many of us reading this can relate to this experience when they first started playing chess. Remember all those random checks and hanging piece? Yup, that's what AlphaZero did. It's amazing to think that the world's best chess engine played as badly as us when we first started! During this learning period, DeepMind measured AlphaZero’s progress by making it play one-second-a-move tournaments with Stockfish, and the previous versions of itself. After a mere four hours of self-play, AlphaZero learned enough to exceed Stockfish’s rating, while examining only about 0.1% of the number of positions Stockfish examined (AlphaZero ran on a software that had a list of 3 candidate moves, and it analyzed all of them thoroughly, unlike Stockfish, who considers all the possibilities.)

Kasparov on A.I.

Who better to give an opinion on A.I. in chess than former world champion Garry Kasparov? Kasparov was the first world champion to lose a match to a chess engine (IBM's DeepBlue) and has witnessed the evolution of technology in chess. Check out this TED talk by Kasparov - 

The Future of Chess

With all of this in mind, what will happen to our game? Should we work with AI to help our chess, or should we leave it be? I, for one, believe that we should stop using chess engines. While this is all incredible, the use of chess engines limits our creativity. I think that we should continue to be creative and come up with new ideas, even if it means worse openings. Let me know where you stand on this in the comments below 

(Sources - https://www.chess.com/article/view/whats-inside-alphazeros-brain)

• Please check out my other posts at https://www.chess.com/blog/BoboTheFlyingSheep67