Full description not available
I**D
Not just a book about two computers playing chess ... finally we can learn something from computers!
In the early days of computer chess, everyone knew that the programs were tactics calculators. Postal players were forbid from using them (at least in the US) because it was an aid for removing tactical mistakes. But strategically, there was a lot to be desired and GMs regularly beat the computers due to their high level of strategic sense. Deep Blue came along and overcame strategic limitations and found ways with good opening books and endgame tables to defeat the World Champion, ending the idea that a humans could compete with computers. But even then, it was due to the tactical resources that were inherent in a position. In an Over-The-Board (OTB) position, a human would buckle under the tactical complexities when faced with a strategic sacrifice that created an imbalance in a position.So, what has changed? With the advent of deep learning and vastly improved AI techniques rooted in neural network processing (nearly 30 years old in the most primitive form), a complete paradigm shift was possible to the point where a computer could teach itself to play chess at the highest level by just playing through 8 hours worth of games (with games finishing in less than a millisecond). This means that moves without direct tactical benefit could be evaluated with the results stored as not better in terms of number of pawns won, but rather how much does this move increase the probability that the game would be won. The exact measure that GMs have used since the first GM title was awarded by the Tsar of Russia. And quite honestly, if you talk to a strong player, they really don't count pawns, they evaluate strategic opportunity which is heavily tied to tactical means. Entering the scene is AlphaZero playing against Stockfish, a program that many of us have lost to (perhaps countless number of times). And personally, I've never learned ANYTHING strategically from a loss to a computer. All of my losses are based on tactical errors or if stretched the realization that tactical resources were available to the opponent (computer chess program) that rendered my approach as either inexact or just bad.NOW TO THE BOOK itself ... The book first introduces the development of chess programs and then places AlphaZero in that timeline. The authors are clearly passionate about their work, and create a sense of excitement for the importance of their work. Which in a nutshell is how AlphaZero stratetgically approaches the chess game and doesn't follow almost any of the old notions about computer chess. And this is the scary part, has a human-like intuition for a move and a position. This book could have easily been called "Intuition in modern chess programs" because most of the text and analyses written are showing how AlphaZero approaches positions. Something that is incredibly enlightening for me as an "advanced" player. In the end, for humans, it's all about ideas and provides a ton of fresh thinking and a second and third look at resources in any given chess position.The book is well written, and as you may know, the authors have done a ton of YouTube videos that provide a similar narrative. At first I asked the question, why bother the book, I can just watch the videos! But the authors provide a great deal more insight that I've been able to process at a reading pace (versus a watching pace). This is personally important for me because I tend to stare at positions a bit more and really think about what I would do in those positions. It's really this process that makes the book valuable as learning tool.I definitely give this book five (GOLD!) stars, it makes me feel smarter just reading it! If you're an advanced player, it is well worth the read.
I**R
Excellent Overview of a Redirection of Emphases
"Game Changer" is a summary meta-analysis of the contests between the A.I.-driven AlphaZero process and the full-calculation-driven Stockfish 8 one. The book is focused on showing how AlphaZero's drive for piece mobility, open lines, and disruption of the opponent's castled position - despite AlphaZero's often being down in material - makes a change from how players evaluate options now and in opening repertoire.Others are better qualified to discuss the accuracy of the analysis. I do want to point out some items.The book is 415 pages long. A bit over a hundred pages of it goes into the history of computer chess and on how the AlphaZero idea was developed. There is an excellent introduction by Gary Kasparov. The rest of the book is made up of games or portions of games, mostly of AlphaZero vs. Stockfish 8, but also of quite a few games played by human masters that provide parallels to what AlphaZero concocted. The formats, type fonts, and layouts were very attractive.I am not in the 2000+ rating category. Still, I enjoyed playing out games from the two matches on on-line data bases and watching the YouTube analyses by famous grandmasters. I had some favorite games and hoped to see most or all of them reasonably annotated in this book.While I enjoy the book as it is, its real purpose is to pluck out the segments of games where AlphaZero does something special. The authors have chapters based on themes, like "Piece mobility: activity" with a set of games and portions of games therein. Complete games can be rather lightly annotated in their openings and endgames but more intensely viewed in the early middle game.My biggest annoyance is that there are no unique identifiers of games. They all get a title like "AlphaZero Stockfish 8 London 2018", followed by 1. d4 etc.. Thus there is no index of games in the back of the book. The on-line data bases have some ad hoc layout, like a page for 2018 games and a listing like (my invention) "023. W: AlphaZero B: Stockfish 8 Queen's Indian Defense 165 moves [Many games did go over 100 moves.] 1-0". I don't see why the organizers of the event couldn't have come up with an official listing, like Match # - Game # for all these games. If I want to compare an analysis in the book with one from outside material, I have to flip through the book to see if I can find a matching diagram or not.Nevertheless, the writing is good, the paperback book is well made, and the ideas are the freshest around.
M**E
What can machine learning teach us about chess?
A very creditable analysis, seeking to extract practical knowledge for chess played by humans, from chess games played between AlphaZero and Stockfish (repetitive, though: too large fragments of text are copied in different places, with the excuse of analyzing different aspects of Alphazero)
Trustpilot
1 month ago
1 day ago