This is the 1st (!!!!!) tech article on my blog! I’m very excited to share my thoughts about it in direct relationship to its recent impact on the game of chess (a passion of mine), via a brand new technology invented by Google call AlphaZero! I will try to take some of the complexities out of it for easier reading,… so without further ado……(!!)
Imagine this,….. (!!)
A game so complex, its almost impossible to master,…
A game invented a little under 1500 year ago and still popular to this very day, … 3/4 of a millennia of everything from written to digital tactical and strategic history,….
A game where only 10 moves into the game, you could easily have over 26 trillion moves to evaluate and thats ONLY if you think 1 ply (forced, no variations on the theme),….
This I describe is the game of Chess….
Its 1997 and you are the reigning World chess champion by the name of Gary Kasparov, and 1997 is the year which will mark an important time in chess history,.. it will be the first time where a sitting World Champion GrandMaster is bested by a chess algorithm, this time in the form of a IBM super computer by the name of DeepBlue in a Championship match. Computers from here on out are now the best players in the World,….
Fast forward 20 years later to 2017, there are a plethora of strong chess algorithms, all relatively similar in strength and all that play well beyond the capabilities of the best human GrandMasters.
Now you are the computer algorithm World Champion by the name of StockFish and you can beat any and all other human GrandMasters and any and all chess computer algorithms,… keeping in mind that all of you have the EXACT same access or are programmed with that same 1500 year history of study, of databases with millions of games, opening theories, tactical and strategic play, end game table bases, etc, but you Stockfish, though not undefeated with your massive database, are 2016 Chess World Champion. And you have the highest Chess GrandMaster ranking EVER and is the strongest player in the World!
In 2017, along comes a self learning algorithm by the name of AlphaZero. It knows absolutely NOTHING beyond the simple rules of the game. No opening theories, no access to the 1500 year history of acquired and documented chess knowledge, no pre-programmed databases, no end ame tablebases, no nothing. You (AlphaZero) learn and in turn, continuously teach yourself the game and you continuously improve with every game. This is ONLY playing games AGAINST YOURSELF (!!)
You do this self learning and teaching for just 4 hours.
In December 2017, in a 100 game chess match, you (AlphaZero) go on to obliterate the reigning 2016 World chess Champion StockFish. And now you AlphaZero,.. with only 4 hours of teaching yourself the game of chess,.. is by far the strongest chess entity in the known Universe (!!!!) And its not even close!!!
Deep Mind’s AlphaZero, what is it and how does it work?!?!
The Deep Mind (a UK company acquired by Google) division of Google has tackled the game of chess and they wanted to conquer chess and appears they have with their program called AlphaZero. AlphaZero is an algoritm and is equipped with two very important key components, one is a deep Neural Network. In short, a Neural Network is a way to process information that inspired by biological systems, such as the brain and thats counter to how a CPU processes data. And the second highly important component is a general reinforced learning algorithm. Typically, a computer algorithm is designed (or coded) to crunch massive amounts of data, or in chess a huge number of possibilities. The faster the computer’s processor or CPU, the faster it can crunch said data. It will look at every single possibility.
But that’s not the way the biological systems like our brains work, nor the Neural Network of AlphaZero. Our systems remember to eliminate things that we know by experience are inferior, so no need to re-process the same data or instances over and over again. This is the very concept of self learning.
And what that means in the case of AlphaZero, is that without knowing absolutely nothing about the game, beyond the basic rules, (how to move the pieces) learned STRICTLY by playing itself and again learned from absolute scratch. Basically AZ played itself and continuously learned to improved with each and every game ( and every iteration of the algorithm), all while crunching far less data than a multi-ply algorithm in a CPU mainframe.
I actually own StockFish and it’s very very powerful.
Deep Mind used their AlphaZero to challenge Stockfish to a 100 game match and it proceeded to lose not one game. Ratio, AZ won 27 game and drew 73,.. and that by chess GrandMaster standards is considered a crushing victory against ANY World Champion, even more a modern-day Deep Blue type program like Stockfish..
Now sit back,… relax,…. take a deep breath…….
and just let this sink in for a moment….>>>
Chess is almost 1500 years old…. and all these years spent acquiring knowledge about the game, compiling it into the thousands and thousands of books, databases, theories, developing chess computer algorithms designed to best the worlds best human chess players, etc… and AlphaZero in only 4 hours of self learning, becomes the strongest chess entity on the planet! I dont know about you but Im still in shock!!
What is Hypermodernism (?)
Hypermodernism is a school of chess that emerged around 1920. It challenges the idea that the strongest play should be direct play for the center of the board, which in now considered the Classical school. The Hypermoderns demonstrated their new ideas with games and victories that games could be won through indirect control of the center, breaking with the Classical view that the center must be occupied by pawns. Aron Nimzowitsch one of the main founding Father of the Hypermodern school, advocated controlling the center of the board with distant pieces rather than with pawns thus inviting the opponent to occupy the center with pawns, which can then become targets of attack. This was part of the hypermodern framework, which Nimzowitsch encapsulated in his seminal book “My System”, which greatly influenced many chess players including myself. I (and many others) consider it the very first modern thesis on chess. It introduced and formalized concepts of the pawn chain, overprotection, undermining, prophylaxis, restraint, rook on the seventh rank, knight outposts, the dynamics of the isolated queen’s pawn, and other areas of chess.
Although none of the primary exponents of the Hypermodern school ever achieved the title of World Chess Champion, they were amougst the world’s strongest players.
In practice, Hypermodern has not replaced the classical theories, as now modern chess textbooks, Grandmasters and scholars describe Hypermodern and an addition or extension to classical theory.
When I was a kid, I was heavily into the Hypermodern style of play and use to employ 2 system almost exclusively, one the Pirc Defense and the other the Modern Defense. And honestly I was practically unbeatable on the novice level and even beat a few GrandMasters at that time of my early development. One thing for me it allowed was for early transposition (sort of like Jazz improvisation!) into other openings (if I deemed via the structure) on the black side, hence I saw many inter-relationships and that in turn strengthened my game. As white, I played a more classically rooted game. So my all around game was also a mix of the classical concepts, openings, endgame theories, etc.
Hypermodern openings include the Réti Opening, King’s Indian Defence, Queen’s Indian Defence, Nimzo-Indian Defence, Nimzowitsch Defense, Grünfeld Defence, Bogo-Indian Defence, Old Indian Defence, Catalan Opening, King’s Indian Attack, Alekhine’s Defence, Modern Defence, Pirc Defence, Larsen’s opening, Sokolsky Opening, and to a lesser degree the English Opening. Openings such as 1.a3 do not constitute hypermodern openings since, although they delay the occupation of the centre with pawns, they also delay piece development.
Why do I personally call AZ’s approach to the game the new Hypermodern (?!)
Before I considered penning this article, I researched and also downloaded Deep Mind’s 14 page “white paper” titled “Mastering Chess and Shogi by Self-Play with a General Reinforcement Algorithm” describing AZ’s Neural Network concepts and its path to self teaching. I also studied in depth the 10 games out of the 100 that were released by Deep Mind.
I also discovered that even Deep Mind’s founders and developers were surprised at the rate of exponential self learning growth A had with every iteration of the algorithm!
I am the FIRST person I know who is calling AZ’s play the new HM. And if you play chess and know the game, you might also consider this new axiom! Ive NEVER seen any play like it!
In at least one game I studied, AZ moved ALL of its pieces off the back rank (!!!!) in the midgame! This type of structure was previously unheard of and certainly not a part of standard Grandmaster play! Even the greatest Bobby Fischer didnt do that! Probably never considered it! This literally goes against the very basic fundamental concepts of chess itself! And it still wins the game! Actually its UNDEFEATED!!! Its probably 100 levels ahead of the next in-line formidable chess entity! And again, with no previous knowledge of the game!
I also realized something else about how AZ views the game,.. since it came in with no prior knowledge of the game, it seems not to care about nor does it consider the point value(s) of the pieces! So it will (and did!!) sacrifice a Rook (5 pts) for a lowly pawn (1 pt) in the midgame (!!) strictly for tactical play! Now of course thats not the first time that has happened (rook for pawn), but usually in Grandmaster play, it’s with gained tempo or a weakness in the structure for a quick all out assault for mate, not just for a mere seemingly tactical advantage! But it wasnt mere as you watched it exploit that same weakness for checkmate 50 moves later!!!
It will move its King toward the center of the board (even in the opening!!) and forgo the safety of castling! It also seem not to like the standard first move of the King pawn, and that these days is not so usual, (but for a computer algorithm it’s not the usual!!) but personally I equate that to the short amount of time its has to self learn, though I might be totally wrong in this view.
It also doesnt play well with an opening called the Sicilian Defense, which is part of EVER GrandMasters playbook and is a frequently played opening in tournament play and is considered a dynamic opening! It also seemed to likes the French Defense, which is considered draw-ish or even passive by most of today’s Grandmasters, but AZ had its highest winning percentage with that opening!
Foe me, observing that kind of play was like having an Alien sit down across from me and view Other-World-like moves, played with pure brilliance while simultaneously teaching the entire Planet how the game should REALLY be played!!
So in light of this fresh, new approach of going against well established, long held chess norms, and giving Masterclasses with eciting new theories, I’m dubbing it “The New Hypermodern”! (You heard it here first!!!) Frankly, its made me seriously re-evaluate my own approach to the game and start to consider some of these alien-like moves!
There is of course a LOT more to AlphaZero and its Neural Networks, TPU vs CPU, its use of the Monte Carlo Method and more, but I wanted to give you all who havent heard of this exciting new technology a glimpse into what the future holds with this new self learning artificial alien-like intelligence! Whether you care or even realize it or not, RIGHT NOW is a VERY exciting time with the emergence of this new artificial intelligence technology! And not just with the Deep Mind algorithm and its many applications, but also it in conjunction with Google’s new Neural Network framework! That also is a new emerging technology of great potential!! I will at some point cover these other areas, but below I will copy/paste a bit about 2 of them for your further reading curiosity!! I hope youve enjoyed this article!
The Monte Carlo method (from Wiki)
Monte Carlo methods are a broad class of computational algorithms that rely on repeated random sampling to obtain numerical results. Their essential idea is using randomness to solve problems that might be deterministic in principle.
TPU vs CPU (as described by Deep Mind)
The tensor processing unit was announced in 2016 at Google I/O, when the company said that the TPU had already been used inside their data centers for over a year. The chip has been specifically designed for Google’s TensorFlow framework, a symbolic math library which is used for machine learning applications such as neural networks. However, Google still uses CPUs and GPUs for other types of machine learning. Other AI accelerator designs are appearing from other vendors also and are aimed at embedded and robotics markets.
Google’s TPUs are proprietary and are not commercially available. Google has stated that they were used in the AlphaGo versus Lee Sedol series of man-machine Go games, as well as in the AlphaZero system which produced Chess, Shogi and Go playing programs from the game rules alone and went on to beat the leading programs in those games. Google has also used TPUs for Google Street View text processing, and was able to find all the text in the Street View database in less than five days. In Google Photos, an individual TPU can process over 100 million photos a day. It is also used in RankBrain which Google uses to provide search results.
Compared to a graphics processing unit, it is designed for a high volume of low precision computation (e.g. as little as 8-bit precision) with higher IOPS per watt, and lacks hardware for rasterisation/texture mapping. The TPU ASICs are mounted in a heatsink assembly, which can fit in a hard drive slot within a data center rack, according to Google Distinguished Hardware Engineer Norm Jouppi