Have you heard about the game AlphaZero? Nine years ago Demis Hassabis, a very strong chess player who used to be ranked as the no. 2 junior in England, founded a company called DeepMind Technologies in London with the idea of establishing a neural network that mimics the short-term memory of the human brain.
Demis Hassabis: “The start-up began working on artificial intelligence technology by teaching it how to play old games from the seventies and eighties, which are relatively primitive compared to the ones that are available today. Some of those games included Breakout, Pong and Space Invaders. Artificial Intelligence (AI) was introduced to one game at a time, without any prior knowledge of its rules. After spending some time on learning the game, AI would eventually become an expert in it. The cognitive processes which the AI goes through are said to be very like those a human who had never seen the game would use to understand and attempt to master it. The goal is to create a general-purpose AI that can be useful and effective for almost anything.”
The results of the research were very encouraging and Google acquired DeepMind in 2014 in keeping with its initiatives to development machine learning. As you know, Machine learning is a method of data analysis that automates analytical model building. It is a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns and make decisions with minimal human intervention.
In 2016 the company made headlines after its Go program, named AlphaGo, beat Lee Seedol, the Go world champion in a 5-game match. As our BW readers know Go is a board game like Chess, seemingly simpler but at the same time more complex. Why do I say that?
Well, simpler because unlike chess in Go all pieces are the same, just black and white, and in Go the pieces do not move around the board.
And also more complex. Chess is a hierarchical game where the object is to catch the king. Go is an imperial game where each player seeks to enclose more territory on the board than their opponent. At the opening move in Chess there are 20 possible moves. In Go the first player has 361 possible moves.
Having conquered Go they then developed AlphaZero, the goal of which was to conquer chess. The machine learning they developed is so highly sophisticated that they merely fed the algorithm machine the moves of chess, it started playing thousands upon thousands of games with itself, and, shockingly, in 9 hours it was already strong enough to beat the strongest grandmasters in the world.
To prove their point AlphaZero played 12 100-game matches against Stockfish, the computer chess world champion. The difference between the two programs is that AlphaZero taught itself to play while Stockfish was programmed to play by human beings — they taught it all the mating patterns they knew, how to play against certain pawn formation, the relative value of the pieces, etc etc. The result was a big blow to humankind — AlphaZero won 290 games, drew 886 and lost 24.
What was even more impressive was the style in which AlphaZero plays chess. Former British champion GM Matthew Sadler wrote a book on Alphazero and comments that: “The first thing that players will notice is AlphaZero’s style, the way its pieces swarm around the opponent’s king with purpose and power.” Underpinning that is AlphaZero’s highly dynamic game play that maximizes the activity and mobility of its own pieces while minimizing the activity and mobility of its opponent’s pieces. Counterintuitively, AlphaZero also seems to place less value on ‘material,’ an idea that underpins the modern game where each piece has a value and if one player has a greater value of pieces on the board than the other, then they have a material advantage. Instead, AlphaZero is willing to sacrifice material early in a game for gains that will only be recouped in the long-term.”
“Impressively, it manages to impose its style of play across a very wide range of positions and openings,” says Matthew, who also observes that it plays in a very deliberate style from its first move with a “very human sense of consistent purpose.”
“Traditional engines are exceptionally strong and make few obvious mistakes, but can drift when faced with positions with no concrete and calculable solution,” he says. “It’s precisely in such positions where ‘feeling,’ ‘insight’ or ‘intuition’ is required that AlphaZero comes into its own.”
Is this the end of chess? Not yet! The hardware running AlphaZero is awesome and costs several hundred thousand dollars a minute to run. Their time will come, but not yet immediately.
What us mortals can do is to study its games and to learn lessons from it. Yes, we are now the ones who the machine is teaching to play chess. And we are starting to see its results. In the recent FIDE World Cup the following game was played. First, let me introduce the players:
Niclas Huschembeth (born Feb. 27, 1992, 27 years old) is a German International Grandmaster originally from Hamburg who learned to play chess at the age of 5. He is a two-time German Chess Champion (2010 and 2019) and represented his country twice in the Chess Olympiads of 2008 and 2010.
GM Niclas managed to beat his first round opponent, GM Arkadiy Naiditsch, with an idea he got from AlphaZero and GM Jorden van Foreest. Naiditsch is a 33 year old GM originally from Riga, Latvia. This fiery tactician moved to Germany and represented his new country from 2005-2015 before moving once again to Baku, Azerbaijan where he currently resides.
Huschenbeth, Niclas (2620) — Naiditsch, Arkadij (2643) [C18]
FIDE World Cup 2019 Khanty-Mansiysk RUS (1.1), 10.09.2019
1.e4 e6 2.d4 d5 3.Nc3 Bb4 4.e5 c5 5.a3 Bxc3+ 6.bxc3 Ne7 7.Qg4
White can also play 7.Nf3 or 7.h4, but I have always thought that the text is the most dangerous to meet for Black.
There are people who like 7…0–0 but it has always looked too dangerous to me, light waving a red flag in front of a bull. Back in the 80s Yasser Seirawan’s Inside Chess Magazine ran a short series to prove that the line is refuted. Some new defenses have been found to counteract his recommendations and currently I don’t know the status but I will always choose white in this line! Here is a short crush which should serve as a warning to Black players. 8.Bd3 Nbc6 9.Bg5 Qa5 10.Ne2 cxd4 11.f4 Kh8 12.0–0 dxc3 13.Rf3 Nf5 14.Rh3 Qc5+ 15.Kh1 Nce7 16.Ng3 Ng8 17.Nh5 f6 18.Bxf5 exf5 19.Bxf6 1–0. Gullaksen, E. (2376)-Williams, S. (2411) Oslo 2004.
8.Qxg7 Rg8 9.Qxh7 Qc7 <D>
POSITION AFTER 9…QC7
What we have on the board is a fairly standard position from the French Winawer. Black’s queen is attacking both the c3 and e5 pawns, and in 9 out of 10 instances White would continue 10.Ne2 Nbc6 (10…Qxe5? 11.cxd4 Qc7 12.Bf4 Qb6 13.Qd3 the center has stabilized, White has an extra pawn and it is passed on the kingside. Obviously, the first player has a big advantage) 11.f4 dxc3 12.Qd3 and Black has either 12…Bd7 or 12…d4, in both cases with a full blooded struggle ahead.
Huschenbeth’s maneuver inspired by AlphaZero and GM Jorden van Foreest, who also studied Alphazero’s games and used this new idea in one of his games last August. The idea is to play cxd5 to stabilize his center. The drawback though is that the e5–pawn is now undefended and Black can take it with check.
When this position came up in the AlphaZero versus Stockfish match the watchers were surprised at White’s 10th move, and even more surprised when Stockfish declined to take the free pawn on e5 and instead continued 10…dxc3 11.Nf3 b6 12.h4 Ba6 13.Qd4 Bxf1 14.Kxf1 Nd7 15.h5 Rc8 16.h6 Qc4+ 17.Qxc4 Rxc4 18.Bg5 Rh8 19.g3 Nc6 20.Rh5 Rh7 21.Re1 b5 22.Kg2 Re4 23.Be3 Ra4 24.Bc1 a6 25.Re3 Rc4 26.Re2 the two engines fought all the way up to the 123rd move although ultimately the game was drawn. AlphaZero (Computer)-Stockfish (Computer) London ENG 2018 1/2 123.
11.Ne2 dxc3 12.Qxc3 Nbc6 13.Qxe5 Nxe5
This is the AlphaZero lesson. Before this game White had preferred to keep the queens on, but the silicon mind looks at the position differently — with the queens off the board Black’s attack is no longer dangerous and, with the bishop pair and a passed pawn on the kingside, White can look to the future with confidence.
The plan is to follow-up with Bb2 and Nh5 to de-stabilize Black’s center.
14…f6 15.Bb2 Kf7 16.Nh5
There is an immediate threat now of Nxf6 followed by f4.
16…Nd7 17.0–0–0 b6
The van Foreest versus Karthik game continued 17…a6 18.Be2 e5 19.Rhg1! b5 20.f4 Bb7 21.g4! d4 22.Rdf1 Ke6 23.g5 Be4 24.gxf6? (A pity. 24.Ng3! Bh7 25.Bg4+ f5 26.Bd1! (to protect c2) 26…Rac8 27.fxe5 Rxg5 28.Ne2 Rxg1 29.Nxd4+ is almost winning) 24…Rxg1 25.Rxg1 Nxf6 26.Rg5 Nf5 Black is already out of the woods. Van Foreest, J. (2626)-Karthik, V. (2462) Zurich SUI 2019 1/2 37.
18.Rg1 Bb7 19.g4 Rh8 20.f4 Rh6
This was the defensive formation that Naiditsch was getting at. But it does not work.
With the obvious threat of g5.
21…Rah8?! 22.Nxf6! Nxf6
23.g5 Rxh2 24.gxf6 Nf5 25.Bd3 R8h7 26.Bxf5 exf5 27.Bd4!
Very alert. 27.Rde1 right away to threaten Re7+ is met by 27…d4! 28.Bxd4 Be4 and Black is still alive.
[28.Rg3 followed by Rdg1 leaves Black defenseless]
28…Rxg7 29.fxg7 Be2 30.Rg1 Bg4 31.Re1 Be2 32.Rg1 Bg4 33.Re1 Be2 34.a4 Kg8 35.Kb2 a5 36.Ka1 Kf7 37.Rb1 Rh6 38.Rg1 Kg8 39.Rg3 1–0
Naiditsch resigns. Clearly White’s rook is going to c3 then c8 and then, with the black pieces tied up defending against the g7–pawn, he will harvest the Black pawns. 39.Rg3 Bh5 40.Rc3 Re6 41.Rc8+ Re8 42.Rc6 Bd1 43.Kb2 Rb8 44.Be5 Rd8 45.Rxb6.
The book that GM Matthew Sadler wrote on AlphaZero (in collaboration with WIM Natasha Regan) was entitled Game Changer. That is certainly starting to ring true.
Bobby Ang is a founding member of the National Chess Federation of the Philippines (NCFP) and its first Executive Director. A Certified Public Accountant (CPA), he taught accounting in the University of Santo Tomas (UST) for 25 years and is currently Chief Audit Executive of the Equicom Group of Companies.