Things Will Never be the Same
Things Will Never be the Same by IM Malcolm Pein from CHESS Magazine March 2019 Issue
Chess will never be the same again. There’s been an earthquake and I’ve felt it. AlphaZero, the chess-playing entity that taught itself using deep reinforcement learning algorithms, is now the dominant chess-playing entity.
AlphaZero was given only the basic rules of the game and improved by playing 44 million games in just nine hours against itself, while updating its neural networks with the knowledge learned from experience.
It gives me a lot of pride to say it was a British artificial intelligence company, DeepMind, based in central London, that developed it. DeepMind was co-founded by former chess prodigy Demis Hassabis. Having given up the game for much more important things when he was a teenager, the development of AlphaZero has rekindled Demis’s love of chess, particularly because it plays so beautifully, so originally and so much more like a human than the software programmes like Houdini and Stockfish which we are used to.
Writing in the journal Science, Garry Kasparov noted: “I admit that I was pleased to see that AlphaZero had a dynamic, open style like my own. The conventional wisdom that machines would approach perfection with endless dry manoeuvring, usually leading to drawn games. But in my observation, AlphaZero prioritizes piece activity over material, preferring positions that to my eye looked risky and aggressive. Programs usually reflect priorities and prejudices of programmers, but because AlphaZero programs itself, I would say that its style reflects the truth. This superior understanding allowed it to outclass the world’s top traditional program despite calculating far fewer positions per second. It’s the embodiment of the cliché, “work smarter, not harder.”
It was the phrase ‘because AlphaZero programs itself, I would say that its style reflects the truth’ that really resonated with me.
Just prior to the London Chess Classic in December, DeepMind released over 200 games of AlphaZero. They had been kept under wraps until a peer-reviewed scientific paper was published in the journal Nature. AlphaZero proved much stronger than the strongest conventional chess engine Stockfish which, like its peers, is programmed to evaluate positions according to criteria honed by human programmers.
AlphaZero was a clear winner in two different contests. When the games were played from the starting position, and AlphaZero could select its openings, it defeated Stockfish +35 =72 -3. In games that started from positions used in computer chess competition (TCEC), the score for AlphaZero was: +17 =75 –8. This negates a criticism one often heard in the early days of computer chess that one program had a better opening book that skewed the result.
As well as playing aggressively, one thing that stood out to me about AlphaZero is the high value it attributes to piece mobility and activity over everything else. It pays far less regard to material gain than conventional software programs and sacrifices pawns with abandon. This also seems far more human.
Back in June of last year, Demis Hassabis allowed Dominic Lawson, Chris Flowers and I a sneak preview of AlphaZero on condition the games were kept under wraps until the peer-reviewed scientific paper appeared. David Howell acted as our advisor. The games against AlphaZero were kindly offered by Demis to the highest bidder at the closing dinner of the 2018 London Chess Classic with the proceeds going to Chess in Schools and Communities.
We played two games against AlphaZero. In the first, AlphaZero played the Berlin Defence and we went for the dullest, symmetrical line. We were incredibly lucky, as the first 17 moves followed a previous game of David’s against Dmitry Andreikin that we had been literally analysing in our heads just before the game. The pre-match plan went something like, “Let’s play the Lopez and hope it plays a Berlin Defence”, and so it came to pass. David mentioned a line we could play and miraculously it came up.
Both games were played with two hours for us, one hour for AlphaZero and a 10-second increment for both sides. It was hugely enjoyable and like all human versus computer games, a slightly strange experience as there is no human player opposite. So we were all pretty nervous, yet excited at what we might see, and it was really strange when we played 1 e4 and AlphaZero started to think. David, a real time trouble addict at points in his career, quipped: “This is the first time in my life I’ve been ahead on the clock!”
I’m in the middle of reading Game Changer, the new book by Matthew Sadler and Natasha Regan. It’s all about the creation and playing style of AlphaZero, as well as its thought processes. The authors have selected some stunning examples of AlphaZero’s play and discuss what it can teach us.
Game Changer is a wonderful read and I can thoroughly recommend it already. It includes a foreword by Garry Kasparov and an introduction by Demis who makes the following point: “At DeepMind we believe one of the keys to AI is the notion of generality, whereby a single system is able to perform well across a wide variety of tasks, much like the brain.”
This is the great hope. By using games as the test bed, and they are perfect for it as the results are clearly measurable, DeepMind hope to be able to employ the reinforced learning techniques in wider areas. AlphaZero is the logical development from AlphaGo which has revolutionised the ancient Japanese game and defeated Lee Sedol back in 2016.
There is also a lovely chapter consisting of an interview with Demis and then a discussion of how AlphaZero thinks. This is particularly fascinating as Matthew and Natasha were let ‘under the bonnet’ and examine AlphaZero’s thought processes in some of its most beautiful games.
I think every chess player should have this book, so I won’t reveal too much but this game, with five pawn sacrifices, illustrates many aspects of AlphaZero’s style and is analysed in depth. For a small preview and more from Matthew and Natasha do see pages 32-33 of this issue.
Vlad all Over
It seemed inevitable, yet it’s still saddening to report that Vladimir Kramnik has decided not to play at the top level anymore. The 14th world champion made this announcement after a chastening experience at Wijk aan Zee – see inside. In recent years Kramnik played some wonderful games as he
changed his playing style to become more risk-taking, almost cavalier in fashion. Yet it’s been producing indifferent results.
It seems that Kramnik can no longer motivate himself to do the huge amount of work required to play the big events and so henceforth he may restrict himself to some rapid chess and exhibitions. His contribution
to chess development and popularisation has been immense. Quite simply he is one of the greatest ambassadors for the game we have.
The only possible upside of his retirement is we will hopefully see him in the commentary room more often. As we know from the London Chess Classic, Vlad is just brilliant at explaining games with the utmost clarity, in terms lesser players can understand. I look forward to a new best games collection and some insights into how he toppled Garry Kasparov back in 2000.
Perhaps Vlad is also a candidate for FIDE President at some time in the future – he has the stature. He may have to revisit his relationship with the Russian Chess Federation as their Chairman Andrei Filatov said just a few days ago that he hoped Kramnik would play the Russian Super-Final this year and even the Chess Olympiad at Khanty-Mansiysk in 2020, which is not at all what I think Vlad has in mind.
England at the World Teams
Following the tremendous result at the Batumi Olympiad, where England came fifth, we have been invited to the World Team Championships at Astana in Kazakhstan, which begins a few days after UK readers receive the magazine. This is the first time we have qualified since 1997. The 10 places are offered to the defending champions, the four continental champions, the hosts, a wildcard place and the top three placers at the preceding Olympiad not already qualified.
Ordinarily, it would not have been possible for the ECF to support our participation. My budget as ECF Director for International Chess only allows for one representative team tournament per year and funds are already earmarked for the European Team Championship in October. Thanks to support from the Scheinberg family, who are great supporters of chess on the Isle of Man, we can field a strong team of: Mickey Adams, Luke McShane, Gawain Jones and David Howell, with Jon Speelman as analyst and reserve. I will captain the team, as John Nunn is unavailable, which is a shame, as he was tremendous at Batumi.
The line-up for the tournament is: China, Azerbaijan, Iran, USA, Egypt, Russia, Poland, England, Kazakhstan and India, while the lineup for the Women’s World Team Championship is: Russia, China, USA, Egypt, Ukraine, Georgia, Hungary, Armenia, Kazakhstan and India. A full report next time (see our April 2019 issue) .
Editorial by IM Malcolm Pein
For live updates, follow Malcolm on Twitter @TelegraphChess
You can subscribe to Chess Magazine www.shop.chess.co.uk