Can this be true? AI, Go and chess
The game of Go is one of the most ancient and most popular games in the world. As a result of its elegant and simple rules, the game has long been an inspiration for mathematical research. For example, in a remarkably advanced calculation, Shen Kuo, an 11th century Chinese scholar, estimated in his Dream Pool Essays that the number of possible board positions is around be ten to the power of 172.
In 2016 research into the game by Cambridge mathematician John Conway led to the number of legal positions being confirmed as 10 to the power of 170 , which makes the computer free lucubrations of Shen Kuo, one thousand years ago, even more astonishing.
By way of comparison, Chess has ten to the power of 40 (legal) positions, far fewer than Go. Nevertheless, the complexities of chess, driven by different pieces operating with different mobilities, is still more than sufficient to baffle most human brainpower.
Now, it seems , a human player has comprehensively defeated a top-ranked AI system at Go, in a surprise reversal of the 2016 computer victory against the human world champion, that was seen as a milestone in the rise of artificial intelligence.
The improbably named Kellin Pelrine, an American player, who is rated one level below the top amateur ranking, beat the machine by exploiting a previously unknown flaw that had been identified by another computer. Nevertheless, the series of head-to-head confrontations, in which he won 14 of 15 games, was undertaken without direct computer support. This is quasi miraculous news, not least because there is a tremendous gulf between the leading amateur players of Go, and the committed professionals.
This counter-intuitively lopsided triumph was claimed as pinpointing a weakness in the best Go computer programs, that is shared by most of today’s widely used AI systems. Russia, Putin and the guardians of western cyberspace, please take note.
According to reports last week in both The Times and the Financial Times, the winning plan was then ruthlessly delivered by Pelrine. “It was surprisingly easy for us to utilise this system,” said Adam Gleave, chief executive of FAR AI, the Californian research firm that designed the program which detected the chink in the Silicon Brain’s armour. Gleave’s software contested more than 1 million games against KataGo, one of the top Go-playing systems, to find a “blind spot” from which a human player could profit. The winning strategy revealed by the software “is not completely trivial but it’s not super-difficult for a human to learn and could be used by an intermediate-level player to beat the machines” said Pelrine.
As we have seen, AlphaGo, a system devised by Demis Hassabis’ Google-owned research company DeepMind, defeated the human world Go champion, Lee Sedol, by four games to one in 2016. Sedol attributed his retirement from Go three years later to the rise of AI, saying that it was “an entity that cannot be defeated”. AlphaGo is not publicly available, but the systems Pelrine put to the sword are considered to be of equal strength.
In Go, two players alternate in placing black and white stones on a board consisting of a 19×19 grid, seeking to encircle their opponent’s stones and enclose the largest amount of space. The huge number of different combinations means it is impossible for a computer to assess all potential future moves. The tactics used by Pelrine allegedly involved the gradual stringing together of a large “loop” of stones to encircle one of his opponent’s groups, while distracting the AI with moves in other corners of the board.
Unbelievably, the Go-playing bot did not notice its vulnerability, even when the growing encirclement was almost complete, according to Pelrine. “As a human it would be quite easy to spot.” This procedure seems almost too easy.
“The discovery of a weakness in some of the most advanced Go-playing machines points to a fundamental flaw in the deep learning systems that underpin today’s most advanced AI” added Stuart Russell, computer science professor at the University of California, Berkeley.
The precise cause of the Go-playing systems’ failure is a matter of conjecture, but it is apparently common to identify flaws in AI systems , when they are exposed to the kind of “adversarial attack” used by Pelrine against the Go-playing computers. One is tempted to write ad hominem, but in this case ad machinam is doubtless more apposite.
In spite of all the evidence above, alarm bells are ringing in my mind, as I write. I have, as yet, been unable to track down any moves or game scores from Pelrine’s slaughter of the silicon brains. Somehow, it seems too good to be true. My hunt for verification will continue and results will be reported here.
Go is a game which tends to absorb its devotees in epic fashion. One story will stand for many in the Japanese Lore which has built up about the game:
Although its Go column had been suspended in March 1945, The ‘Mainichi’ newspaper supported the third Honinbo title match. Iwamoto Kaoro, 7-dan, was the challenger to Hashimoto Utaro. Finding a venue in bombed-out Tokyo was impossible, but Kensaku Segoe came to the rescue; with the help of friends, he arranged for the match to be played in his hometown of Hiroshima. The players had been warned by the Hiroshima police chief not to play within the city, as it was too dangerous. They took advantage of his temporary absence to play the game as scheduled, on 23-25 July, ignoring the rain of bullets on the roof from strafing aeroplanes. Iwamoto won.
When the police chief heard on his return what they had done, he was furious and absolutely forbade them to play any more games in Hiroshima. It was arranged to play the second game in a house in Itsukaichi, an outer suburb of Hiroshima, on 4-6 August.
In the third day of the game, the players saw the Enola Gay fly in. Hashimoto later stepped out into the garden and saw the flash of the atomic bomb and the mushroom cloud rising above the city. A great blast of wind broke the windows, made a mess of the room and upset all the pieces, but since they had reached the endgame, they set up the position again and finished the game, which led to a five point win for Hashimoto. It was not until evening, when the streams of survivors pouring out of Hiroshima began to reach Itsukaichi, that they realised the magnitude of the disaster and just how lucky they had been. The house where they would have played was destroyed.
Both players survived into the 1990s and Hashimoto Utaro died in 1994.
_________
Meanwhile, apropos my last week’s mention of the poet John Betjeman, I have been contacted by world famous chess collector, Jon Crumiller:
“I thought you might like to know that John Betjeman’s grandfather, George Betjemann (yes, two ‘n’s at that time), had a chess connection. He made premium chess set boxes for some of the top chess set makers, and his label is on some outstanding antique chess boxes (or caskets), such as this one, the star auction lot of the 2007 Christie’s Cholet auction. This set and box sold for 150,000 GBP (hammer + buyer’s premium).”
A FINE AND RARE GERMAN CARVED IVORY ANIMALIER CHESS SET
RETAILED BY WALTER THORNHILL, CIRCA 1870
Depicting owls versus mice, the owls partially stained brown, the kings and queens crowned and robed, the bishops wearing mitres and holding croziers, the knights wearing plumed helmets, carrying shields and holding swords, the rooks perching on ruined turrets, the pawns individually modelled as warriors, whilst some modelled as wounded warriors, all with inset glass eyes, on integral square section and moulded stepped bases, each fitted into the box with a brass pin, the coromandel veneered and satinwood lined casket with brass lock plate stamped BETJEMANNS & SON, MAKERS, LONDON.
_________
The games completing this week’s columns are both absolutely compelling. In the first, fresh off the presses at the WR Chess Masters in Dusseldorf, is the second round victory for our old favourite, Nodirbek ‘Nodi’ Andusattorov vs. Andrey Esipenko which features a kingside storming; in the resigned position, Nodi’s h-pawn will wander up the board to devastating effect.
But even this assault pales into insignificance compared to the sacrificial attack whipped up by that great British Master and master trainer, Bob Wade OBE. Having won Black’s queen, Wade then proceeds to return his own for a mate the following move, in a thrilling Australian championship game from 1945: Wade vs. Shoebridge
Raymond Keene’s latest book “Fifty Shades of Ray: Chess in the year of the Coronavirus”, containing some of his best pieces from TheArticle, is now available from Blackwell’s. His 206th book, Chess in the Year of the King, with a foreword by The Article contributor Patrick Heren, and written in collaboration with former Reuters chess correspondent, Adam Black, is in preparation. It will be published later this year.
A Message from TheArticle
We are the only publication that’s committed to covering every angle. We have an important contribution to make, one that’s needed now more than ever, and we need your help to continue publishing throughout these hard economic times. So please, make a donation.
