Appreciating Famous Games

by Shuzo Ohira, John Fairbairn (Translator)

Genres: Go

282 pages, Paperback

First published April 1, 1998

About the author

Shuzo Ohira (大平修三) was a Go master, rank 9.

Reviews

[Manny]

[Original review, Feb 7 2016]

About a month from now, it's possible that the most important Go match in history will take place in Seoul, Korea. The two players will be Lee Sedol, widely regarded as the best Go player in the world, and AlphaGo, the breakthrough AI program recently announced by Google's Deep Mind group in London. The prize will be $1M.

There is a speculation about this unique event all over the web. The most interesting pages I've found so far are the Deep Mind paper in Nature (I think this actually isn't supposed to be freely available, so the link may break soon), and the GoGameGuru match site. Both of them are pretty uncompromising, but there is some fascinating stuff there. In particular, I've been looking at a long video on the GGG site in which Myungwan Kim, a top Korean Go professional, gives a very detailed and insightful breakdown of AlphaGo's strengths and weaknesses.

It seems clear from Kim's analysis that the version of AlphaGo which in October 2015 won the already famous match against Fan Hui, the European Go champion, would have poor chances of beating a truly world-class player. It made several mistakes which at that level were close to outright blunders. But Deep Mind will have had another five months of development time to improve it. Also, Google must be very confident if they're rushing things and holding the match so early and in such a high-profile way.

So like everyone else, I have no idea what's going to happen! But there was a period of a couple of years back in the 80s when I was very keen on Go, and now I'm dusting off my old Go books and trying to remember what this amazing game is about. I want to be able to follow the match as well as I can, and I'm thinking that Appreciating Famous Games, available for free download here, might well be part of my training program...
___________________
[Update, Mar 9 2016]

Oh wow. AlphaGo, clearly very much improved after the Fan Hui match and with a completely revised, ultra-aggressive style, just won the first game. Historic moment.
___________________
[Update, Mar 10 2016]

AlphaGo won the second game too, very competently getting Lee Sedol into time trouble and then squeezing him - a technique also familiar to chess players.

This is starting to feel a little scary. Though hey, people are still better at writing poetry.
___________________
[Update, Mar 12 2016]

And AlphaGo wins the match 3-0.

There was a really weird and impressive moment near the end. The machine was ahead a long way on territory and Lee Sedol, Black, played a last-ditch invasion; the whole game hung on whether he could make his invading group live or not. A complicated fight arose. The commentators thought Black's group was dead, but they weren't 100% sure. Suddenly, the machine ignored its opponent's latest stone and played in a completely different part of the board to gain some extra points; it had calculated that the invasion was already dead even if it gave Black an extra move, sort of the Go equivalent of "I can whip you with one hand tied behind my back". The commentators, who had been looking away while they analysed a variation, were so surprised that it took them a good twenty seconds even to figure out where the machine's last move had been.
___________________
[Update, later on Mar 12 2016]

The comments on the GoGameGuru site are rather striking:

With his back against the wall, and armed with the results of his research, Lee took advantage of the first move by playing a fast paced and active High Chinese formation. Things seemed to be going as planned when Black created a wide position up to Black 11 and AlphaGo entered Black’s large moyo with White 12.

Lee responded by denying White a base with Black 13, and the game became exciting. It was the first time we’d seen AlphaGo forced to manage a weak group within its opponent’s sphere of influence. Perhaps this would prove to be a weakness? This, however, was where things began to get scary.

Usually developing a large sphere of influence and enticing your opponent to invade it is a good strategy, because it creates a situation where you have numerical advantage and can attack severely. In military texts, this is sometimes referred to as ‘force ratio’. The intention in Go though is not to kill, but to consolidate territory and gain advantages elsewhere while the opponent struggles to defend themselves.

Lee appeared to be off to a good start with this plan, pressuring White’s invading group from all directions and forcing it to squirm uncomfortably. But as the battle progressed, White gradually turned the tables — compounding small efficiencies here and there. Lee seemed to be playing well, but somehow the computer was playing even better. In forcing AlphaGo to withstand a very severe, one-sided attack, Lee revealed its hitherto undetected power.

Move after move was exchanged and it became apparent that Lee wasn’t gaining enough profit from his attack. By move 32, it was unclear who was attacking whom, and by 48 Lee was desperately fending off White’s powerful counter-attack.

I can only speak for myself here, but as I watched the game unfold and the realization of what was happening dawned on me, I felt physically unwell. Generally I avoid this sort of personal commentary, but this game was just so disquieting. I say this as someone who is quite interested in AI and who has been looking forward to the match since it was announced.

___________________
[Update, Mar 13 2016]

Lee Sedol came back and won the fourth game, but I'm looking forward to seeing detailed post-match analysis. The commentators again had trouble following what was going on. Late in the middle game, they were almost sure AlphaGo was winning, then something happened. Perhaps Lee Sedol, despite being in severe time-trouble, found a brilliant combination which made his desperate invasion of AlphaGo's territory work and turned the tables. But it's impossible not to wonder at least for a moment whether the Deep Mind team, who had already won the match, didn't throw the game. The machine played some really odd moves which looked plain bad and had the commentators shaking their heads.

Well, I'm sure we'll know more soon.
___________________
[Update, later on Mar 13 2016]

Oh, what a relief! The experts have examined game 4 carefully and given it the thumbs-up. Lee Sedol just played very well. He'd thought carefully about what strategy to use, and decided that his chances would be best if he went "all-in" and staked the whole game on a single tactical fight. He let AlphaGo create a big territory, then invaded it. If his invasion lived, he would win, and if it died, he would lose, simple as that.

The YouTube commentators thought his invading stones were dead, but despite having almost no time left Lee Sedol found a move so brilliant that they couldn't even see the point after he'd played it. Top Chinese player Gu Li, who was watching, said he'd missed it too and called it "the hand of God". So AlphaGo is indeed still vulnerable to human insight and cleverness.

[Full GoGameGuru writeup here}
___________________
[Update, Mar 16 2016]

Well, I can see that people are going to be writing about this match for years, most likely decades, to come. Now that the experts have had time to do more analysis, it seems that the comforting picture from the post above isn't actually correct. Lee Sedol's move 78 in game 4 was indeed a brilliant resource, but it wouldn't have saved him if AlphaGo had taken its time, rather than answering quickly. It seems (today, anyhow) that it still had a clever way to preserve a small advantage and win. Needless to say, the analysis is so complicated that I'm only repeating what I heard in the YouTube video. But if you're a conspiracy theorist who believes Google, Deep Mind or the computer threw the game, you can get back to work.

And so to the fifth and final game, in which the machine again both showed its class and confused the hell out of the millions of online spectators. It started by getting into a tactical fight in the lower right-hand corner and missing a relatively obvious tactical trick which gave Lee Sedol a substantial advantage - trying to translate into chess terms, it felt like winning an exchange for minimal compensation. But the machine made the most of that compensation, and after its initial accident slowly played itself back into the game. What happened next is something I've seen hundreds of times in chess matches where a top player is managing a bad position against a weaker opponent. It carried on accumulating small advantages, got the human into time trouble, and ground him down in the ending, finally winning by a couple of points.

There's no reasonable doubt about it any more. The AIs are now better at this game too.
___________________
[Update, Mar 18 2016]



The Korean Baduk (Go) Association presented AlphaGo with an honorary 9-dan certificate. As far as I know, this is the first time a piece of software has received an official award normally given to human beings.
___________________
[Update, Jan 6 2017]

AlphaGo is back. It's been playing anonymously on the web against top players and winning all its games.

If you know something about Go, look through the game given on the page above. It's impressive and also rather scary to see how the machine smashes the world #1 with its inhumanly aggressive style.
___________________
[Update, June 3 2017]

Last week, AlphaGo finally played its match against Ke Jie, the current world #1, and easily beat him in all three games. The most impressive moment was perhaps near the end of the first game. AlphaGo played a clever tactic which the commentators had missed. They told us what the continuation would be: AlphaGo's play won several points, though there was a remote chance that it would allow Ke Jie a desperate counterattack. But then the machine surprised them a second time. It chose a different line, refusing the points it had apparently won but also blocking off the counterattack. The commentators shook their heads in admiration. They'd seen it do this before. It doesn't choose the play which is going to give the biggest winning margin, but rather the one which maximizes the chance of victory. In the end, it won by half a point, the least amount possible, but it never gave Ke Jie the slightest chance to fight back. It must have calculated all this when it turned down the chance to take the extra points earlier.

In chess terms, a rough translation would be that it had refused to win a piece, giving the opponent a remote chance at a sacrificial attack which would almost certainly not work, and instead chosen to play a rook ending where it could be quite sure of winning by one tempo, twenty moves later. We're in Korea at the moment, and I had a chance to discuss the match briefly with an amateur 6-dan. He said he thought AlphaGo was "one and a half stones better than Ke Jie, maybe even two stones". If he's right, it could have offered the world champion a small handicap and still have had a decent chance of winning.

The Singularity is getting closer.
___________________
[Update, June 24 2017]

If you know something about Go, check out "The future of Go: AlphaGo versus AlphaGo games". The machine isn't just an amazing tactician. It's worked out completely new strategies that the top players didn't even suspect might exist.
___________________
[Update, July 2 2017]

If you search for "AlphaGo versus AlphaGo", you find what looks like an exponentially proliferating set of pages. The whole Go world has thrown itself over the 50 games which Deep Mind released, and they have been studying them as though they were the Word of God. A particularly striking quote from the download page:

Shi Yue, 9 Dan Professional and World Champion said the games were “Like nothing I’ve ever seen before - they’re how I imagine games from far in the future.”

___________________
[Update, Dec 30 2017]

I got chatting today with a young Chinese-American 6 dan who's written a book about the AlphaGo versus AlphaGo games. So far only available in Chinese, but he says the English translation will be ready soon.
___________________
[Update, January 11 2018]

I've been watching Michael Redmond 9p's video commentaries on the AlphaZero-AlphaMaster games (there are links on the relevant AlphaGo games page). I have to hand it to Redmond: when I hear his explanations, I have the illusion that I can follow what's going on, which definitely isn't the case when I play through the games on my own.

It's amazing to see how quickly the Go world is absorbing this new information. I wonder if the top players' ratings are going to start going up as a result? That's what happened when computers became good enough to beat the top chess players.
___________________
[Update, August 26 2018]

I just watched another of Redmond's excellent videos on the AlphaZero-AlphaMaster games. (He says these are being turned into a book, though I don't think it's out yet). Redmond has spent a huge amount of time analysing the game, which AlphaZero won after a fantastically complicated fight. Chris Garlock, who's acting as the straight man, asks him at the end what Master did wrong. "I can't find any mistakes," says Redmond. "And you've looked!" says Garlock. "Yeah," says Redmond with a wry smile.

[Jon · Rating 3 out of 5]

Great set of 1800s games that aren't too difficult for early learners. Good commentary. Would recommend