Life In 19x19

We in the West have enjoyed the same live coverage of the AlphaGo-Lee Sedol match that the East has enjoyed. But with millions of fans and hundreds of pros sharing their viewpoints, analysis there has gone into greater detail. Jennie Shen 2P found this article particularly interesting. It was written by Li Zhi 6P, currently ranked 30th in China. It was translated by Chun Sun with help from Michael Chen and Tong Yi. Enjoy! (More to come...)

Game 1 - Lee Sedol’s Strategy and AlphaGo’s weakness

Super interesting - thanks Brady.

I wonder if this is the same pro who left the pro scene to go to grad school in philosophy. Sounds like it could be him from the way he writes.

Also, his translated post from round 2 is up, in which he mentions (dismisses) the chatter about Lee Sedol secretly agreeing to not start ko fights. Can't wait for his post on round 3, and for other news as to whether such conspiracies still exist.

mhlepore wrote:I wonder if this is the same pro who left the pro scene to go to grad school in philosophy. Sounds like it could be him from the way he writes.

Yes, Li Zhe (李喆) went to Peking University to study philosophy in 2012. (Incidentally, the Chinese wiki page mentions that the chess grandmaster and former Women's World Chess Champion Hou Yifan went to Peking University to study International Relations in the same year)

Great, insightful article. I was pleased to see Li Zhe said the same as I had with regards to black 7 becoming inefficient, and that 3-3 was better than iron pillar to protect the top left corner, though he didn't explain that one.

In the article "[Game 1] Lee Sedol’s Strategy and AlphaGo’s weakness" on
https://massgoblog.wordpress.com/2016/0 ... -weakness/
Zhe Li makes the following mistakes in the commentary on game 1:

1) It is premature to conclude that Black 7 is refuted by White 8 + 10 because Li does not discuss alternative black moves 9 to 15.

2) Li speaks of "open and complex board positions" in a too general manner and, as a side effect, draws a wrong conclusion. Instead - for the context of Li's discussion -, non-local positions must be classified into at least two types: a) a complex local fight involving strings on the open outside (such as the game 1 opening fight on the upper side); b) a complex fight involving large scale to global interaction between different parts of the board (such as the game 4 fight around move 78). By confusing the two types, Li makes the wrong judgement that, in game 1, Lee would have managed to create a global, long-term fight, while it should be called a large local fight, whose tactical complexity is mostly confined locally and whose global judgement can be done at quiet positions after the fight cools down. AlphaGo won game 1 because it could handle the fight as an, although large and tactically complex, local fight. I had expected this from AlphaGo, so Li's statement "nobody expected this" also is wrong.

3) According to Li, Lee knew what a good response to the AI would be. I disagree. During the first three games, Lee had no clue. Only in game 4, Lee had become aware that his play must be with global and long-term interaction.

4) Li speaks of human intuition without clearly relating AlphaGo's calculations to justify the use of the phrase.

5) Li writes that, in a closed area, AI could exhaust all possible moves to complete the calculation (brute force). Especially for AlphaGo, this is wrong because MC / NN algorithms are designed to NOT exhaust all possible moves. Instead, MC / NN rely on sampling and pruning.

6) Li claims that the ability to evaluate positions was the most difficult part of the game to master. It is the easiest part of the game to master because a) for quiet positions reading is relatively very simple and b) for fighting positions the difficult part is not the judgement but is the reading (i.e., another part of the game to master) and nevertheless necessary reading for the sake of judgement is simpler than usual reading because the reading task is restricted by conditions of what / when / how to judge. Where, in a closely related context, Li writes that [it] was the most difficult [...] for professional players, he should not learn from the unnecessary difficulty of professionals unaware of powerful judgement theory but he should study such theory.

Here are Li Zhe's thoughts on the second game. I am told there is more to come.

FOR THE FIRST TWO GAMES, NOBODY WOULD HAVE DONE A BETTER JOB THAN LEE SEDOL!

RobertJasiek wrote:2) Li speaks of "open and complex board positions" in a too general manner and, as a side effect, draws a wrong conclusion.

I'm not going address the critical judgement, but I do think the point of terminology is worth hashing out.

In the article a notorious dynamical lepidopterist is invoked. The Wikipedia article on the "butterfly effect" say this was an unnamed meteorologist, invoked by Edward Lorenz (not Mandelbrot, as I and I suppose others lazily thought). In any case, Li says he is talking about fights where a small change makes for largely divergent trajectories of the resulting development.

In other words, for mathematicians, he is talking about a "hyperbolic" situation.

RobertJasiek wrote:Instead - for the context of Li's discussion -, non-local positions must be classified into at least two types: a) a complex local fight involving strings on the open outside (such as the game 1 opening fight on the upper side); b) a complex fight involving large scale to global interaction between different parts of the board (such as the game 4 fight around move 78). By confusing the two types, Li makes the wrong judgement that, in game 1, Lee would have managed to create a global, long-term fight, while it should be called a large local fight, whose tactical complexity is mostly confined locally and whose global judgement can be done at quiet positions after the fight cools down.

I'm going to pass on the matter under debate, not being strong enough to have an opinion. I note though that Robert is making a distinction in an ontology of fights. My current work on Wikidata makes me familiar with co-existing ontologies. Robert's (a) is roughly "open system", and (b) roughly "coupled system".

Let me get my own contribution in now. As a somewhat faux mathematician these days, I think I can still point out that hyperbolic belongs as an adjective in a family with parabolic and elliptic. Or, in another sense, with Euclidean and elliptic, for geometries. In any case there is a clear antonym, "elliptic", and a boundary case.

We know, roughly, what the antonym to "running fight" is, namely a localised fight that will reach one of a definable list of outcomes (i.e. has a status once you know who will play first). So, e.g. classic life-or-death for a group, capturing race, to name the ones every player has to learn. Or, indeed, come down to endgame plays, though there are some weasel words around that.

So, the parabolic case might be the bone of contention? We know some running fights can be described, for example in joseki dictionary shorthand, as "both players jump out". It is quite possible that this quasi-evaluation is an artefact, created for example by looking at one corner of the board and arbitarily "ruling out" coupling to other areas, and other fights, on the board now or implicit in frameworks.

It is also quite possible that modern go theory has got beyond this bit of ontology.

"Neither player needs to add a stone to the running fight, for the time being." If playing out the fight further has foreseeable consequences, this might stand up. The game will flow in laminar rather than turbulent fashion, to introduce another metaphor.

The commentary Game 2a is good, mostly except that I disagree with the conclusion that Lee was the best possible player for the match.

RobertJasiek wrote:The commentary Game 2a is good, mostly except that I disagree with the conclusion that Lee was the best possible player for the match.

If Lee Seedol isn't the best person for the match, who is?

The person must be very good at opening, positional judgement and reading. I am not sure who qualifies but somebody said that Ke Jie would meet these criteria.

Here is the translation of Li Zhe's third passionate post on the AlphaGo series.

I've enjoyed all three articles very much. I love this quote from the most recent:

For me, the most astounding part comes from the game records. AlphaGo played a lot of moves outside of human experience, moves that weren’t understood by many professional players. However, I start to see a spectacular world beyond these moves, a world that was not created by AlphaGo; a world that has been hiding inside of the game of Go itself for thousands of years.

Thanks to Chun Sun 5d, Michael Chen 7d, Yi Tong 3d for their effort translating this passionate pro! Hopefully there is more to come.

A great fraction of moves surprising apparently quite a few professionals has not surprised me because I am used to flexible moves or moves emphasising the center. E.g., game 5 move 70 was the first move I considered. Nowadays, there are too many professionals emphasising side territory over center influence; they should be more flexible even if they do not become as extreme as Takemiya.

RobertJasiek wrote:A great fraction of moves surprising apparently quite a few professionals has not surprised me because I am used to flexible moves or moves emphasising the center.

However no one is surprised when a weaker amateur makes strange moves. When professional level bots play unusual moves, it is more worth the discussion.

A great fraction of moves surprising apparently quite a few professionals has not surprised me because I am used to flexible moves or moves emphasising the center.

Then why haven't you become a professional? If you're so much superior at flexibility, and have better positional judgment than Michael Redmond, surely you should have reached 9p by now.