AlphaGo vs. AlphaGo: 50 Self-Play Games (May 2017)

[John Fairbairn]

Yes, as I discovered a few years ago, there is at least one actual proverb along those lines. One I can loosely translate as, "If you don't know where to play (locally), play somewhere else."

You both might be thinking of a proverb along the lines of "If a local play is worth 15 points, tenuki". It surfaces in different forms, with a different number of points, and has been around for at least 50 years. It appears to be a western concoction, and the likely origin is Russia (or CCCP as it then was). It appears in David Mitchell's book on proverbs.

[apetresc]

Any chance we can put the large number of embedded sgfs behind hide tags? This page stops rendering halfway through on my iPhone, and I have a hunch that's the reason. It's slow but functional on my laptop.

Yes, fixed that – good call

[Bill Spight]

Yes, as I discovered a few years ago, there is at least one actual proverb along those lines. One I can loosely translate as, "If you don't know where to play (locally), play somewhere else."

You both might be thinking of a proverb along the lines of "If a local play is worth 15 points, tenuki". It surfaces in different forms, with a different number of points, and has been around for at least 50 years. It appears to be a western concoction, and the likely origin is Russia (or CCCP as it then was). It appears in David Mitchell's book on proverbs.

I saw it in one of Go Seigen's books a few years ago. Not worth tracking down now, I don't think. What I loosely translated as "don't know" was 分からない, as I recall.

Ah! I found a reference to it on Sensei's Library. Here it is:

分からない時は手を抜け

at http://www5.plala.or.jp/hasebehp/igk/igk_unyou.htm#f

[palapiku]

I wonder why these games feel so much more "bot-like" than the mostly-human feeling of the moves AlphaGo played against Ke Jie and Lee Sedol.

Because as soon as Ke Jie plays enough inefficient moves in the opening, Alphago takes advantage of them. From that point Alphago is in the lead, and is interested in solidifying and simplifying the situation on the board, so it plays natural-looking, simple moves instead of the extremely sharp, tactical probes it plays when it feels behind.

[John Fairbairn]

Ah! I found a reference to it on Sensei's Library. Here it is:

分からない時は手を抜け

Not quite the same thing, though, is it, Bill? "Te wo nuku" is part of the normal language for missing something out - usually something that is actually needed (and so often has a bad sense, as in "cut corners"). Here the usual meaning would be to defer a move until you know better how to continue, but the implication is that you do still have to continue locally. The "toki" also tells us that this is a "whenever" situation not an optional "if."

The specific go usage is almost always "tenuki (wo) suru" (with derivatives tenuki sareru, tenuki dekiru etc). The "te wo nuku" form can occasionally be used (e.g. for elegant variation, when you have ignored one move by the opponent and then ignore another, you might say "sara ni te wo nukimasu" (skip yet another move), though even then you can perfectly well say "sara ni tenuki suru"). But more than that, tenuki is (?) always defined in Japanese as deliberately not responding to an opponent's move even when there is a perceived need. It is an optional, conscious decision where you do know what you are doing. You have gone elsewhere and, just like Katie, you are satisfied with your peanuts and crackerjack and don't care if you never get back. Accordingly, the equivalent Japanese proverb is 手抜きは最高の手段. (Tenuki is the supreme resource) - which is much more like your "tenuki is always an option."

[Bill Spight]

Thanks, John.

Suffice it to say that we agree that there are subtleties, but we disagree about the pragmatics of language. As for 手抜きは最高の手段 I think that overstates my proposed saying. It reminds me of the contract bridge saying, in British English, No bid is the best bid. 'No bid' being the British version of 'pass'. The great bridge writer, Victor Mollo, once joked that small children should be made to stand in front of a mirror for hours per day, repeating, "No bid."

[Kirby]

Ah! I found a reference to it on Sensei's Library. Here it is:

分からない時は手を抜け

Not quite the same thing, though, is it, Bill? "Te wo nuku" is part of the normal language for missing something out - usually something that is actually needed (and so often has a bad sense, as in "cut corners"). Here the usual meaning would be to defer a move until you know better how to continue, but the implication is that you do still have to continue locally. The "toki" also tells us that this is a "whenever" situation not an optional "if."

The specific go usage is almost always "tenuki (wo) suru" (with derivatives tenuki sareru, tenuki dekiru etc). The "te wo nuku" form can occasionally be used (e.g. for elegant variation, when you have ignored one move by the opponent and then ignore another, you might say "sara ni te wo nukimasu" (skip yet another move), though even then you can perfectly well say "sara ni tenuki suru"). But more than that, tenuki is (?) always defined in Japanese as deliberately not responding to an opponent's move even when there is a perceived need. It is an optional, conscious decision where you do know what you are doing. You have gone elsewhere and, just like Katie, you are satisfied with your peanuts and crackerjack and don't care if you never get back. Accordingly, the equivalent Japanese proverb is 手抜きは最高の手段. (Tenuki is the supreme resource) - which is much more like your "tenuki is always an option."

I don't quite get the distinction being made here. Are you saying that the meaning is different since Bill's quote used "手を抜け" instead of "手抜き"? Or maybe that "手抜き" is always a conscious decision?

Not sure how much you can trust a wikipedia article, but there's some description of tenuki here:

一般に、戦いの最中や、大きな欠陥を残すような場面で手を抜くのはよくない。しかし手を抜いても大きな損害が出ないような場合なら、思い切って手を抜き、大場に先着するのも重要な戦法である。また、完全に形を決めてしまわず、後に味や含みを残しておくために手を抜くこともある。

Roughly, I'd translate this as: Generally, in the middle of a fight, or in the case where you're leaving a big defect, "手を抜くのはよくない". But if it's the case where there won't be a lot of damage, resolutely "手を抜き", playing first in a big area is an important tactic. Also, doing "手を抜く", not completely deciding on a shape, for the purpose of leaving aji or other implications for later also sometimes happens.

In the description here, they used "手を抜く" and "手を抜き".

I never imagined much difference between "手を抜く", "手を抜き", or "手抜き". Is there an implication that "手抜き" always happens consciously when one is aware of the situation? Or is this just how pros typically use the term?

Is it possible for amateurs to "手抜き" without knowing what the heck is going on?

[John Fairbairn]

In the same way that all professions have their own jargon, go pros use tenuki suru as the much preferred technical term rather than te wo nuku. When they don't, there is typically a reason that we should try to be sensitive to. I have outlined a couple of scenarios above, but there are bound to be others. Of course, it may all just depend on what side of the bed the writer got out of, but it is my experience that there usually is a valid reason, just as in the example above we can detect a different emphasis - skipping a move (but actually only deferring it) because we don't know what to do yet and skipping a move because we think we know exactly what we are doing - we have found something bigger.

You have to make up your own mind how careful a reader you want to be, just as in go you can read one move ahead or 20. In go I'm a one-move man but in my professional life, dealing with legal documents and other crafted documents, I have had to be not only a 20-move man but also to read between the lines. And that's all without mentioning the joys of linguistic analysis and explications de texte (my favourite task in school). So it's second nature for me to look at go texts carefully (but maybe not the diagrams!).

It's over-refined for some people, I know, but I think this approach is paying off in establishing the proper meaning of thickness, for example.

[Kirby]

Is your view of the nuance of tenuki solely based on how you've seen pros use the term in the past? If so, I guess I'll just have to take your word for it, since there's no way to really discuss further.

[ez4u]

The "proverb" under discussion sounds like an amateur one to me. I watch a lot of television commentaries (well, three a week anyway from the NHK and Ryusei). The issue of not knowing what to do just does not arise. From the pro point of view the problem is that the available local plays are known but all lead to poor results. In other words the player has been outmaneuvered locally. In that case, tenuki is the only appropriate choice. The player hopes to change the overall state of the board so that one of the rejected local plays becomes favorable in that overall context. I don't think a pro would describe this in terms of not knowing what to do.

Many amateurs like me settle for making the best of a bad bargain locally (in gote more often than not!) and wonder why we don't get stronger. Getting to pro involves getting over that mindset among many other things.

[ez4u]

Meanwhile on the original topic of the release of 50 self-played games (out of a hundred million or so). I can not help but think of one of Khalil Gibran's aphorisms in Sand and Foam, "How mean am I when life gives me gold and I give you silver, and yet I deem myself generous."

AlphaGo forever!
DeepMind and Google, not so much.

[Uberdude]

Meanwhile on the original topic of the release of 50 self-played games (out of a hundred million or so).

They said these 50 were self-play with long time limits so presumably stronger play, whereas the millions of games used for training the networks were faster and thus potentially weaker and embarrassing. Nevertheless, it sure would have been nice to see some 3 stone handicap games between AlphaGo Lee and Master versions (Aja confirmed on facebook they did actually play with 3 stones, it wasn't a proxy for win percentage).

P.S. Inseong Hwang 8d has made a nice video lecture taking a quick look at some of the interesting moves in these games, particularly a lot of attachments: https://www.youtube.com/watch?v=iO_JmGH8Iu8

[ez4u]

Ask yourself how likely it is that the program played a hundred million silly, embarrassing games against itself and then suddenly transcended to the AlphaGo that we see today.

Alternatively, consider the early 3-3 invasion. How many thousands (millions?) of games must they be holding that illustrate all the alternatives that both sides have tried both for and against the play? Not to mention any indication of what it did not try!

Thx, I'll have to have a look at Inseong's videos.

[Uberdude]

Ask yourself how likely it is that the program played a hundred million silly, embarrassing games against itself and then suddenly transcended to the AlphaGo that we see today.

Yeah, I don't really buy that argument myself

[lightvector]

A good rule of thumb from computer chess I seem to recall is 30 Elo points per doubling of computing power, although as you get closer to the top I think that shrinks a bit. I don't actually know whether that's exactly right any more, and for AlphaGo this is heavily complicated by the fact that you need both CPU and GPU and whichever one is more constrained becomes a limiting factor, but anyways 30 Elo points per doubling looks very roughly consistent with Deepmind's original Nature paper showing how earlier versions of AlphaGo scaled with computation.

If the newer versions of AlphaGo don't scale terribly differently, then ignoring things like fixed-per-search overheads and other issues with this kind of extraplation, back-of-envelope calculations suggest that AlphaGo at 1 second/move should be still at a minimum at the level a top pro who has a normal-slow-game-amount of thinking time, and actually likely still a bit beyond it, but possibly not enough to win every time any more against a human who has very long thinking time.

(Evidence from the master games, even taking into account the fact that they were blitz and therefore disadvantageous for humans, puts a pretty confident floor on Master's possible single-machine Elo rating of about 3800. It's more likely a couple hundred points higher than this absolute floor, and coupled with improvement since then the Master games, you can easily drop a few hundred rating points in the process of restricting to 1s/move and still end up higher than top pros like Ke Jie around 3600 - https://www.goratings.org/en/).