Have we been duped by AlphaGo?

[Jhyn]

First, can AlphaGo teach us new things about the fuseki?

This is not completely relevant to the discussion, but this is 2016. We were supposed to have the flying cars already, where is my pro-level go commenting software for (Unix) laptops?

[Mike Novack]

First, can AlphaGo teach us new things about the fuseki?

This is not completely relevant to the discussion, but this is 2016. We were supposed to have the flying cars already, where is my pro-level go commenting software for (Unix) laptops?

a) Because nobody has written it.
b) Because nobody has handed you a run exec complied and linked for running under linux.

OK, I really want to talk about the first part but put the second part there to straighten out a misconception. Software isn't written for particular operating systems, hasn't been for longer than many of you have been alive (yes in the early days, not only operating systems but hardware taken into account, but probably few of you can remember things like spacing instructions on the drum of an IBM 650). When we say "available for Windows but not for linux" we usually mean you can easily get a complied and linked run exec (an executable). Strictly speaking, free software IS "available" provided you are able and willing to do that compile and link yourself or have a friend to do it for you or money to pay somebody to do it for you.

But like I said, the first part. What folks need to understand is that this is TWQ separate things, a program able to play at pro level and a program able to comment on a prod level game. The mistake here is to imagine that that these are closely related but they are not. In other words, shown a given move and being required to say what is good about that move (the comment part) is very different from finding and choosing that move in the first place (the playing part). The mistake is in thinking "of course would have to know what is good about a move to have selected it" in order to play because, for example, the MCTS programs do not << they have determined that it IS the move most likely to win the game, not WHY it is most likely to win the game >> The newer neural net programs (program plus training) are selecting "gut feeling of the distilled experience of the training database" and so again can't say why.

Pretend there were TWO pros involved. One of them is playing the game but cannot communicate with you (maybe doesn't speak English, etc. *) The other is looking at the moves made and explaining to you why that move was made.

* Or for a better metafor, is a severely handicapped savant, plays brilliantly but nobody can tell why.

[Uberdude]

First, can AlphaGo teach us new things about the fuseki?

This is not completely relevant to the discussion, but this is 2016. We were supposed to have the flying cars already, where is my pro-level go commenting software for (Unix) laptops?

I think AlphaGo is already teaching us new things about the fuseki, in that pros are trying idea from it in their games, for example the opening from game 5. I collected some examples in another thread: viewtopic.php?f=13&t=12953. This may just be a fad for a while, not some ground-breaking revolution in opening theory like "you can play 4-4 points, not just 3-4s you know", but these are early days yet. It might not be doling out expertly crafted natural language explanations for its moves (though maybe neural nets will do that in a few years/decades, with the caveat that neural nets have all sorts of complex and mysterious emergent properties and it may recognising patterns and learning concepts that we as human Go players haven't even recognised or named), but I thought JF was a proponent of the learn Go by example method, rather than the RJ style of trying to define and explain and spell-out everything. And as I think Jhyn is alluding to, it's only played 5 games!

[Bill Spight]

humans have an inherent minimum time-scale for decision making, of the order of seconds. Making strategic plans and reading in Go takes seconds or minutes, not milliseconds or nanoseconds. Yes some strong players can read faster than others, maybe at a glance you can read some moves in 1/10 of a second, but not 1/100. There are certain pattern-matchy, intuition, neural net sort of things going on in the brain (even if Robert Jasiek doesn't think so) which might be sub second, and I can play KGS blitz at one second a move and maybe still be dan level (which is kind of amazing, much like neural net only no MCTS bots), but I don't think anyone can play at 1/100 of a second a move, and not only for the physical problem of playing a stone / clicking a mouse.

I don't think that this does justice to the massive parallel processing that goes on in the brain when playing go. It is not just a question of pattern matching. There is real computation going on. There is analysis, there is synthesis, there are feedback loops. Decisions may emerge through some process of competition in the cerebral cortex. OC, we are unaware of nearly all of this. Consider visual processing. The last time I studied any of this, we had discovered that the brain constructs at least 18 different images before synthesizing what we think we see. Consider language processing. We occasionally recognize ambiguities in language, but as we make sense of speech we eliminate hundreds of ambiguities per second, without conscious effort. This is not just a matter of pattern recognition, as we have to choose among a number of patterns that fit what we hear, and construct a meaningful interpretation (usually one) as we go along. By comparison our conscious processing is linear, and painfully slow. But even our parallel processing is slow by comparison with computer processing. We may unconsciously process hundreds of possibilities while the computer is processing hundreds of thousands, or millions. If we weren't better at it, there would be no contest.

More thoughts at This 'n' That. ( viewtopic.php?f=12&t=12327&p=204275#p204275 )

[Bill Spight]

I think AlphaGo is already teaching us new things about the fuseki, in that pros are trying idea from it in their games, for example the opening from game 5. I collected some examples in another thread: viewtopic.php?f=13&t=12953. This may just be a fad for a while, not some ground-breaking revolution in opening theory like "you can play 4-4 points, not just 3-4s you know", but these are early days yet. It might not be doling out expertly crafted natural language explanations for its moves (though maybe neural nets will do that in a few years/decades, with the caveat that neural nets have all sorts of complex and mysterious emergent properties and it may recognising patterns and learning concepts that we as human Go players haven't even recognised or named), but I thought JF was a proponent of the learn Go by example method, rather than the RJ style of trying to define and explain and spell-out everything. And as I think Jhyn is alluding to, it's only played 5 games!

My impression is that not only AlphaGo, but MCTS computer programs tend to have a cosmic style. Perhaps AlphaGo is showing us that that is not just an artifact of evaluation based upon randomized playouts. In any event, it is strong enough to make such plays respectable again.

[Jhyn]

a) Because nobody has written it.

They used the same excuse for flying cars. Quit slacking!

Joking aside (I hope it was clear that I was tongue-in-cheek), I would be perfectly happy with a Crazystone-style comment system : here are the moves where your (estimated) win probability decreased most compared to then engine's choice. See http://www.remi-coulom.fr/CrazyStone/an ... index.html if you don't see what I mean. Of course by "happy" I mean "ready to pay good money".

Maybe I should call that an "analysis" instead of a "commentary"? Anyway it seems to me that building this feature from a working go engine would require less work than a full-fledged human-aimed commentary engine, and it is literally the only feature that I would use (I would have bought Crazystone long ago if it worked on my machine).

[gowan]

There is some evidence that slower games can produce better games. Just watch fast server games and you almost always see bad moves that the player wouldn't make if there were more time. There are numerous examples of pros making errors in byo-yomi, too, so it seems likely to me that a pro would generally make better moves given longer thinking time. I think the term "soba go" isn't describing what makes longer time limits related to better go. The crucial benefit of longer time limits is the opportunity for deeper reading, rather than judging probabilities. Forty years ago a pro commentator on the NHK TV tournament in Japan (sorry I can't remember just which pro) said that TV games produce more fighting games than slower tournament games and gave the reason being that in long-time-limit games the players can pretty much completely read out the fights.

It is obvious that for commercial purposes shorter time limits will attract a larger audience of fans watching the games. When a server is broadcasting a long-time-limit game, such as a Honinbo match game, many people ask questions on the online venue regarding why it is taking so long for moves to be made and how boring it is. Sponsors get more exposure from faster games. When I watch such games on line I like to think along with the players, I don't mind taking an hour to think about what will happen next, there is a lot to be learned that way. Saying one kind of go is "better" than another is a question that is difficult to answer. A response that dodges the question is that better go has fewer or no mistakes, is closer to perfect play; the players know that their moves are the best moves, there is little or no estimation of probabilities.

[Kirby]

gowan, games with fast time limits make mistakes, but so do games with longer time limits. It is biased to point out only the former.

Popular YouTuber Haylee (Hajin Lee) recently pointed out that pros generally agree that pros from today are stronger than pros from the past (who didn't play with the ****-ing so called Mickey Mouse time limits), because today's pros can see mistakes from pros from the past.

It's not a bad thing - it's called human progress.

Despite this, pros still study older games that didn't have fast time limits.

That's because, unlike us, they are not obsessed with arguing about Mickey Mouse - they are interested in actually learning from one another. That's how progress was made in the first place.

[sorin]

The point about Go becoming dominated by younger players due to decreased time limits is very interesting.

Is it true though? If we look at title matches for that top 3 titles in Japan, we can see that the average age of title holders has decreased over time, while the time limits stayed the same pretty much since the 60's or so.
To me, this shows that it is a change in Go training/playing that resulted in this shift towards younger player dominating the professional Go scene, not the change in time limits.

I don't think a 3h game quality is much different than a 9h game quality.

[gowan]

gowan, games with fast time limits make mistakes, but so do games with longer time limits. It is biased to point out only the former.

Popular YouTuber Haylee (Hajin Lee) recently pointed out that pros generally agree that pros from today are stronger than pros from the past (who didn't play with the ****-ing so called Mickey Mouse time limits), because today's pros can see mistakes from pros from the past.

It's not a bad thing - it's called human progress.

Despite this, pros still study older games that didn't have fast time limits.

That's because, unlike us, they are not obsessed with arguing about Mickey Mouse - they are interested in actually learning from one another. That's how progress was made in the first place.

I suppose it really depends on what you want to get from playing go. If the most important thing is winning, making the game very complicated might increase your opponent's chances of making a mistake and this approach might be a reasonable strategy. This approach might be effective with shorter time limits because your opponent might not be able to read out the complications. In longer time limits this approach wouldn't be effective because both players could read out the fight. I would like to see the best possible go played. Of course, mistakes are also made in long time limit games as well as in shorter time limit games, though I'd guess that there are more mistakes in the shorter games. As for modern pros being stronger than the historical greats, I'm not so sure that that can be judged by modern pros seeing mistakes in the older games. Modern pros have the advantage of all the discoveries and work of the older players; if there were a level playing field who knows which would be stronger.

[Kirby]

I suppose it really depends on what you want to get from playing go. If the most important thing is winning, making the game very complicated might increase your opponent's chances of making a mistake and this approach might be a reasonable strategy. This approach might be effective with shorter time limits because your opponent might not be able to read out the complications. In longer time limits this approach wouldn't be effective because both players could read out the fight.

Oversimplification. Having six-hours of time does not mean you can "read out the fight". Your argument works if we are discussing the endgame. In the endgame, the situation can be completely read out. So longer time limits might result in more optimal play.

For middle game situations, your argument is purely speculation.

I would like to see the best possible go played. Of course, mistakes are also made in long time limit games as well as in shorter time limit games, though I'd guess that there are more mistakes in the shorter games.

Yes, you are guessing. I probably agree with you about the endgame. But I don't think your argument holds for other parts of the game.

As for modern pros being stronger than the historical greats, I'm not so sure that that can be judged by modern pros seeing mistakes in the older games. Modern pros have the advantage of all the discoveries and work of the older players; if there were a level playing field who knows which would be stronger.

Yes, modern pros have the advantage of earlier discoveries. That's the point, and it's why the level of Go has progressed. It's not to say that historical greats were not impressive for knowing what they did at their time. But pros are getting stronger as time goes on.

[John Fairbairn]

Kirby: Gowan's remarks about longer time suggesting higher quality are not pure speculation. There have been thousands upon thousands of commentaries. A simple survey will show that longer games tend to come with many more "brilliant move" type comments and far fewer "mistake" type comments than fast games. Even in Korean commentaries. I have done research on this field for ICOB papers.

Furthermore, extra time is not necessarily devoted to reading out fights. In fact, in my experience it is hardly devoted to direct reading at all. It is devoted to evaluation (including but not limited to counting). I have seen at first hand, many times, how pros see very deep lines instantly. Then at the end of each line they stop and think - is that stone really on the right spot given that group on the other side, and stuff like that. One example I recall was being in the pressroom with Ishida during a title match. A move was played in the middle game. By definition he had never seen that position before. Once the move was relayed in he slapped about 40 stones down on the board (no ladders). Paused briefly, picked up about 20 of the stones and slapped 20 more down, paused again, looked at Kato and asked whether A was better than B. (As it happens, I was not amazed by that, because I'd seen it all before. The reason I remember it vividly is because I've never seen anyone pick up so many stones so quickly, mostly in one hand).

This is well supported anecdotally. For example, Go Seigen commenting on why Kitani played so slow. He said it was nothing to do with seeing the moves, as Kitani could see instantly more moves than anyone else. He spent his time comparing the variations. Go's comment in turn is supported by examples of vast depth and complication (yose-kos and the like) that Kitani had apparently seen instantly and which made fellow pros gag.

In fact it is a truism that the strongest players tend to be strongest whatever speed they play at. In chess, Carlsen is the world's highest rated player in classical, rapid and blitz play. In go, the last 8 in the NHK Cup tends to reflect the slate of title winners in the slow major events. It's not the quality of the player that time affects; it is the quality of the game (I have argued that that quality loss is reflected not just in more countable mistakes and fewer countable brilliancies but in a change of style; that's a different debate but it doesn't affect the countability of the number of mistakes).

I am not alone in my views. Even in Korea. As you well know, but seem to choose to ignore, many Korean pros, led by Cho Hun-hyeon, have decried the impact of faster time limits in their domestic game. But too many modern tournaments everywhere are becoming a feeding frenzy instead of a banquet. My own grandchidren prefer KFC fast food over Michelin stars. Love then even more than I love you; so much so I've even been prepared to accept greasy fingers. But when it comes to a question of food alone, I prefer still to sit in a proper restaurant.

Sorin: Time limits in Japan have decreased markedly since the 1960s. Whether that explains the rise of young players fully or at all we can't be sure, because it has also been paralleled by reforms giving young players quicker access to the higher regions of tournaments. But personally I'm sure it's a factor because older players have not adapated to the change in style required. It's interesting to me that very many young Japanese players have visited China and Korea to seek improvement. They seem to feel that the new style cannot be acquired properly at home. You view that the quality of a 3h game is not much different from a 9h seems beside the point. At pro level a game is so finely balanced that one mistake can decide the outcome. So even a tiny difference assessable only in a longer game can have a catastrophic effect.

If I may go back to my point about commentaries. If you presented me with two pro games, one fast and one slow, and asked me to say which is which, I would be quite lost. I rely on the commentaries to tell me. But once I know, I prefer the higher quality game (normally the slower one) simply because it teaches good habits for me, and the game has greater internal consistency. It's not quite true to say that fast, mistake-laden games teach bad habits, because if that style of play is best for modern tournament play, maybe that should be your role model. But in my case I don't play in tournaments and I'm not looking for cheap thrills. In fact I don't find fast games exciting at all. I have watched countless late-night fast games on Baduk TV in Korean hotels as I struggle to overcome jet-lag. I find they are quite a good way of making my eyes glaze over.

[Uberdude]

Kirby: Gowan's remarks about longer time suggesting higher quality are not pure speculation. There have been thousands upon thousands of commentaries. A simple survey will show that longer games tend to come with many more "brilliant move" type comments and far fewer "mistake" type comments than fast games. Even in Korean commentaries. I have done research on this field for ICOB papers.

Interesting, are these long games new ones (as in last 10 years), such as Honinbo / Kisei 8 hour title match games, or older ones like Go Seigen's ten-game matches? My impression (not done any research) from An Younggil's commentaries of the former on gogameguru.com is he doesn't find it any harder to find mistakes in those games, compared to the 2-3 hour or so (do you consider those Mickey Mouse?) games of international/non-Japanese tournaments. But maybe that's just a reflection that the current top Japanese players (except Iyama) aren't so strong.

[dfan]

Some perhaps-useful analogies with chess, which I do know something about:

As John Fairbairn says, the strongest players are the strongest players whatever the time control. Sometimes someone will be particularly strong or weak in blitz, but this generally means something like they are #20 in the world at classical time controls and #5 in blitz.

Longer time controls do make for games of higher quality. This is particularly noticeable in the endgame. Modern games are not adjourned, due to the quality of computer analysis, so must be finished in one session, the result of which is that during the endgame modern players are often "living on the increment" (the equivalent of byo-yomi) and are taking ~30 seconds per move to think instead of ~3 minutes like the players of old.

In general older players do better (relatively) at short games than at long games, because they can use their considerable intuition and don't have to maintain concentration for 5+ hours. It seems that go may be different in this regard?

In contrast to John Fairbairn's experience with go, I like to learn from fast pro chess games more than slow ones. A grandmaster playing a blitz game is using his instinct, and that's instinct that I would like to learn to imitate. If he would play a different move in a slow game, it's likely because in the slow game he's spent long minutes doing a difficult calculation, something that I could not hope to duplicate and that is less likely to be useful in improving my own tournament play.

[Kirby]

Kirby: Gowan's remarks about longer time suggesting higher quality are not pure speculation. There have been thousands upon thousands of commentaries. A simple survey will show that longer games tend to come with many more "brilliant move" type comments and far fewer "mistake" type comments than fast games. Even in Korean commentaries. I have done research on this field for ICOB papers.

Interesting, are these long games new ones (as in last 10 years), such as Honinbo / Kisei 8 hour title match games, or older ones like Go Seigen's ten-game matches? My impression (not done any research) from An Younggil's commentaries of the former on gogameguru.com is he doesn't find it any harder to find mistakes in those games, compared to the 2-3 hour or so (do you consider those Mickey Mouse?) games of international/non-Japanese tournaments. But maybe that's just a reflection that the current top Japanese players (except Iyama) aren't so strong.

This was my impression, too: mistakes can just as easily be found in slower games, too. They might be different types of mistakes, I suppose.

You can call your own opinion "research", but it doesn't make it true - "even in Korea".

FWIW, faster time limits have led to new ways of studying. It's not uncommon to study opening strategies extensively so that significant time isn't needed during the opening. That way, more effort can be spent on the middle game. In some ways, this could lead to a more interesting game, since the core of the brilliance behind the game happens during the middle gamme - where the result is often determined.

Just because you don't spend a significant amount of time at the board while the clock is running doesn't mean you're not spending a lot of time. Top pros study tremendously these days, and surely that helps in making faster and accurate decisions.