Are pros being underestimated?

[Kirby]

Pros recommended a lot of things when people ask about getting stronger. Sure, replaying pro games is one of them. But life & death, playing games, and reviewing are also popular recommendations.

Practicing what you actually want to get good at has got to have great practical value.

Pros also have some incentive to recommend methods of study that require the existence of pros. After all, their fans are what keep them in business.

[CDavis7M]

Michael Redmond has Go lectures on YouTube. He is recently doing English lessons on proverbs which are sort of a modern refresh of his old NHK lecture series on proverbs.

My favorite parts of these lectures are when he mentions the moves that AI would play and how he does not recommend such moves to amateurs.

Splitting the side: AI will not play it in the opening (maybe some exceptions), but it's a valid play even at the pro level.

That's great. Because I can understand and use that move.

[pwaldron]

My favorite parts of these lectures are when he mentions the moves that AI would play and how he does not recommend such moves to amateurs.

Good point. When you look at what the AI there's usually not much difference between any of a number of plausible moves--less than a point by KataGo's estimate. I'd happily give up a point or two to get a game that I'm comfortable with. After all, I'm going to lose 50 points through the middle game.

[Kirby]

There's also something to be said regarding the type of game you are playing. Even before AI, sometimes I would see some joseki pattern and not totally understand it.

I'll actually use a pattern that became popular after AI as an example:

$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X X . O . . . . . . . . . . . . . |
$$ | . . O X . . . . . , . . . . . , . . . |
$$ | . . O a . . . . . . . . . . . . . . . |
$$ | . . O X X . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |

Click Here To Show Diagram Code

[go]$$Bc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X X . O . . . . . . . . . . . . . |
$$ | . . O X . . . . . , . . . . . , . . . |
$$ | . . O a . . . . . . . . . . . . . . . |
$$ | . . O X X . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |[/go]

I saw this pattern for awhile, always feeling uncomfortable about the cutting point at 'a'. I didn't spend enough time thinking about/studying it, and played the pattern in a tournament. Given that I was in a tournament, I started worrying about that cut again, and played something weird, because I didn't understand the shape well. I lost the game, and regret not studying the pattern before.

Why did I play a pattern that I knew I didn't understand well??

On the other hand, if I am playing a non-tournament game for learning, pushing the boundary of what I understand and trying to play something that I know is probably correct - even if I am not an expert on it yet - it would seem to be a good way to learn. My opponent may play in a way that I didn't think of, and it's a good opportunity to learn and understand *why* it's a good shape.

In the same way, I believe that there are a lot of good AI moves that I don't understand, yet. Trying them out and experimenting with them during my games is a good way to get a feeling for how that pattern develops and/or what kind of shapes result.

But if it's a tournament game that I want to win, it's probably best to stay in comfortable territory.

[pwaldron]

In the chart, I plotted moves outside my top 5 as a score of 7 (the lower the better match I have with AI).

Qualitatively the graph looks different than a professional's. John noted in his "First Teenaged Meijin" that in the last game, the AI largely agreed with the pros. The agreement tended to be "clumpy"--they played sequences that the AI agreed with, but then would make a mistake in direction. Once the new direction was selected, though, they seemed to do quite well again.

I am curious what the longer-term effects of the AI on professional play with be. Many of us remember the new-fangled idea of playing online go, but it was about 10-15 years before those players who grew up on online go became professionals. Pretty soon we'll see a new generation who have always had an AI available to them.

[Ferran]

Human pros are extremely strong. Stronger than amateurs can fathom. Borrowing from another sports, I recently saw a top tennis coach, who attained professional level himself, address the question whether a high ranked amateur would be able to beat John McEnroe, given his old age and dito style.

I saw a bit of a match, once, between a resurfaced Navratilova and a young promise in the top... 5? Graf? Can't recall. Thing is, Navratilova's precision and reading had the younger player an inch from throwing her guts, teenage power and all.

Take care

[Mike Novack]

John noted in his "First Teenaged Meijin" that in the last game, the AI largely agreed with the pros. The agreement tended to be "clumpy"--they played sequences that the AI agreed with, but then would make a mistake in direction. Once the new direction was selected, though, they seemed to do quite well again.

What you are saying is that "direction A is OBJECTIVELY better than direction B". And you need to look at the percentages at the end of each best play line, not just after the first move.

But the pro might NOT be making a "mistake" because a real game between two particular pros is not purely objective. In the real game between two particular pros there is a matter of the current state of the game (who is ahead) and whether the direction A or the direction B best fits the style and strengths of each player.

Thus if player 1 is slightly behind, if direction A results in a simple best sequence but direction B leads to a complex sequence where there are lots of alternatives, hard to pick out the best sequence, more chance to go wrong, then B might be the best PRACTICAL choice. Or perhaps player 1 is a "fighter", player 2 is not, and direction A is "quiet" compared to direction B.

When rating the two directions, AI is assuming it is playing against an opponent equal to itself and is not allowing for "which direction offers most chances for a mistake". .

[CDavis7M]

Found this topic again because I was thinking more about how well human pros do in matching AI plays/recommendations, and I was also thinking about a (non-pro) human commentary technique I've seen, which I sort of don't like, and then wondering how often I see pros do this...

I don't like when commentary points to a "mistake" according to the AI. I do like when a mistake is pointed out and an explanation is given as to how it was refuted or why the game result started to turn sour because of that play, but all that could be done without saying things likes like "this move lost a few points." It's just that this type of description doesn't tell me much about the game itself (as others can have described better here on L19 and elsewhere).

And this seems to be a shift in the perspective of commentary. The player made a mistake according to the AI analysis. Not that their opponent took the initiative. Of course, game commentaries always liked to point out moves that were slow, or which were the losing move, but I feel like this tendency to phrase the move in the context of a "mistake" seems to be growing. Most mistakes hardly seem like an actual "mistake" in judgment.

I've already said that I don't prefer this commentary style. But I don't think it's a bad thing and I wonder if others like it. It is good to have an "expert" opinion on the moves. But I just appreciate the discussion of how the other player seized the opportunity, not identifying how many points the mistake cost.

[John Fairbairn]

I was also thinking about a (non-pro) human commentary technique I've seen, which I sort of don't like, and then wondering how often I see pros do this...

I think your instincts are spot on here. I have thought a lot about the art of commentary recently because it is what is at the heart of my Museum of Go Theory project. One thing that struck me about the latest work in this series, the Yi Mo of 1662 (not yet issued), is that the commentator (Ji Xinxue, a top master) makes sure he praises both sides if possible, and his praise is frequent. And the same goes for criticism of mistakes. But more importantly, he strives to let the reader know what is going on in the game.

I find this is typical of pro commentaries in general. It is the amateur commentator who harps on mistakes. Probably this is because mistakes are easier to spot (especially now, if you have a bot handy).

I agree that we need to know when a mistake is a mistake, but it should be like seasoning. We need salt on our food, but a plateful of salt is no substitute for a plate of roast beef, Yorkshire pudding, two veg and gravy.

Yet another way of looking at it is journalism, which is after all just a form of commenting in the world. A typical news story tells us what, who, when and where about an event. Valuable information - but how do we process it? Probably just through our existing prejudices. We may be more knowledgeable, but no wiser. A discursive feature, however, adds "why" and "how" to the mix. That's far, far more valuable.

I think the message is: stick with the pros.

[pajaro]

I don't like when commentary points to a "mistake" according to the AI. I do like when a mistake is pointed out and an explanation is given as to how it was refuted or why the game result started to turn sour because of that play

This is obviously the best case possible.

If possible, it's always desirable to have a comment from a human, even if this human is not a pro. Usually, just a stronger player than us will do, because it's what most players always have. But I think that this is stating the obvious. Only a very strong player can benefit only from AI suggestions, with no more explanations.

So, what to do with "mistakes according to AI"? Use with caution, I'd say. In my case, I look at the AI evaluation, and I try to see if it matches my opinion. Or when things go wrong for somebody, I look for the critical move and try to see why.

As long as you don't use AI to say "look, Iyama made a blunder, what a noob" or other hooligan-ish comments like that...

[Gomoto]

Aren't we flogging a dead horse?

I would like to encourage everybody to use their cone cells in addition to their rod cells when contemplating about this topic.

[Kirby]

I am surprised that there is so much discussion on this topic. An AI is just an entity that knows how to play good moves. It can provide sequences an evaluations. It doesn't offer explanations.

This is not that different from pro games in the pre-AI era, which didn't have commentary: amateurs can see examples of high level play, even if there isn't an explanation.

When you have a strong player available to give explanations and context beyond the moves, of course that can be helpful.

Nothing complicated here...

[CDavis7M]

If there's still a horse to flog then the job's not finished.

Anyway, reading my comment again I could have been more clear. I'm not just talking about a comparison between AI analysis and pro analysis, what seems to have changed is the perspective from which commentary is given. It feels bad (to me) for a pro move to be labeled as a "mistake" instead of commending their opponent on a skillful parry or reposte. I think this perspective leads to terse and less informative commentaries. And I think this perspective has taken over a bit with the rise of AI.

In my mind, it's not the "mistake" that lost points, it's the next alternative play that gained points (or an advantage from which to gain points). In plenty of situations, especially in my own games, the best move to play in that context, is a mistake (according to better players, if made against such players). Of course, the context of a pro game is much closer to the context of an AI game. But it's definitely not the same context.

I have a preference for commentaries given from the perspective of how a player can take initiative over commentary labeling moves as mistakes. That's all.

[Kirby]

If there's still a horse to flog then the job's not finished.

Anyway, reading my comment again I could have been more clear. I'm not just talking about a comparison between AI analysis and pro analysis, what seems to have changed is the perspective from which commentary is given. It feels bad (to me) for a pro move to be labeled as a "mistake" instead of commending their opponent on a skillful parry or reposte. I think this perspective leads to terse and less informative commentaries. And I think this perspective has taken over a bit with the rise of AI.

I agree. Though, I would argue that some amateurs did this in pre-AI times, saying things like, "Your move is a mistake, because I saw this sequence in a pro game" - despite not understanding the reasons *why* a pro played a particular way, and without giving credit to the opponent for their response. Trick plays come to mind, though, there might be a difference in nuance.

It's just that now, some pros are apparently doing it, too, in their own commentaries

[John Fairbairn]

I think there is an important aspect that is being largely overlooked, and that alone justifies any equine necrophilia. I think it is encapsulated in the phrase by CDavis "shift in perspective." I'd be more brutal and call it "dumbing down."

It is not a simple pro vs am or pre-AI vs post-AI issue. It is a matter of how much weight is given to each side of each dichotomy.

To restore some perspective, let us look at these two dichotomies. First, go back a wee while in the field of go commentaries. There was a time almost the only way to see a game commented was to buy something like Go World, and the comments were almost invariably by a pro, either directly or through the intermediary of a journalist. But even though they were by pros, these commentaries were far from perfect. To list just some of the faults:

1. The commentary could be too brief, as in Shusai's "121 was good".

2. The commentator would tell us "Black could have played the joseki as in Diagram 1." This sort of comment used to infuriate me. I wanted to know what was going on in this game, not some imaginary game.

3. The pros could be wrong. I have produced lots of books based on multiple prop commentaries - as many as 30 for some games. In such cases you find many instances where pros just plain disagree, even to the extent of "brilliant" against "incomprehensible" or "the losing move" versus "the game is no close." So, somebody somewhere had to be wrong. Related to this are the many more cases where one pro thought a move was worth talking about and yet others ignored it completely.

But what we usefully got from such commentaries was sense of what was happening in the game and (if I may go back to my previous point about "how" and "why") we learned how to looks at a game holistically.

Now the pre-AI vs post-AI distinction. Instant notification of errors is nothing new. We have games like tennis where a decision is made after every rally. We see a player make an unforced error and even if we don't play tennis ourselves, we don't need a Deep Bonk program to tell us that he made a mistake. Furthermore, if we just had a human commentator on tv telling us repeatedly that the players were making such mistakes, we would soon switch off. What we want is someone like John McEnroe telling us why these mistakes are made and why they matter, and how the player can recover. Mutatis mutandis in many sports we've watched for decades before computers were even invented.

Looking at an AI flow chart of a go game just tells us the equivalent of how often a tennis ball is put in the net or out of court. Actually, in a sense it tells us less, because an AI score is just a statistical snapshot, and could prove to be wrong. Furthermore, the alleged mistakes are not 100% errors like a netted ball or a missed throw. In go, they are degrees of error, and again subject to statistical noise. So, if we don't want to listen to amateur tennis commentators prattle on about undoubted mistakes, why would we want to listen to amateur go commentators prattle on about possible mistakes?

Note that I am not saying that bots are not interesting, and I agree potentially useful for those who want to review their own games. And they have already given even pros food for thought. And so on.

But what I think CDavis is talking about is a fast-growing trend for amateurs to produce video commentaries which are the equivalent of "oh he hit the ball in the net, oh he hit that too hard, oh he dropped his racket." To the bling bling of AI they are adding the bling bling of making your own videos. This is by no means limited to go or even games. It's the modern version of expecting relatives and friends to watch your home movies of your latest holiday and pretending you know a lot about France because you had a croissant on the Champs-Elysées.