It is currently Fri Mar 29, 2024 3:05 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 47 posts ]  Go to page Previous  1, 2, 3
Author Message
Offline
 Post subject: Re: Reviewing Lee Sedol's commented games with KataGo
Post #41 Posted: Tue Mar 22, 2022 5:57 pm 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
CDavis7M wrote:
Just to go back a bit... The opening post is about how Lee Sedol's judgment of a move or position may have been biased and how the judgment is sometimes different from AI judgment. Having looked again at Vol. 2 (I don't have Vol.1) I just wanted to point out that even if Lee Sedol's judgment is wrong (I don't think so), his judgment is not the teaching point of the commentary. Lee Sedol teaches techniques for making your own judgment (summarized):
(...)
It should be no surprise that KataGo has different judgment from Lee Sedol. But KataGo cannot provide better judgment-techniques or even assess whether these techniques are good or bad.


Vol 1 provides similar lessons and you make a good case of these being valuable and valid advice, regardless of the tactics and evaluations presented.

Still I disagree with your last statement. KataGo has not just different judgments from Lee Sedol but must on average have better judgment, since it's likely on par with AlphaZero who beat the leading pros 64-0 and pushed them on an estimated 2-3 point handicap. Kvasir has made an interesting case that this 20 point difference on average may come rather from the compound effect of smaller "mistakes" which don't even add up arithmetically to the difference in the end, than from the "bigger mistakes". However, if KataGo's precision is off in any way, the bigger mistakes might even be bigger than quantified.

KataGo doesn't articulate its judgment except for showing sequences underneath its percentage and point wise assessments. It needs interpretation and articulation by a human. At the pro level this is beyond my capacity. At my own level I have been able to discover heuristics from KataGo's evaluations and sequences.

There have been records of professionals analyzing their game right away by showing sequences to each other, because they didn't speak each other's language or even in case they did. Their "hand talk" is often a better communcation tool than speech. I'm sure professionals are learning a lot from AI, despite the lack of articulation.

I'm going to rest my case here and can agree to disagree on the usefulness of AI analysis of pro games, commentaries, articulations and claims.

Top
 Profile  
 
Offline
 Post subject: Re: Reviewing Lee Sedol's commented games with KataGo
Post #42 Posted: Tue Mar 22, 2022 6:09 pm 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
jeromie wrote:
(...) This bring us back to some of the earlier discussion on the first game: the overall result wasn’t bad, but perhaps it was bad for Lee.


In some parts of the analysis he indeed personalized the evaluation in that way: whether the result fitted his style of play. In other parts he made the claims more absolute, like no one should play that way, while KataGo thought it was playable or even best. (and the other way round). I don't blame Lee for that (anymore): AI hindsight is too easy to dismiss a thoughtful commentary by arguably the best player of his time.

jeromie wrote:
This is, overall, an interesting way to look at professional commentary and how to play the game of go. Each of us is making choices, and we can try to direct the game toward one that is favorable to our strengths and preferences. I wonder how many of my clear blunders (which are many at my level!) come because I let myself be lead into a situation that doesn’t play to my current strengths?


I can give that to a professional. At my level I consider myself to have mostly weaknesses and hardly any strengths. The overall degree of idiosyncracy of AI is IMO overestimated. The only "advice" I'm reluctant to take from KataGo is its choice of invasion points and therefore its leniency to opponent moyos forming. I think I will play a little more influential go than what KataGo advises because I find its invasion tactics too complicated. Still that leaves plenty of room for conceptual learning: slow connections, heavy cuts, raw peeps ...

Top
 Profile  
 
Offline
 Post subject: Re: Reviewing Lee Sedol's commented games with KataGo
Post #43 Posted: Tue Mar 22, 2022 7:36 pm 
Lives in sente

Posts: 902
Location: Fort Collins, CO
Liked others: 319
Was liked: 287
Rank: AGA 3k
Universal go server handle: jeromie
Knotwilg wrote:
AI hindsight is too easy to dismiss a thoughtful commentary by arguably the best player of his time.

Yes, I agree. I have a lot to learn from a variety of sources. Disagreement at a high level can help me to understand that professional advice is not absolute, but I can certainly glean something from their thought process!

Knotwilg wrote:
I can give that to a professional. At my level I consider myself to have mostly weaknesses and hardly any strengths.

At an absolute level, this is certainly true. And I’m considerably weaker than you. I am sometimes amazed in review (or even in mid game) at the ridiculously poor level of play I can demonstrate. But I do think it can be helpful, at some level, to think of the type of game I want to play and try to move the game in that direction. Not because I’m truly strong in that style, but because it gives me some strategic direction that might help my moves make sense together.

Top
 Profile  
 
Offline
 Post subject: Re: Reviewing Lee Sedol's commented games with KataGo
Post #44 Posted: Wed Mar 23, 2022 6:40 am 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
It would be very interesting to see how much KataGo agrees with pro opinion in this game., although I may be hijacking your thread a bit . . .

Top
 Profile  
 
Offline
 Post subject: Re: Reviewing Lee Sedol's commented games with KataGo
Post #45 Posted: Wed Mar 23, 2022 9:54 am 
Lives in sente
User avatar

Posts: 714
Liked others: 109
Was liked: 138
Rank: Shokyu
Universal go server handle: CDavis7M
Knotwilg wrote:
KataGo has not just different judgments from Lee Sedol but must on average have better judgment
I think you're missing my point. I don't disagree that KataGo has "better judgment" than Lee Sedol but it is better for itself, not for Lee Sedol, and certainly not for much lower ranked players. A lot of AI suggests have miniscule benefits if playing as an AI and many AI suggestions are too difficult for even pros to pull off.

Knotwilg wrote:
I'm going to rest my case here and can agree to disagree on the usefulness of AI analysis of pro games, commentaries, articulations and claims.
I don't disagree. It is useful. My only disagreement is that the original post implied that Lee Sedol's judgment wasn't on-point because KataGo said otherwise, but I think Lee Sedol's judgement was correct for him.

Top
 Profile  
 
Offline
 Post subject: Re: Reviewing Lee Sedol's commented games with KataGo
Post #46 Posted: Wed Mar 23, 2022 4:19 pm 
Lives in sente
User avatar

Posts: 714
Liked others: 109
Was liked: 138
Rank: Shokyu
Universal go server handle: CDavis7M
Speaking of discrepancies between Pro commentary and AI, I looked up one of my favorite plays with KataGo.

Unnamed Pro (Shusai?): "Incomprehensible - a 4-dan idea"
KataGo: "Meh, -0.2% (but not 'a' +0.5%)"
Me: "I hope to have a 4-dan idea someday."

Click Here To Show Diagram Code
[go]$$c
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . O . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . X , X . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . W X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . X . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . . . , O . . |
$$ | . . . . X . . X . . O . a . X . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Top
 Profile  
 
Offline
 Post subject: Re: Reviewing Lee Sedol's commented games with KataGo
Post #47 Posted: Thu Mar 24, 2022 11:19 am 
Lives in sente

Posts: 905
Liked others: 22
Was liked: 168
Rank: panda 5 dan
IGS: kvasir
CDavis7M wrote:
Speaking of discrepancies between Pro commentary and AI, I looked up one of my favorite plays with KataGo.

Unnamed Pro (Shusai?): "Incomprehensible - a 4-dan idea"
KataGo: "Meh, -0.2% (but not 'a' +0.5%)"
Me: "I hope to have a 4-dan idea someday."


Honestly to myself "a" looks like overthinking but KataGo likes this way with enough playouts. It is normal to try to go out when you are about to be sealed in and I think most players should follow such instincts (play the game move) most of the time and that will help more with improving than trying to come up with a fancy move. Granted that you may play a normal looking move and realize later that it was a mistake, that will lose games but it is also an opportunity to learn.

If you think about it the game move creates some problems by giving black outside influence facing the enclosure. White will have to reduce and can we be sure white has enough...to me it seems like overthinking it if you consider all of this and then conclude that white can't play like that :scratch:

This doesn't have anything to do with the Lee Sedol commented games per-se but it goes to show that KataGo and pro commentaries can equally be useless because they can agree not to play normal moves without obvious reasons. On the other hand if you have played the game move countless times it will be interesting that there are other good moves.


This post by kvasir was liked by: Elom0
Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 47 posts ]  Go to page Previous  1, 2, 3

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group