Rethinking Human Chess Calculation: Beyond Kotov's Tree

Рет қаралды 3,480

Күн бұрын

Fundamental Chess Calculation Skills Just Won the 2023 Best Tactics Course Award! www.chessable.com/blog/announ...
www.chessable.com/fundamental...
🔵 My Chessable Courses: chessable.com/drcan
♟️ Find me on Chess.com: canka19
♟️ Find me on Lichess: cantosh
🏆 2022 Chessable Community Author of the Year! www.chessable.com/blog/announ...
Connect on / kabadayichess
chessable.com/drcanpro
00:00 Carlsen's Derived Move
04:10 Kotov's Ideas
07:37 Kasparov's Derived Move
11:01 Svidler's Derived Move
13:32 Endgame Study Derived Move
17:11 Homework
This video talks about the cognitive psychology behind human calculation and how it differs from engines. Alexander Kotov, in his influential book "Think Like a Grandmaster", proposed a model for calculation that resembles how computers calculate. Put simply, he said we should first identify all the candidate moves that look reasonable and then calculate them one by one, never revisiting our earlier analysis - as that would be a waste of time.
But years of experience and verbal protocols from grandmasters paint a different picture when it comes to human calculations. Humans revisit old branches and engage in progressive deepening while calculating. It is a much less structured process than what Kotov proposed. Humans analyze one candidate move and discover something, and then they use that discovery in another branch, changing their move order in the process. This video shows several examples of such "derived moves" and how humans discover those moves.
Surprisingly, most of these findings were already articulated by Adriaan de Groot in his Ph.D. thesis "Thought and Choice in Chess" in 1946. Those ideas stood the test of time, unlike Kotov's.
Books mentioned:
- Thought and Choice in Chess by Adriaan de Groot
- Think Like a Grandmaster by Alexander Kotov

Пікірлер: 96

@chimbiepaladin4629 5 ай бұрын

This was the funnest and most satisfying chess youtube video Ive ever watched. I felt like it was candy for my brain. Searching for the solutions was enjoyeable and even if I missed a point, seeing it was very satisfying. Stuff like this makes me stronger. And happier. This is real chess.

@Dr.CansClinic 5 ай бұрын

I am so happy to read this... Such comments are extremely motivating ❤ I am doing it to help you improve your game and spread my passion for the game.

@jonnyjansson7320 5 ай бұрын

The first position is very interesting, because actually d6 was the first move I looked at because of the pattern Qc3-Nd5-Qb8. The inadequately defended Night on d5 fulfills the 3rd of Silman's precondition for when a combination might exist. But I didn't, at first sight see anything after ...exd6. In order to get a feel for the position I used a method, which I learned recently from one of Anna Cramling's videos about how to quickly evaluate a position, where she said that you have to look in descending order of importance: King position, material, piece activity, pawn structure, and space. After I done that, I noticed that the knight on d5 doesn't have any squares. f4 then immediately came to my attention, and I saw the hidden purpose of Qb8, because the knight can go to g4 and you win it back with the fork on f4. Ok, can I improve this? Starting with Rg1 followed by f4 seems like an idea, but what does Black do in between? Here, I started to look for Black's threats and it dawn on me that White cannot allow the c-file to be open, because after Rg1 cxd5 exd5 e6, Black threatens Rc8 pinning the queen. If Rg1 cxd5 f4, black can just ignore the threat and play e6 giving the knight a chance to escape via the square on d7 while threatning the pin. And here the circle closes, when I realise that the idea of d6 is to cut the connection between the queen on b8 and the square on f4. Now, everything becomes simple: d6 exd6 f4 winning the knight. The interesting thing here is that what I did is as far from Kotov as you can be. Here, I think, I was following closely to what IM Andras Toth says in his channel: You first calculate a variation and evaluate, then you try to disproving it by either stepping forward or stepping backward through the the variation trying to find improvements. I think seeing the move d6 immediately or very late if at all is dependent on how much tactics training you have had, and I have had a lot. I hope that seeing my thought process when I looked at this position is interesting and relevant. I found your discussion, of the position and the enormous difference of how top players actually think in contrast to Kotov, afterward most interesting.

@Dr.CansClinic 5 ай бұрын

Thank you so much for taking your time and writing this thought protocol! ❤️ It was very interesting to follow your thoughts there. Good that you realized the potential pin on the c-file. And Rg1 may give Black several choices to save the knight. Yes, good tactical vision would help solve this one easier. The falsification of the immediate f4 was also crucial - many people would have wishful thinking at that moment.

@Gingnose 5 ай бұрын

Prioritizing a promising move is also implemented into the Monte Carlo tree search in machine learning. Probably all strong engines like stockfish and alphaZero incorporate this "promising move first" method because of the reduction of calculations, often called "pruning" (because it is like cutting the tree branch). However, in terms of calculations/move, GM has the upper hand because GM will calculate 100 moves in average before making a move compared to engines like alphaZero which calculates 10,000 moves before making a move. So, GM is much more efficient in that respect. This discrepancy is probably caused by the deep knowledge that grand masters have in many many positions that they played or analyzed others playing or positions from books, that they know somehow what to start searching from (nuanced, probably very difficult to put in language and gms couldn't explain it well too, one easy example is checks, captures, threats but this is most people learn in early stages). It would be interesting to see the advancement of engines with more efficient searching methods and humans might be able to learn from them too. Sorry for diverting but somewhat felt tangent to the topic because I'm currently learning about ML. Thanks for the great content as always!

@Dr.CansClinic 5 ай бұрын

I love your comment, thank you for taking your time and writing it! Thank you for mentioning this similar process in the Monte Carlo tree search! Absolutely, GM's large number of chunks in their long-term memory allows this reduction. Novices tend to look at 'irrelevant' lines that masters discard instantly. @@Gingnose

@Gingnose 5 ай бұрын

I thought I originally posted but I was accidentally replied to a comment, sorry for that.

@peterintoronto6472 5 ай бұрын

As Deleuze and others noted, our thinking is more rhizomatic than tree-like.

@Dr.CansClinic 5 ай бұрын

As always, lovely comment Peter!

@adomaskuzinas2137 5 ай бұрын

Can said i am a great player! Yahoo!

@Dr.CansClinic 5 ай бұрын

You are a great player for finding that Bd6 move!

@joeperry1188 5 ай бұрын

This is a really really awesome video. I really love content like this. It's so useful.

@Dr.CansClinic 5 ай бұрын

I am so happy that you liked it ❤️

@alwaysprepared 5 ай бұрын

Great video and great procedure to improve calculation! So obvious, yet so often overlooked!

@Dr.CansClinic 5 ай бұрын

Thank you so much for your kind words 🙏

@csouza608 2 ай бұрын

Amazing and intructive videos!!! Congrats!

@Dr.CansClinic 2 ай бұрын

Thank you so much for your motivating thoughts! Please do not hesitate sharing the channel with your chess friends. That is very important for me to be able to produce similar content.

@user-gg6df4xr3i 5 ай бұрын

Very instructive,looking for what stops the calculated candidate move or the check.I did actually get this one...

@Dr.CansClinic 5 ай бұрын

Thank you! 🙏

@Eudaimonist 5 ай бұрын

This is a fascinating subject because I will soon start writing an AI chess player as a hobby for a game that I'm writing at home on a home computer. While I don't have anything against adding alpha-beta tree search analysis to the AI, instead of making the fastest and best tree searching AI (which I will never manage as a hobby programmer anyway with limited resources), I actually want to write an AI that is good at strategic analysis, positional evaluation, and making principled moves that might make one think that a human being is making those moves. I'll be taking notes.

@christmackey 5 ай бұрын

You should look at the maia chess project. There are three bots on lichess you can play against to test them. Their code is open source and they have a package to train the bot on your own data. The bot is designed never to search, only to play the most human like move.

@Dr.CansClinic 5 ай бұрын

Fascinating! I made a video about how Maia predicts human mistakes and how we independently came to the same conclusion! kzbin.info/www/bejne/aGK0e6R5f5yMjs0 There is also Noctie.ai from Sweden. The idea is to create AI that plays like human, and plays on different rating ranges. It does not calculate but makes 'intuitive' decisions!

@christmackey 5 ай бұрын

Thanks Doc

@Dr.CansClinic 5 ай бұрын

My pleasure! ☺️

@KF1 2 ай бұрын

Dr. Can, at 8:20 I found bishop takes e4 by working backwards. Without the ability to play d4, white cannot hold on, so the bishop sack comes naturally by rechecking the calculation starting with the deflection sack provided by the b7 bishop. Cool stuff!

@KF1 2 ай бұрын

At 11:40, instead of your solution rook to e2, can you explain why queen to g3 is not better? I don't see the defense. Edit- oh, knight to c6 can block the alignment. And if we take the knight first, white gains a tempo to play bishop e3. Removing the bishop then opens their queen to protect laterally. That was a tough one

@Dr.CansClinic 2 ай бұрын

Great way to solve that problem. Probably Kasparov also used similar thinking!

@Dr.CansClinic 2 ай бұрын

...Qg3 fails to Nc6!, and White shuts down the bishop on b7, thus preventing mate. ...Re2 solves that problem!

@KF1 2 ай бұрын

@@Dr.CansClinic Thank you. Yes, I looked again and found this, very cool! You have an excellent teaching style!

@AndytheHumphrey 5 ай бұрын

Was staring at the final solution for ages trying to find black's defensive resource but once I found it the move was easy. The defense involves opening up the bishop to defend the g1 square so taking rook then the knight so the solution is to move the knight from c3 with the biggest threat.... Na4 hitting queen either winning the queen or mating with the plan - incredible as would never think I could find a move like this. Top quality content for chess improvement have recently subbed to you, keep this up and you will earn many more subs in the future!!

@Dr.CansClinic 5 ай бұрын

Beautiful solution! Thank you so much for your motivating comment! 🙏

@genius_127 5 ай бұрын

I was looking at the final position and found your comment, I thought Nd5 was a solution as it seems to serve a similar purpose. What would be better? please recommend.

@AndytheHumphrey 5 ай бұрын

Having a look back I think Nd5 achieves the same thing so is an equally good solution(?) It feels less natural to me as it blocks the light squared bishop but since the bishop is not crucial to this particular combination I think it also works just as well

@AndytheHumphrey 5 ай бұрын

Had a look at other comments and Nd5 doesn't work as after cxd5 the queen guards f6 which is important as you need that resource e.g. Rf4, exf4, gxf4, Kh8, Rg1, Rg8 and white no longer has Qxf6# which they do have in Na4 line

@adomaskuzinas2137 5 ай бұрын

In the endgame study, it is important to include why Nd5 is the only possible resource for black - trying to stop the pawn with Nd7 fails due to the variation a5 Nf6 a6 Nd7 Bc5!! and the pawn promotes. Nd7 also fails in the d5 variation, also because of the beautiful Bc5, deflecting the knight and promoting by force!

@Dr.CansClinic 5 ай бұрын

Absolutely, very good point!

@tobiass3540 5 ай бұрын

I actually already knew the endgame study. Finding d5 was the hard part for me, Bd6 I saw instantly. But yeah, d5 took some time, the first time I saw that study :) Anyway, great video! Thanks!!!

@Dr.CansClinic 5 ай бұрын

Nice! Good that you knew that study!

@kitsune090 5 ай бұрын

Of course you do see D6 even without seeing the follow-up move... you could play D6 in other to block the queen's defense of the knight since it is only protected by it at that moment and your queen is attacking the knight... but after your opponent captures your pawn with their pawn and it is now protected again, you realize you can now attack it with a pawn, EVEN If you still not seeing it is trapped, but probably you already notice it even without properly calculating it... if it's a classical game you can calculate all those moves before, but if it is a bullet match you may use that "play first and think later technique", where you calculate only 1 move ahead and think about the follow-ups and variations when it's your opponent's turn...

@Dr.CansClinic 5 ай бұрын

That describes a trial and error process that is very inefficient in real world settings. Would this blind mechanism also work in Kasparov's game that I later show in this video?

@luca4xxx 11 күн бұрын

@@Dr.CansClinic I'm a positional player, i felt the need to close the c file and bury the enemy Queen. Also noticed the black Knight is almost trapped, without a secure outpost, dominated by White bishop.

@luca4xxx 11 күн бұрын

@@Dr.CansClinic Am I wrong? Weaker chess players tend to apply Trial and Error machine learning process, kind of Alphazero MCTS method; As you said it's slow and require very strong GPU hardware compared to Stockfish alpha-beta pruning faster and less CPU hardware intense. For that reason they say Stockfish thinks like a human GM...

@luca4xxx 11 күн бұрын

@@Dr.CansClinic I feel CM-M and below apply inefficient and blind Trial and error combined with "opponent threats" and "null move" what if I can play 1-3 moves in a row (reach dream position), myabe you named reversal thinking. Maybe GMs developed a faster instructional process, faster and efficient. We tend to be lazy and avoid calculation at lower level, we rely too much on static evaluation of the position....forgetting the dynamics...for that reason we apply general rule in the wrong position; GMs have strong and quick calculation skills, they don't need Trial and error, just like alpha beta pruning.

@luca4xxx 11 күн бұрын

@@Dr.CansClinic In Kasparov I saw the check Be5, and the opponent's best response d4. But you can apply trial and error since Bxe4 is a forcing capture move. Both moves are aggressive candidates moves: check and capture. I also notice Bxe4 after dxe4 i need to have something tactical, since the Qh3 isn't no more disconnetted from the queenside flank. I felt it's an anti-positional sac! I didn't saw immediatly the backrank mate, but first thing I noticed that the king's safety: the white king's lack of potential escape squares.

@chessophiler 5 ай бұрын

11:13:. . .Qg3 forces mate on g2 (if position is correct). If W defends with Bd5 and/or Qe4, or Be3, all to no avail. Pf2 pinned! Did not use Kotov's tree analogy.

@Dr.CansClinic 5 ай бұрын

But doesn't White have Nc6 against it?

@chessophiler 5 ай бұрын

Yes! Missed Nb8 defense. Oh well, back to the tree! Have u seen Botvinnik-Kotov, Groningen, 1946? A Kotov krusher and one of my favorite games. BTW subscribed!

@aldocaraig 5 ай бұрын

Kramnik and Aagaard proposed the 3 Question rule. Cant discussed it here for copyright law. But you can find it in Chessable Kramnik Thinking in Chess and Aagaards Grandmaster Preparation Calculation

@Dr.CansClinic 5 ай бұрын

Yes, I am familiar with those questions. They are used as a guide to help find the best candidate move. This video mostly discusses the 'comparison' and 'choice' between those candidate moves. (e.g. should we 'revisit' the branches we looked at?)

@derkamon 2 ай бұрын

homework: took me one day, still not sure of the answer the knight allows a discover control of the bishop on g1, so need to get rid of it first, maybe knight a4 and gain a tempo on the queen then rook f4

@Dr.CansClinic 2 ай бұрын

Beautiful! Time well spent! Na4!! wins!

@user-ot8bb3ng7o 5 ай бұрын

Really good video 🎉I found it very difficult to think like tree of analysis idea ,I came across this idea in middle game in chess ,by kotov and keres,the process is distracting,i depend on intuition after which which I begin to investigate deeper, sometimes we have to keep adjusting to the opponents move,a thinking formula is hard to follow

@Dr.CansClinic 5 ай бұрын

Thank you! Yes, it is hard to come up with a rigid algorithm that always gives the best move.

@Opferschach 5 ай бұрын

I spent a lot of time on the final position, but I wasn/t able to come up with anything better then 1. Qxf6 Re8 2. Qg5+ Kf8 3. f6 Re6 4. Qg7+ Ke8 5. Qh8+ Kd7 6. Bxe6+ That just wins the exchange and it is too long, so it/s probably not the intended solution. I bought your calculation course in early January, but I just clicked at some variations, havent/t really studied it yet. In great part because looking at your videos and not being able to solve a single position made me think it is above my current level. Hopefully we can see the solution in the next calculation video.

@Dr.CansClinic 5 ай бұрын

Thank you for your comment and buying my calculation course! What is your rating if I may ask? When it comes to the final position, the main idea is Rf4 exf4 gxf4, and we give mate by Rg1+. But Rf4 does not work directly, due to dxc3 and the bishop covers the g1-square. So we have inflict a surprising looking move first, to mess up with Black's only defensive idea against Rf4. What move can that be?

@Opferschach 5 ай бұрын

@@Dr.CansClinic We also may move the c3-knight away. Although I can hardly imagine that is what you meant when you talked about 'the surprising move'. But if Black plays dxc3, we are not putting anything on g1 in this game. I am usually in the high 1200s chess/com blitz and high 1700s on LiChess blitz. But my skills are all over the place. I took a chess acessment test a while ago, and it came out that tactics is one of my weakest areas. In pawn play, which is admittedly my favourite chess topic, I was placed at the respectable 2000+ level. I saw an early review saying that the calculation course is great for 1500s aiming to reach 1800. From what I saw that might be fair. For now I started working with a puzzle book for beginners, it is in my language. Even though it is aimed at beginners, I don't score 100%. I mean to study your course when I am done with that puzzle book. I know full well that I have no right to complain that I am not improving if I don't study stuff that I bought.

@Dr.CansClinic 5 ай бұрын

Yes, Na4! is the solution. Move the knight with tempo and disallow ...dxc3. Then we go for the Rf4 idea. What chess assessment test was it? I am curious to see. Yes, it makes sense that you tackle that puzzle book first before tackling my calculation course. Please let me know about your progress! @@Opferschach

@douwehuysmans5959 5 ай бұрын

I'd call d6 an interference sacrifice. I know usually sacrifices refer to pieces, but if we want to properly label this tactic I'd go with this.

@Dr.CansClinic 5 ай бұрын

Yes, d6 interferes with Black's main plan!