The Trick
This might well be the most important post from the last years.
So what's the trick then?
Understanding and scenarios are closely related. You can't learn the trick without scenarios. But scenarios are not the essence of the trick.
The trick is to see what a move does. You can find out what a move does by fiddling around with standard scenarios. Take for instance the following scenario:
 |
| White to move |
r1b2rk1/p1qn1ppp/1p1bpn2/2ppN3/3P1P2/1P1BP3/PBPN2PP/R2QK2R w KQ - 0 3
In this diagram, you are close to a standard scenario in the Colle Zukertort. It is not winning yet, although Stockfish gives it about 0.5. Black must do a few things wrong and white must do a few things right before you can count a point.
The idea is to open the long diagonal by two tempo moves (Nxd7 and dxc5), then throw the two bishops to the king position to get rid of Nf6 and h7 (again, with tempo) and mate the black king with Queen, Rook and / or Knight.
I was fiddling around and asked myself the question, can't I start the attack a move earlier? The answer is no. I was surprised. But by playing around, it became clear what the problem is: I need the rook. The rooklift is prepared (f4), but the rook isn't in place yet. After 1. O-O it is.
You need to develop a feel for what a piece does in relation to a scenario. The scenario provides the purpose, but the Trick is to see what the pieces do in relation to that scenario.
After white plays 1. O-O Stockfish provides the following evaluation:
- 1. ... c5xd4 takes away the possibility to open the long diagonal with tempo
- 1. ... Be7 takes away the tempo after 2.dxc5 AND protects Nf6
- 1. ... g6 closes the diagonal d3-h7
- 1. ... Rd8 opens an escape route for the black king via f8
- 1. ... Re8 ditto
- 1. ... a5 dunno, I must fiddle around a bit more with this one
- 1. ... Qd8 protects Nf6
Every opening has 30 scenarios. A scenario consists of 8 positions. Every position has 7 moves for black and 7 moves for white that need to be fiddled around with. This means that you must absorb the patterns that belong to 3360 moves per opening. The numbers are arbitrarily chosen. But it gives an idea of the work that need to be done to accomplish the Trick.
ReplyDeleteI reckon that when push comes to shove, that there is a lot of overlap between the moves. Otherwise grandmasters wouldn't be so good at Freestyle Chess too.
You might define the Trick as the different techniques leading to a certain purpose. A scenario is the lowest level of abstraction of a purpose.
ReplyDeleteIt starts with the 4 sitting ducks:
ReplyDelete* king
* weak pawn
* invasion square
* promotion square
The lines of attack lead towards the sitting duck.
Every LoA (line of attack) has its own set of scenarios.
Every scenario has its own set of positions.
Every position has a set of relevant moves for black and white.
Forget visualization. Replace visualization with logic. The Trick is to let system 1 map visualization patterns to logic. You don't need system 2 for that.
ReplyDeleteIn an interview Kaan Erdogmus explained that his training was mainly based on tactical puzzles "when he was young". So there are many ways to learn the trick. Focus on what all the methods of the prodigies of this time have in common: a deep understanding of the inner workings of tactics and technique. Concrete calculation. Quality is more important than quantity, while speed is taboo. Speed is for testing, not for learning.
ReplyDeletelearn more
There are two things:
ReplyDelete* knowledge (WHAT are my goals). This can be rule-based. It is relative simple to acquire knowledge. It only takes time to gather the appropriate knowledge. If you want to make the difference with knowledge, you must focus on openings. Knowledge is quite democratic. A novelty will soon be adopted by everyone.
* concrete calculation (HOW do I realize my goals). The Trick. Skill. This is what makes the real difference between one level and another.
Books and coaches tend to focus on knowledge. Hence the progress they induce is rather limited.
ReplyDeletePART I:
ReplyDeleteTemposchlucker summarizes the distinction between KNOWLEDGE and SKILL:
KNOWLEDGE:
WHAT are my goals?
Declarative
KNOW-THAT
Can be acquired quickly
Academic credentials
Demonstrated through formal “regurgitate on command” tests
Can be taught through formal pedagogical methods (education)
Can be obtained fairly quickly (cramming)
SKILL:
HOW do I realize my goals?
Prescriptive
KNOW-HOW
Apprenticeship/On-the-Job Training
Demonstrated expertise through DOING
Can only be acquired through practical experience
Can only be obtained via repeated practice over a long period of time
Aristotle:
We are what we repeatedly DO. Excellence (SKILL) then is not an act, but a habit.
Ralph Waldo Emerson:
SKILL to do comes of DOING.
One of the best illustrations of the distinction between KNOWLEDGE and SKILL is Gary Klein’s story about his cooking experience in Sources of Power: How People Make Decisions, Chapter 10, The Power to See the Invisible.
PART II:
ReplyDelete~+~+~+~+~+~+~+~+~+~+~+~+~+~+~+~
Example 10.2, Recipe for Disaster.
I am going to have to cook a special dish for some guests, and I don’t know what to make. A friend offers me a recipe for an outstanding dish. It is important to impress my guests, but it is also important not to fail as a cook.
Should I make the dish? The answer depends on my expertise.
I have little experience in cooking. I’m not even such a hot shopper. My family still recalls the time I was sent to the supermarket to buy a head of lettuce and returned with a cabbage.
For someone like me [a novice], it would be a mistake to use the recipe. A recipe, we all know, is a set of procedures. It includes a list of ingredients and a list of operations to perform on these ingredients.
None of this is as simple as it seems.
Let’s say I have decided to use the recipe. I start cooking away a few hours before the guests arrive. The recipe calls for a red pepper, and I don’t have one. Can I use a green pepper as a substitute? The recipe calls for raisins. Can I use grapes and add sugar? The recipe calls for two potatoes. I dig out two and notice they look small. Should I add another one or two?
In short, HOW DO I IMPROVISE?
How do I know the size of a typical potato? I am in trouble even before I have started cooking.
I realize I need to prepare side dishes, so I look in the refrigerator to see what there is. I must ignore the issue of how the dishes will taste together since I do not have the experience to imagine how my new recipe will taste. I can stare at its ingredients and directions and not be able to anticipate its final flavor. I look at an ingredient like “flour” and wonder what taste the flour is going to add. An experienced cook would know that the flour is there to thicken the dish, and by adding too much flour I have just made sure it was going to congeal. The experienced cook could imagine how the originators tried it without flour, and decided to thicken it. I just see the word flour.
Now I am cooking the dish, grating the potatoes. But the work is taking too long, and the potatoes are turning an unappetizing brown. I was supposed to add these potatoes fifteen minutes ago, but it is hard to grate such small potatoes, and my fingers are getting tired. I lack the metacognitive skills to take into account the time it takes me to perform the operations. Should I turn the heat down under the pan while I am finishing the potatoes, or will that spoil some chemical process?
Speaking of finishing, what is the dish supposed to look like when it is finished? I’ve never made it before so I can’t tell. The recipe says to bake it at 350 degrees for one hour, or until brown on top. I don’t know what to make of “brown on top,” so I better stick to the one hour. Is my oven calibrated accurately? It already looks done on top, but will it still be mushy inside? Even worse, my guests won’t arrive for another twenty minutes. If I turn the oven off now, will it get too cold? Should I leave the heat on low, or will that dry the dish out?
An experienced cook would notice that the dish wasn’t bubbling around the edges, so it wasn’t done. The reason it was getting dark so quickly was that I had left the rack too close to the heating coil. I don’t see the missing event—the failure to achieve bubbling.
~+~+~+~+~+~+~+~+~+~+~+~+~+~+~+~
The novice relies on the formal surface aspects because that is all he can “SEE.”
The expert relies on the hidden aspects because he can “SEE” HOW everything fits together based on personal experience.
The TRICK is more complicated (to acquire) than simply by amassing 10,000 hours of so-called “practice.” Anyone with experience with the MdlM or Woodpecker Method can readily attest to that truth.
TANSTAAFL
I'm sure I would make a perfect dishaster.
DeleteIt makes perfectly clear why so many attempts have failed. And it helps to describe what is missing.
ReplyDeleteI'm missing especially endgame knowledge. Yesterday I had a somewhat better endgame. I saw a lot of things that I wanted to do But I couldn't decide on what was important. Should my king go to the left and help the pawns there? Or to the right where my main group of pawns might be vulnerable. I didn't know the WHAT, so no skill would help me here.
It might well be that when I acquire the knowledge, that I discover that I lack some skills too. But I must start at the beginning. (which I do not, because I discovered even bigger holes in my bucket in the skill department of the Vukovic gap)
TANSTAAFL. But my own lunch is often too charred to eat..
I lost the endgame, btw.
DeleteHave a look at Gobet
ReplyDeleteThey haven't published part 2 yet, if I'm not mistaken.
DeleteWhat conclusion can we derive from this video?
System 2 is apparently biased towards proposals of System 1. It leads to tunnel vision.
Why does it seem that titled players are less biased? They are biased too, but their System 1 has absorbed the second solution too.
How can it then be that amatures score so well when there is only one solution? They must have absorbed this solution too.
Maybe because we can find these solutions too with the aid of System 2, as long as there is no distracting alternative produced by System 1. Hence the test must be done while measuring the time. This gives an indication about which system is used.
It looks like System 1 directs the saccades, and System 2 must follow. This means that the only way to avoid tunnel vision is to educate System 1 better.
DeleteSo far, Part 2 is members-only. That's a shame. I read Gobet's The Psychology of Chess. It's a small book with a lot of good information in it. Unfortunately, my copy disappeared when I "cleaned house" in anticipation of down-sizing - which hasn't happened. Oh well, if I need another copy, I'll buy another copy. I'm sure Gobet won't mind.
DeleteI consider my two problem sets (The Checkmate Patterns Manual and 1001 Chess Exercises For Club Players) as finished for the moment. Yesterday I started with a new problem set: 1001 Chess Exercises for Advanced Club Players.
ReplyDeleteI'm excited!
I'm curious about your training process. Do you work on the problems by just visualizing the solution (along with recognizing the patterns involved) without an actual board or do you set up a board with each position and shuffle the pieces around in order to "cement" the visual patterns into memory?
DeleteI usually work directly with the book and don't spend the time to set up each position on a board.
Gobet part 2. I didn't encounter a paywall.
ReplyDeleteWhen I went back to it, I could play it. The "members only" restriction apparently lasts for a day or two, and then the general public can catch up.
DeletePart 2:
Two things stood out for me:
Lazybot at 2800 Elo is based on pattern recognition only.
Intuition is trained pattern recognition guiding a cognitively bounded search.
It would be interesting to know how much patterns Lazybot has stored.
DeleteOk, that definition of intuition is more or less acceptable. To avoid confusion, I probably will not talk about intuition that much.
Gemini estimates the neural network has 750,000 parameters
DeleteIn the first video Gobet talked about 300,000 patterns as an under limit for a grandmaster. If I remember it well.
ReplyDelete300,000 patterns in 10 year is 82 patterns per day at average => 300 patterns per rating point. That sounds a bit much.
300,000 patterns total ÷ 300 patterns per rating point = 1,000 rating points (if my math is correct). It's unclear as to how that relates to a GM's rating of 2500+. Given 300,000 patterns as the lower bound for a GM and a GM rating floor of 2500, wouldn't that be 120 or fewer patterns per rating point?
DeleteIMHO, there is no clear linear relationship between patterns and rating points. Why not? Because what are considered to be patterns are not discrete entities. There is considerable blurring because of simple patterns becoming blended into more complex patterns. As noted by Hofstadter and Sanders, there are no sharp dividing lines between categories of any kind, in spite of our valiant attempts at Platonicity.
AFAIK, all such estimates are SWAGs [Scientific Wild-Ass Guesses] usually projected from relatively simple computer programs, rather than a researcher "popping the top" off someone's head and counting the number of discrete patterns. [Sarcasm off]
Besides, there is nothing prescriptive in total pattern counts nor in projecting "average patterns per day". There is no training program that can feed 82 patterns per day for 10 years into the average< person (who doesn't exist except as a statistical calculation).
Just my jaundiced cynical opinion.
"My jaundiced cynical opinion" may have more validity than I thought. Holding, D. H. (1985). The psychology of chess skill. Hillsdale, NJ: Erlbaum suggested that experimental results obtained by Chase and Simon regarding chunks could be accounted for by as few as 2,500 chunks rather than 300,000 chunks.
DeleteGiven the huge variability (10,000 to 300,00 chunks) "guess-timated" to be required for GM -level skill, there is no solid range estimate of how many patterns/chunks/templates are actually required to sustain a given skill level.
Even if we knew a more exact number, we still would be no closer to developing a training process that would provide the required exposure to those patterns, the required repetition of exposure, and the personal factors (such as focused attention) necessary for absorption.
In short, we are still on our own.
If MDLM became 600 points better by just using the 1100 problems of Art 3, then I assume that more nuance is needed.
DeleteI'm interested in these kind of numbers since they might reveal flaws in my approach.
DeleteThe existence of one Trick seems to be consistent with Lazybots trick of pattern-recognition-only resulting in a 2800 rating. Which is the same as the best humans.
ReplyDeleteGobet doesn't seem to give an answer on how to achieve that other than start way before your 12th birthday.