A very interesting article. I will give my experience and insights, such as they are. I hope this may be helpful.
Struggling with chess as an adult improver, I quickly came to think that I have cognition problems (in an area or areas that affect chess). I felt I maybe had spatial perception problems because after all chess is a 2D space and even a of 3D arena if one includes time. So I did a spatial perception test on line. It was the kind of test which uses dice shapes with patterns on each side. The dice is presented as a cube with 3 faces visible and also as an unfolded flat surface. Then one must work out or "see" if the flat shape has patterns consistent with the visible dice faces or vice versa. There were also combined cube shapes like say the gold, silver, bronze podium boxes (as a compounded shape) also presented isometric and unfolded flat.
I noticed that sometimes I could immediately just "see" or "tell" that an unfolded shape matched a cube or other compound 3d shape (or vice versa) but that other times I could not. In the latter case, I had to use steps of conscious logical thinking to try to work out the answer. The possible answers were multiple choice with 5 (IIRC) options. I found that my spatial perception was reasonably good, possibly even a bit above average, but not exceptional.
The interesting thing to me was the fairly clear demarcation between rapid intuitive thinking (in just "seeing" the right answer straight away or almost straight away and just feeling it was right) or being forced to go to step by step via logical or algorithmic thinking to try to work it out. What is the difference and what is the cause of this difference? I think the answer is in neural plasticity.
I posit (and I am sure this is not an original thought) that just "seeing" the right answer (very rapidly) occurs because a neural pathway has already been laid down for that pattern or very similar patterns. That pathway is laid down by play and training (in chess) when people are young, often very young. These pathways are essentially firmware in the brain: sets of neurones / synapses grown and strongly interlinked under the challenge of solving specific patterned problems of the type trained in. On the other hand, step by step logical or algorithmic thinking is a general (though still neurally encoded) tool we are taught, and learn to different degrees of proficiency, to solve problems not directly addressed by bespoke pattern recognition and automatic "component shuffling" in our firmware grown via neural plasticity under challenge
Adult improvers have a very hard time in chess and in any number of other physical and mental pursuits. Their neural and physical plasticity has declined. Their brains and bodies are "set jelly" not "still setting jelly" and cannot be conformed as quickly or easily to new tasks: or at all once they get old and sclerotic.
Just a note on the terms physical and mental, or body and mind. The body and brain are both physical. It follows (from this physicalist point of view) that the mind is all physical too. Descartes was wrong IMHO. Happy to debate this. I consider I have a neat proof that Descartes was wrong and will post it on request. All the answers we are looking for re this problem (human cognition) are entirely in the physical realm. A strict physicalist conception even requires that we find, philosophically and then scientifically, a way to unify the ontology of real system and formal system interactions. Formal systems are really just a special category of real systems. Formal system knowledge, as information, exists as patterns in real media: real media being brains, books, computer memory storage etc. The codes of formal knowledge and the codes of algorithms for encoding, decoding, manipulating and applying formal knowledge all exist as patterns in real media containing information. Real agents, as humans or computers for example, operate on and via this information. Happy to debate this too but maybe not on this general forum.
not only small brain but noisy channels. yet our visual cortex gets all the signals. The eye gaze and the attention might share voluntary intent, and where the precise retina wanders in the whole FOV is part of that, but our visual low level channels for chess are not that noisy.
That A0 gets also low level full peripheral vision acuity might be in your direction (op) of argument, but the attention problem is also there, and it might be even made experimental now in some developments of LC0 (attention NN layers, or modules, not sure how they did it, just read some mention somewhere, stopped following them for a while now).
We have no means without going into the internal of the NN weigths to make wet or silicate implementations comparison about the pattern space size that each internal models of chess being shaped with experience would compare. We even have no notion about the size of external chess board in term of exact patterns. All we have as notions of sizes are cardinality bounds calculations. those assume that every distinct position is equally "different" from each other, if differing by one mode for example (that is the game world) or one piece location difference (I guess if counting through distinct FEN it could be just any difference that would count as one in the cardinality point of view.
But yes. we can't hold our breath until the end of a game like A0 would before making inferences back to upstream positions. We build more chunking in the depth direction and use forecasters from clues on the board along the way. Our chess theory is not only turn by turn model like SF species have been built on through exhaustive tree searches, but it is not also about connecting each outcome to all the positions of the games in equal or many games redistribution of associations like the training of LC0 NN would do.
There is that added fog. That we have surrogate signals chunking such breadth. So that is not as exact as the judicious back propagation onto all the weights of each game outcome reward gain (or objective function distance about the win ratio from 1, forgot which exact functional formulation). The fog might be in how we value the combined signal as odds predictors, but there might be board non foggy corresponding patter that trained GM can have in common with the NN as they might also be part of how this backpropagation corrections might affect the position pattern recognition along all positions of the selfl play generated games. It would need research. but the arena emergency leitmotiv, is kind of obscuring that curiosity, another way it could actually be helping humans instead of fleeing forward in the stratospheres of ELOs.. in some pool of unidentifiable engine ecosystem.
The question is a good one. It is not as easy as the calculation model of SF and the Kotov model comparison. We are now in the world of patterns and another type of size of chess world. For such comparative question. there is work to do, on a rahter fundamental or dare I say, mathematic level.
I think the adult learning theories are inexistent and assuming that GM already learned thinking, or childhood intensive exposure requirement (in general, I gather), is the only implicit learning theory ever needed.
And the red herring of opening theory based memory task compeition, that is indeed an important factor as opening are already a chess game in itself, as one can allready have decisive offs imblanace if only knowing how to handle positions for their on board signals as a basis of life long experience. Increasingly so, the current tournment grooves, based on closed book preformance, as like the blog arguing to avoid cognitive offloading, for memory retrieval preformance task context.
(a more recent same line of experimental tools and cognitive theory, then changed the nature of task being measure to something I call a problem solving task, where the conclusion was not to put the book down for ever, it depends on what it is that is being the objective of learning. Is problem solving having anything to do with the chess passion. It is for me.
is it for OTB? So I think we should not jump to conclusion about adult learning. One thing adults have in their luggages, is some self-awareness of how they might have best learned difficult or complex tasks in the past. And self-awareness on their immediate understanding or ignorance, and whether a positoin is familiar or not. I think that sort of tool is not factored in the rarely explicit de facto leanring theory underlying lots of chess expreience or leanring tools. or teaching. I apologize for those which own activities would prove me wrong.
I may have missed it in the OP but it seems like the author is not considering de Groot's chunking experiment where masters were able to reconstruct actual chess positions much faster than non-masters, but when given positions with pieces placed randomly on the board masters were no better than non-masters at reconstructing the position.
I think this ability to "chunk" chess knowledge is of prime importance to achieving master level chess, but can it be learned later in life or is it something that needs to be hard wired at a young age?
With this "chunking" ability masters are able to prune the search tree drastically so that they are analyzing fewer lines than non-masters. They don't consider dubious moves like I do.
I'm guessing that someone like Praggnanandhaa isn't seeing the same board I am. I see pieces on squares. I imagine that Pragg is seeing a network of connections or relationships, which allows him to see "harmony" or "discord" and identify hot spots that need attention.
The original post is very interesting.
Thank you!
@SummerThereof said in #14:
> I may have missed it in the OP but it seems like the author is not considering de Groot's chunking experiment where masters were able to reconstruct actual chess positions much faster than non-masters, but when given positions with pieces placed randomly on the board masters were no better than non-masters at reconstructing the position.
>
> I think this ability to "chunk" chess knowledge is of prime importance to achieving master level chess, but can it be learned later in life or is it something that needs to be hard wired at a young age?
>
> With this "chunking" ability masters are able to prune the search tree drastically so that they are analyzing fewer lines than non-masters. They don't consider dubious moves like I do.
>
> I'm guessing that someone like Praggnanandhaa isn't seeing the same board I am. I see pieces on squares. I imagine that Pragg is seeing a network of connections or relationships, which allows him to see "harmony" or "discord" and identify hot spots that need attention.
>
> The original post is very interesting.
> Thank you!
Yes, not in the OP but very relevant to this topic! That is an excellent example of the kind of cognitive process that must indeed allow experts to transcend some of the key constraints on memory, perception , etc. Look for some future posts about De Groot and other researchers who have explored specific cognitive mechanisms that may support stronger chess play (and thanks for reading this essay!).
Interesting read. Looking forward to more.
@franksgambits said in #16:
> Interesting read. Looking forward to more.
Thank you for reading! Look for a new essay in the next few days.
Although most chess players believe that DeGroot's theory that chess is ll about recognizing patterns, recent research by Dr. David Lane at Rice University has shown this is not the case. His work is based on a simple thought experiment. Is chess similar to playing a piano where one has memorized the sheet of music so carefully that one can play all the notes correctly with the correct intonation even though the piece requires one to play very quickly? This might be for true bullet chess or when you memorize lengthy opening sequences or endgame positions. However, at slower time rates in novel positions chess might be more like music composition which is widely recognized to use thinking skills that go beyond simple pattern recognition toward actual problem solving. Yes it's true the composer must know all the scales, chords, music theory so the composer must use pattern recognition but there is much more to it then that. Musical composition requires problem solving skills and imagination. As you might expect chess has requires an enormous amount of pattern recognition but it goes beyond that and incorporates higher levels of thinking than just pattern recognition. That is what Dr. Lane found and that is what must modern cognitive psychologists agree with. In most chess positions there is no tactical device to achieve an advantage. Having your students study tactics will take them far but not all the way.
Also Degroot, I think was considering linguistic coginitoin as model for chess, if I not mistaken.
Reading music, and fingering skills are not the same kind of patterning, as board configuration informat and the dynamics it might have as logical possible futures. Improvising and composing music thought, might beusing higher level pattern about this execution skills set, and might be more like the 2D position spatial configuration skills, and their dynamic mechanics.
I think it would be time for a new more visuo-spatial point of view about patterns, as there may be new cognitive science experimental methods avaible, that then more sequence or how many words can a chess player store as patterns in the live expert brain.
There has been a lot of conceptual papers on chess cognition, but it seems they would all build on degroot experiements, and propose theories, but no with verification or testing. Also a lot of the existing studies keep forcuing on the already learne cognition, and we are left with extrapolation from general cognition and plasiticity development stage differences (although some past conception about losing plasticity, in adults, might be losing ground, as losing neurons, is not the same as variations in the synaptic pathways. am I wrong, that there are reconsideration of that monotonic decline in plasticity?).
I think Degroot may have great reference for then, with linguistics cognition serving as basis of modeling, but there have been more work with artificial visual cortex inspired architectures since them, and would not eye gaze tracking help build new experiements, and this time not just about expert, but all phases of learned states.. we might need a different models of what is chess skillset, that that of strenght or rating though. maybe that could also be explored. the variablitiy of the eye gaze behavior possibly givng hints of what on the board is being paid more attention to, across same strenght population "slices"...
Yes, this is indeed interesting. Our brain does not follow the same algorithm prior to each move (like an engine does) but it might be that we calculate deeply (and correctly) in position 1 but that we blunder in position 2 (because we just "move" instead of following the proper thinking routine). Even GMs sometimes make apparently "stupid" mistakes (like Fischer's famous Bxh2 move against Spassky, or Erigaisi in the game Abdusattorov - Erigaisi, where Black completely unforced blundered a whole rook (overlooking Bf6+ with double attack against K and R).
Sometimes we "know" immediately after the move that it was wrong. This is not a matter of lacking chess understanding but of lack of concentration or of too much psychological pressure (the engine never suffers from these issues). Some players are very cool under time pressure others make plenty of mistakes (and I once "forgot" about the clock altogether and lost a totally won position - but to be fair, I won once a game because my opponent "forgot" about the clock).
I think that these psychological factors (constant thinking routine, behaviour in time pressure, behaviour under pressure like a "must win" situation due to the team situation etc.) are very important. Some players have a great chess understanding (and pattern recognition) but suffer from these factors - which often results in won positions which are not "used" to get the full point (instead they end in a draw or even worse).
On the other hand there are players who are not extremely strong, but have a constant and psychologically stable attitude - even in lost positions. These players sometimes win "by luck" - because the (stronger) opponent got nervous and blundered his won position away (f.ex. by overlooking something very obvious).
I once managed to win like this myself. I played the opening and middlegame in a horrible fashion, my king was exposed and my opponent just had to find the "knockout" move with queens still on the board. He did not - I was lucky and escaped into a (queenless) endgame. But this endgame was also not pleasant for me. I defended and my opponent got a little impatient and made some inaccuracies so the game was finally equal (a N endgame with pawns on both sides).
He declined my draw offer and "played for a win" (because of the team situation - see above). But I activated my queenside majority - and finally won the game. This was - of course - partly luck, but on the other side my patient (but certainly not spectacular) defense payed off.
