Pattern Recognition—Fact Or Fiction?

Chess players, chess authors and chess psychologists attempt to teach and explain chess playing with the concept of "pattern recognition":

Pattern recognition is one of the most important mechanisms of chess improvement. Realizing that the position on the board has similarities to positions you have seen before helps you to quickly grasp the essence of that position and find the most promising continuation (van de Oudeweetering, 2014).

This may sound familiar:

"The acquisition of chess patterns is the main ingredient for chess mastery" (Silman, 2010, p. 638).

"After working with this book, an increasing number of positions, pawn structures and piece placements will automatically activate your chess knowledge. As a result you will find the right move more often and more quickly" (van de Oudeweetering, 2014).

Or:

"Once the reader has started applying the patterns in IYCPR ("Improving Your Chess Pattern Recognition") in their own games, they will find that the post-opening phase of the game becomes easier and they will more often build up a strong position" (GM I. Rogers, quoted by van de Oudeweetering, 2014, p. 7).

However, closer examination reveals causal, conceptual, epistemological and practical problems needing to be dealt with.

Practical Problems

Nature (Amidzic et al., 2001, p. 603) informs us that GMs need to learn 100,000 patterns and that pattern familiarity is what distinguishes grandmasters from more ordinary players. Does this mean that GMs, when asked, can set up pattern nr. 75,836? 54,723? 47,628? 17,319? Stating there are a 100,000 patterns, plausible would seem to assume that we could separate them individually.

However, the number appears rather random. How are numbers of patterns delimited and measured? Are lower-rated grandmasters from 2500-2600 familiar with fewer patterns than grandmasters from 2600-2850?

If "pattern" is understood as "piece configuration," acquisition of chess skills appears even more cumbersome, depending on how fast one can set up different positions either on a board or on a computer screen.

On average, with 10 positions learned a day, it takes 27 years to acquire 100,000 patterns which, in turn, makes it hard to explain how young super-grandmasters, like Magnus Carlsen (grandmaster at 13, learned chess at age 8, which suggests 20,000 patterns a year and 55 patterns a day) can at such a tender age can be much stronger than older grandmasters who have had much more time to acquire far more patterns.

Does anyone else think eight seems a late learning age for a world champion?

Alternatively, do different grandmasters have familiarity with different sets of patterns which reflects the rating differences? Since nobody knows every possible pattern, there is no way to know if players are acquiring useful patterns or if they are wasting time. Perhaps the way to beat a grandmaster is to get non-pattern positions on the board. This will shut down much of what the grandmaster has on his personal "hard drive."

If pattern recognition is the key to chess strength, we are hard pressed how to explain why performance decreases with age as one would think that a chess player could with some effort bring back relevant patterns and keep their performance at peak.

Do older grandmasters' performances decrease because other cerebral factors (otherwise unaccounted for) interfere with the grandmasters ability to reproduce relevant patterns during a game? This illustrates that chess playing is more than mechanical reproduction of patterns, and it is impossible to determine how much is pattern recognition, and what is ability to produce high-quality moves of one’s own accord.

Does chess that easily lend itself to mechanical recipes? Is acquisition of chess skills no more difficult than to say: Learn these gazillion patterns, and you will become a 2800+ player? In the same vein, are grandmasters able to reproduce all the patterns they recognize and are familiar with?

Another fundamental problem is that as long as thinking and learning are subconscious (see our previous installments here and here), there is no way to tell if the brain perceives different positions as patterns, but psychologists and chess writers seem to try to make pattern acquisition look like a mechanical, conscious process. They make a pattern a straightforward way to explain human behavior, but it does not explain the real course of events.

The only way is to reverse the process. What is called a "pattern" is something established after the so-called patterns are learnt, and we rationalize and justify what happens to make it possible to make recipes and write instructional texts. When learning it, we do not know if what we have before us is a pattern so a question is: How and when does chaos/randomness transform into pattern in the chess mind? Learning chess resembles Wittgenstein's idea on how to learn to follow a rule; there must be something going on underlying our ability to understand rule-bound instructions before we know the rule, and it is the same with chess.

Moving on to conceptual issues, we conclude that there seem to be too many problems linked to the practical use of the concept of pattern recognition to give it the explanatory force ascribed to it.

Conceptual Problems

At first glance, "pattern" seems to indicate something like a whole or some sort of totality, repeating itself infinitely in its entirety (otherwise, how else are to know we’re dealing with a pattern?), like patterns on bed sheets or tablecloths, but what constitutes a chess pattern?

Without defining chess pattern, van de Oudeweetering presents us with the following sample positions which he refers to as "positional patterns" (van de Oudeweetering (2014, p. 9):

What makes these positions patterns and not just examples of where a good move is at our disposal? Is it because we know how the pieces move? We know we can plant a knight on d6 or f5 because we know how the knight moves. Where do these positions’ "patternness," or "patternicity," come from? What are we supposed to perceive as the pattern?

(Errata: Sveshnikov-Tseshkovsky is on page 165)

Circular Similarity

"[...] Realizing that the position on the board has similarities with something you have seen before helps you to quickly grasp the essence of that position and find the most promising continuation," van de Oudeweetering (2014) tells us and presents us with this position:

The caption to this position is: "Here's a similar position." (van de Oudeweetering, 2014, p. 257). Similar to what? The position on the preceding page?

Or an even earlier Petrosian game:

Or what about these three thematic IQP positions?

Our sample positions bring us to the core question: When and how does a position become paradigmatic, i.e. a position we later on compare other "similar" positions to?

When the old masters played their games that later on became our patterns, what did they model their play after? What, if anything, served as a previously established model or pattern for their play? The reference to the classics might lead us to believe that all patterns are already discovered, or are there games being played by today’s contemporary masters that will become paradigmatic for future reference?

Due to the diversity of chess and the characteristic of identical repetition inherent in the concept of pattern as a means for explaining acquisition and development of chess skills, the concept seems problematic, and what strikes us when using the word pattern in connection with chess is the apparent abyss of discrepancy between the two, and how are the two to be reconciled?

As long as repetition is an inherent component of the concept, and each and every position appears different, single or individual positions can never constitute a pattern. Unlike patterns on bed sheets or table cloths, where we don’t have to check out the whole bed sheet or table cloth to see if the pattern repeats itself, we have to examine and take in the whole position before making our move.

A problem is how to generalize different positions into the same definition of pattern, which appears impossible, since the positions are different, and no player will live long enough to see if a position repeats itself and thus be able to establish a pattern. Due to the diversity of chess, there will always be a principal problem of formalizing a pattern definition comprising the ever occurring unique positions while not violating the notion of identical repetition.

A fianchetto castling position may to many appear to be a pattern since this specific configuration is known to repeat itself numerous times in different kinds of positions.

To show that this is anything but philosophical nit-picking, we might mention that Kant (1724-1804) brought to our attention the fact that concepts are never defined by their use. The color "red" might illustrate the point in question. If we ask someone what red is, most will point at cars, pictures or books. These are mere instances of red, but they do not define what red is. I.e. they do not delineate red in contrast to, for instance, blue or green. Back to topic: different kinds of castlings or mating images as patterns are problematic for three reasons:

• The positions as instances do not yield any definition of pattern (analogous to red).
• The positions are at once different, where the question of how to incorporate these into a single definition of pattern comprising both (as pointed out) and
• Castlings differ from mating images in that the former appear more static (fixed configurations) than the latter (ever changing configurations of pieces) where more precisely than (static) patterns would seem to be to speak of tactical operations or combinations, which differ from patterns.

Two implications seem to follow from this: If a single definition, incorporating different positions, cannot be given, diverse positions cannot serve as examples of patterns, or we are applying one definition of pattern to each and every position, but this does not solve the problem and leads to a circle. Again we might ask what is it about castlings that make these constitute or make up a pattern, and thus we are right back to the problems of definitions. For the question of development of chess skills to be solved, of utmost importance is that concepts are defined before being used or applied.

The perceptive reader may have noted the difficulty in pinpointing when a pattern occurs, since a single move results in a new position, and the whole position must be considered when evaluating what move to make. Chess has to be played according to specific circumstances, not some generalized idealizations. Will a change in a position of a pawn imply a change in an existing pattern within a position, or are we talking about a different pattern due to a new position?

Pattern vs Structure

Without distinguishing between them or elaborating on their interrelatedness, van de Oudeweetering (2014, pp. 10-11) introduces other related/similar but not identical concepts like ideas and themes, and we might add, motifs (ITM), configurations, constellations and set-ups (CCS), in short: structures, though granted that CCS sound more physical than ITM and even if the latter suggest a more mental or abstract aspect detached from their physical manifestation, they must somehow still be connected to the pieces, either physical or mentalised (Hearst & Knott, 2013; Mechner, 2010).

Contrary to "patterns," "structures" might be defined as a certain smaller configurations or distributions of forces occurring within certain delimited sectors of the board within positions as a whole and which might be repeated without the whole context within which they occur having to identically and infinitely repeat itself. Smaller parts or fragments within a greater totality will repeat themselves more often than whole positions, if at all. Even if only having remnants or fragments of broken or shattered structures, at a glance we recognize the contours of an intact structure within positions as a whole regardless of what the rest of the board looks like.

In other words, when seeing a broken castled position with pawns on f7, f6 and h7/h6, in an instant we spot the possibility of a knight on f5. This has more to do with recognizing a structure on a restricted part of the board rather than an all-comprising pattern.

Familiarity with structures might facilitate the speed of calculation, but still more important appears the ability to produce concrete, correct moves as otherwise we would be hard pressed to explain how super-strong young grandmasters like Carlsen or Karjakin so well handle and play positions never seen before, having had less time than more experienced (by age) players to be acquainted with all these new structures.

Karjakin's thinking face is definitely 2800+.

Also, chess discourse seems to suggest that "pattern recognition" is not the most accurate of terms when explaining acquisition of chess proficiency. On DVDs or books, we never encounter the term "pattern," only "structure;" we "weaken/change/ruin/expand/improve the structure of the position," not "the pattern of the position."

"Pattern recognition" therefore more appears to be an idealized simplicity rather than a concept apt to explain acquisition of chess skills, paving the way for the question; how do we define what a pattern is and if "structure recognition" is acquired by playing and studying chess, how does "pattern recognition" relate to this?

The problems related to the concept of pattern seem to apply to other aspects of acquisition of chess skills as well: How to improve our position without any preconceived idea of what the improvement consists of? How many examples of exchanges do we have to work through before mastering the art of exchanging pieces? Not to mention when not to exchange? And what about the exchange sacrifice? How many examples before one masters the art of relinquishing our bishop pair, grabbing space or playing "the right rook"?

As long as chess is played with the same number of pieces moving the same way on the same number of squares, players will always encounter positions "similar" to something they have seen before, so what relevant sense of similar are we talking about?

Regarding the question of learning and what to look for, studying the games of the masters appears akin to the paradox we face in one of Plato’s dialogues: How to search when not knowing at all what you are looking for? How to search for something you do not know at all? If finding it, how will you know that this is what you did not know?

Epistemological Internalism

"After working with this book, an increasing number of positions, pawn structures and piece placements will automatically activate [italics, ours] your chess knowledge […]" (van Oudeweetering, 2014).

or

"Once the reader has started applying [italics, ours] the patterns in IYCPR (Improving Your Chess Pattern Recognition) in their own games, they will find that the post-opening phase of the game becomes easier and they will more often build up a strong position." (GM I. Rogers, quoted by van de Oudeweetering, 2014, p. 7).

The idea that right knowledge or insight leads to (morally) right action and thought inextricably related has long and rich traditions with perhaps Socrates as its foremost proponent and might be called moral internalism but applies to morally neutral questions, issues and matters as well, denoted epistemological internalism.

If knowledge is to initiate actions or behavior (fingers picking up and letting go of pieces), which are physical effects and therefore need a physical cause (a non-physical consciousness cannot trigger actions, i.e. cause arms and legs to move), assumed chess patterns in our case, need thence be physically represented in the brain with the discussion revolving around how knowledge is represented.

However, a concept of chess pattern still to be defined renders well-nigh impossible to say what this knowledge looks like and/or how it is to be physically represented in our brain (not to mention estimating their number) and leaves us hard-pressed to explain how the concept is causally related to our playing and other chess knowledge, say the afore-mentioned art of exchanging pieces, exchange sacrifices, relinquishing of the bishop pair, playing the right rook, mastering opposite-color bishop endings, exploiting weak color-complexes etc. (Note in passing that the causal problems associated with the presumed knowledge of patterns apply to this knowledge as well) Tarrasch’ aphorism, "It is not enough to be a good player … you must also play well," nicely catches the gap between knowledge and action. The same applies to math: "It is not enough to be good at math … you also have to do the math."

Tarrasch actually learned chess at 15.

Conclusion

Lack of a precise definition of the concept of pattern raises the question of what we are supposed to recognize (no patterns—no pattern recognition) and how the concept causally is to automatically activate our other chess knowledge.

"Automatic activation" of chess knowledge remains problematic as every position has to be assessed and played according to its own accord where blunders reduce the value of assumed pattern recognition to zero, suggesting that the triggering of moves works unaided of the ability to remember patterns or any other chess knowledge. We cannot "start to apply" the presumed patterns both because these patterns are undefined and chess playing is subconscious (or more precisely, a fine-tuned interplay between brain and consciousness.)

Finally, if there were a necessary connection between knowledge and action (behavior), there seems to be reasons to believe the world would be a different place.

In our next installment, we’ll present a method to improve our overall play without having to rely on or resort to static generalizations or idealized simplicity.

.

Bibliography

Amidzic, O., Elbert, T., Fehr, T., Riehle, H. J., & Wienbruch, C. (2001). Pattern of focal ɤ-bursts in chess players. Nature, 412 (6847), 603. https://doi.org/10.1038/35088119

Hearst, E., & Knott, J. (2013). Blindfold chess: history, psychology, techniques, champions, world records, and important games. Farland. https://doi.org/10.1901/jeab.2010.94-373

Lipnitsky, I. (2010). Questions of modern chess theory. Quality Chess.

Mechner, F. (2010). Chess as a behavioral model for cognitive skill research [Review of the book Blindfold Chess by Eliot Hearst and John Knott]. Journal of the Experimental Analysis of Behavior, 94 (3), 373–386. https://doi.org/10.1901/jeab.2010.94-373

Silman, J. (2010). How to reassess your chess 4th edition. Siles Press.                                   

Van de Oudeweetering, A. (2014). Improve your pattern recognition: Key moves and motifs in the middlegame. New in Chess.

Vik-Hansen, R. (2016, 9 February).  Why chess players blunder. Chess.com.           https://www.chess.com/article/view/why-chess-players-blunder  

Update 18 August 2019: Link inserted to "next installment." (Thanks to "Machariel" for making us aware).

Update 4 April 2021: Bibliography updated according to APA 7 ('Amidzic et al., 2001' updated accordingly.)

Update 5 October 2021: Bibliography updated according to APA 7 (Inserted a comma between initial and ampersand ('Hearst & Knott, 2013)

Acknowledgement

Again a heartfelt thank you to our dear friend and colleague, Victoria W. Guadagno, for taking time to proofread both the 2008 version and this reworked edition.

Dr. Philos Ståle Gundersen for clarifying the status of knowledge.