# Objectively, do playing styles exist?

I think nobody here will disagree to call Tal a master of attack and Petrosian a master of defense, or Kasparov more tactical and Karpov more positional. But then, such differences would be moot for a hypothetical 32-piece tablebase. Can, what we call "playing style" (cf. "Spielertypen" by Müller & Engel, claiming there exist 4 1/2 types) be formulated in objective terms? (Offhand example: make a statistic of how close their own officers are to the enemy king in an average game.) Links to scientific papers are welcome.

• what do you mean "such differences would be moot for a hypothetical 32-piece tablebase"? I think that if you had such a tablebase, you could not only always win (against humans) but also choose what playstyle you want to win with. Feb 12 at 17:55
• Why it is relevant: well, I suspect that maybe easy statistical heuristics like your example wouldn't cover it, and you'll need to rely on some game-theoretic measures regarding evaluations, how exactly they try to make the position harder for their opponent (to find evaluation-preserving lines in), etc. Feb 12 at 18:05
• @MobeusZoom: yes, I anticipated this objection :-) and it's reasonable that more than one move is "correct" in an average position, but chess masters had different opinions on that. Tarrasch wouldn't have agreed with you. Feb 12 at 21:04
• Leonid Stein: "I think I have my own style". This question is the subject of the first chapter of V. Eingorn great book, Decision-Making at the Chessboard: gambitbooks.com/books/Decision-Making_at_the_Chessboard.html Feb 14 at 14:37

I seriously doubt there are scientific papers on this. "Player styles" is probably impossible to define "objectively"/scientifically to the point where one can give a list of definite, rigid criteria that can be used by a computer to determine a player's style based on their games in a database such that everyone agrees that the computer is correct. In the end, what constitutes a distinct player style has to be arbitrarily determined at some point. In fact, I'd go so far as call an attempt to "scientifically define" player styles in this way a fool's errand.

The best kind of study one could hope for is one that tries to find clear statistically significant differences between groups of players (as chessplayers) who have different self-reported playing styles (out of a few listed), and in this sense of playing style I would be surprised if the study did not find clear differences between players who claim to be "attacking" vs. players who claim to be "solid", etc. How many broad "styles" that one should try to define is similar to how one should determine the average values of different pieces, i.e., one should do it by experience.

As a final note, one could also ask questions about objectively analyzing a specific player's unique chess style. As an example, if we feed every game that Magnus Carlsen has played up until 2021 into a neural network as training data, would the network be able to successfully pick out the games from TATA steel 2022 that were played by Carlsen? In other words, can a chess game bear the mark of Magnus Carlsen? The exciting thing about this question is that even if we are unable to describe in detail what "the mark of Magnus Carlsen" is, we can still test for whether it's there using this kind of method.

• I agree that it is highly interesting if a NN could detect Carlsen-ness. But you run into the classic Hofstadter trap - if such a NN exists, then a math formula exists, and Carlsen-ness is objectively measurable, even if we have no idea how the NN works. And even if it is only statistical because Carlsen had a bad hair day - statistics is an objective science too. Feb 12 at 21:09
• @HaukeReddmann Of course I realize this, which I thought my final sentence indicated perfectly well. Just because something can theoretically be described in full detail does not mean there is a way to describe it perfectly in a way that is understandable by a human being. The limitations of the human mind, while well-known, often get ignored in these kinds of discussion, and in my opinion often unjustifiably so. After all, what is an explanation if there is no mind able to understand it? Feb 12 at 23:44
• My comment was more directed to MobeusZoom than you; in any case, get an comment uppie from me for hammering a point that never shall be forgotten. Feb 13 at 9:00
• @HaukeReddmann I'm not sure I get it. My point was that, if you do want whitebox heuristics, they are likely to require considerations of computer evaluations and/or the game-theoretic nature of the moves chosen (how exactly Kasparov tries to push for a complicated/difficult position, as opposed to Carlsen). If you are happy with blackbox methods for individuals' playstyles then this has already been done: cs.toronto.edu/~ashton/pubs/maia-personalized2021.pdf Feb 13 at 19:34
• @MobeusZoom: My argument is that the labels "whitebox" and "blackbox" merely measure our own human stupidity :-), but nevermind and many thanks for that link! Feb 14 at 8:28

The only way to answer that question seems to be to perform a blind test. Can player A, who has played B many times before, recognize B (to whatever extent), in a series of blind games against him/her under similar playing conditions. Statisticians would have to decide just how many games that would be involve, and how many A-B pairs would need to play, as well as A-C pairs and games (where C is unknown to A, to guard against false positives, or decide just may be recognized) to produce a result with enough confidence.

If not, ... we probably are no further. But if there is a statistically significant 'recognition', we might as well call it 'style' as long as we don't have a better definition of the term.

I'm afraid that something like 1000 pairs of A-B people, 500 pairs of A-C people, and something like 50 games between each would be required, making a test impractical. But I'll leave that to the experts on statistics.

• See the link MobeusZoom gave in his comment above! Feb 14 at 8:30