Larry Kaufman, Math and "The Development of Chess Style

[CRoberson]

A few years ago Larry and I had a discussion on his Mathematics in calculating the value of the pieces. This is specific to his work published
before his work on Rybka. I claimed that the Math didn't find the perfect values it only finds the values of the data set. IOW, it only computes the
values used by those in the data set. Thus, using a data set of modern games only finds the values for modern play and not the perfect values.

This presents several opportunities.

In the 1980's, I purchased a book by World Chess Champion Dr. Max Euwe called "The Development of Chess Style". It is an excellent book and I recommend it to
all. In it, Dr. Euwe claims that people develop their personal Chess style on a parallel path to how man as a whole developed Chess style. This makes
sense, you don't teach Calculus before learning to Add and Subtract.

His chapters are excellently laid out in chronological order and each chapter depicts a particular era of Chess style. The time periods run from
the 1600's to the 1960's. I found Chapter 5 "Positional Play (Steinitz, 1836-1900)" very interesting and entertaining. Especially entertaining
was the negative public reaction to Steinitz's published theories before he became world champion and how he developed those theories.

So, it would be interesting to apply Larry's computations to each period identifying the differences in the knowledge. Based on the information in
the book, some of the knowledge learned is not in the form of piece values.

I would have sent this to Larry instead of global post, but I thought others might like the book and find this topic interesting. This general
concept leads to many other ideas and opportunities.

[hgm]

I don't think you are right. In my experience the values one finds are highly independent of the values used by the players. If I determine the value of a piece X, say a Chancelor (= Rook + Knight compound) by playing it in an imbalance X vs Queen, with self-play of an engine that uses Q=9.5 and X=9, I see the Queen wins by ~58%. Which confirms the X=9, as with Pawn odds the score would be ~65%. But if I set X=10, the Queen still wins ~58%!

There is just no masking that the Queen is stronger than the Chancellor, and will inflict more damage on the other pieces, which eventually will cause its side to win, as long as it stays on the board long enough. And if the values used are different, one side will try to avoid the trade (and usually be succesful in that, so that they indeed stay for most of the game). It doesn't matter much which side that is. It is about equally difficult for a Queen to force an (ill-advised) trade for a Chancellor as it is for the Chancellor to force a trade for Queen.

So when both players share the same misconception on the piece values, there hardly is any effect. This would not work for grossly wrong values; if you tell the engines a Queen is worth more than a Pawn, there is of course no way the side with the Pawn could avoid it to be traded for a Queen, no matter how hard the Pawn side tries to. The most dangerous case is actually where you use equal values, because then neither side would object to quick trading, and any real difference would be masked by that.

So the measured values ar far from being self-fulfilling prophecies. If anything, setting a value of a piece too high, so that it gets above that of a similarly valued piece, while it should be below, would lower the score of the side having that piece, and thus the value of the piece derived from it. Because you force a trade-avoiding strategy on that piece, restricting its usefulness, while in fact it had nothing to fear, and should on the contrary have forced a trade-avoiding strategy on its rival. The higher the value you put in, the lower the value that comes out! This effect causes the value of Knights and lone Bishops to be equal to such extraordinarily high precision, which would otherwise have been an extreme coincidence for such very different pieces.

[Uri Blass]

I don't think you are right. In my experience the values one finds are highly independent of the values used by the players. If I determine the value of a piece X, say a Chancelor (= Rook + Knight compound) by playing it in an imbalance X vs Queen, with self-play of an engine that uses Q=9.5 and X=9, I see the Queen wins by ~58%. Which confirms the X=9, as with Pawn odds the score would be ~65%. But if I set X=10, the Queen still wins ~58%!

There is just no masking that the Queen is stronger than the Chancellor, and will inflict more damage on the other pieces, which eventually will cause its side to win, as long as it stays on the board long enough. And if the values used are different, one side will try to avoid the trade (and usually be succesful in that, so that they indeed stay for most of the game). It doesn't matter much which side that is. It is about equally difficult for a Queen to force an (ill-advised) trade for a Chancellor as it is for the Chancellor to force a trade for Queen.

I expect the side with the queen to score less than 58% if both evaluate X=10 Q=9.5

Suppose White has queen and black has X and both evaluate wrong X=10 Q=9.5
The following can certainly happen
1)White allow bad trade of queen for X because bad trade win 0.5 pawn based on white's evaluation.
2)Black is going to accept the trade not because of evaluation but because black can search one ply deeper and see that avoiding trade is even worse.

Uri

[hgm]

I expect the side with the queen to score less than 58% if both evaluate X=10 Q=9.5

Well, so what you expect and what actually happens are apparently different.

I guess that it equally-often happens that black could avoid a loss of 0.5 Pawn in other evaluation by trading, but refrains from doing so because he unjustly thinks the trade in itself loses also half a Pawn.