Tweet
Piece valuation methods - Google AI's Q&A
#1)
Based on contents of one Google search AI answer to query: 'did kaufman ever test his piece valuation methods by using a large database of engine self-play games of chess against itself instead of using a large database of 2300+ vs 2300+ human played games of chess'? (note: AI answer per search tried may vary, even if same keywords used):
did kaufman ever test his piece valuation methods by using a large database of engine self-play games of chess against itself instead of using a large database of 2300+ vs 2300+ human played games of chess - Google Search
No, Larry Kaufman never used a large database of engine self-play games to test or derive his piece valuation methods. [1, 2, 3]
When Kaufman conducted his landmark 1999 and 2008 studies on material imbalances (detailed in articles like The Evaluation of Material Imbalances), his datasets consisted exclusively of hundreds of thousands of human-played games where both players had ratings of FIDE 2300 or higher (Grandmaster and International Master standards). [1, 3]
He utilized human grandmaster data to solve the question of material imbalance precisely because he wanted his conclusions to reflect practical, high-level human play rather than the tactical blind spots and algorithmic biases of earlier chess engines. [1, 2]
#2)
Contents of one answer (including the contents of one answer to a shorter follow-up question) to a Google AI query I made (note answers may vary per search even if same keywords used): 'if kaufman's piece values are made using samples of games that include games between players rated as low as 2300 elo can such methods somehow be soundly and similarly used in the case of evaluating the worth of novel fairy chess pieces if for example very few 2300+ elo chess players have played chess variants that use such a given fairy chess piece type'?:
if kaufman's piece values are made using samples of games that include games between players rated as low as 2300 elo can such methods somehow be soundly and similarly used in the case of evaluating the worth of novel fairy chess pieces if for example very few 2300+ elo chess players have played chess variants that use such a given fairy chess piece type - Google Search
Applying IM Larry Kaufman’s statistical approach to evaluate novel fairy pieces is a sound concept, but it requires adaptation since human data is scarce. Instead of using human tournament games, you can soundly apply his methodology using modern chess engine self-play. [1, 2, 3]
Why Kaufman's Method Works in Theory
Kaufman’s core method from his famous "Evaluation of Material Imbalances" didn't calculate absolute point values. Instead, he analyzed a massive database of human games to determine how many pawns were required to perfectly balance a specific material trade (e.g., how many pawns compensate for giving up a knight). [1]
At its foundation, this is a statistical and game-theoretic approach to expected value. As long as the dataset is large enough and the players are capable enough, the method derives the true exchange worth of a piece based purely on its power and outcomes. [1, 2, 3]
How to Adapt the Method for Fairy Pieces
Because very few high-level players practice variants with novel fairy pieces, you cannot rely on human databases. You can easily adapt the approach using the following methodology: [1]
Why Rating Doesn't Matter for Engine Games
In traditional chess, Kaufman filtered games to players rated \(2300+\) to ensure conclusions were based on the play of experts, avoiding the "noise" of beginner blunders. [1]
You said: is dr muller only qualifies person investigating fairy chess piece values
is dr muller only qualifies person investigating fairy chess piece values
While Dr. H.G. Muller is arguably the most famous pioneer in quantifying fairy chess piece values, he is not the only qualified person, nor do his methodologies exist in a vacuum. His work is brilliant, but other researchers, mathematicians, and programmers have developed equally sound, parallel systems for evaluating novel pieces. [1, 2]
The evaluation of fairy chess pieces is split across distinct experts and methodologies: [1]
1. The Mathematical & Formulaic Pioneers
Dr. H.G. Muller: Famous for creating the "Short-Range Leaper Law", which uses a quadratic formula (\(\text{Value} = \alpha[30N + 0.625N^2]\)) based purely on a piece's static mobility (N, the number of squares it attacks). [1, 2]
Ralph Betza: Long before modern engine testing, Betza developed the "Betza Notation" and a highly logical framework for calculating piece values based on components (e.g., valuing a piece by adding its "step" components, "leap" components, and adjusting for color-blindness penalties). [1]
2. The Statistical & Machine Learning Experts Why You Can Trust Multiple Experts
While Dr. Muller's formulaic laws are excellent for estimating a piece's value out of context, game-data analysis (the method used by Kaufman and modern engine devs) is inherently more accurate for actual gameplay. A static formula cannot perfectly predict contextual mechanics like the "levelling effect" (how a weaker piece can easily harass a stronger fairy piece) or complex piece synergy. [1, 2, 3, 4, 5]
Leverage Engine Self-Play: Use a variant engine (such as Fairy Stockfish) to run hundreds of thousands of self-play games. Set the engine to high depth to ensure high-quality play and eliminate blunders. [1, 2]
Isolate Imbalances: Force the engine to play positions with material imbalances (e.g., starting Black with an Archbishop instead of a Bishop and Knight). [1, 2]
Analyze Outcomes: Calculate the win/loss/draw ratio to find exactly how many pawns (or other standard pieces) are needed to bring the expected win rate to \(50\%\). If giving White an extra pawn still leaves the game at \(45\%\) win-rate for the fairy piece, the piece is worth more than that. [1, 2]
The Engine Equivalent: With engines like Fairy Stockfish, you do not need to filter by rating. Modern engine evaluations—even over short time controls—do not suffer from the strategic blunders found in human lower-rated play. The engine's evaluation perfectly mirrors expert human understanding of tactical consequences, meaning the baseline "play quality" condition of Kaufman's method is inherently met...
Fabien Letouzey (and the Fairy Stockfish contributors): Authors of modern chess variant engines do not guess piece values. They use logistic regression over millions of computer self-play games to observe win/loss ratios. This acts as the exact modern extension of Larry Kaufman's database approach, substituting human GMs with high-elo engines. [1, 2]
Data Scientists / Bloggers (e.g., "ubdip"): Independent researchers have published automated scripts that use machine learning tools to calculate piece values from scratch for any custom rule-set by training a winning-chance predictor against material differences. [1, 2]
#3)
One Google AI answer to query: 'what is alpha zero average value for a single knight and a single bishop and what bonus value on average does alpha zero give if one side holds the bishop pair' (note answer may vary per query even if same search keywords used)?:
what is alpha zero average value for a single knight and a single bishop and what bonus value on average does alpha zero give if one side holds the bishop pair - Google Search
AI Overview
AlphaZero calculates the value of a single knight at 3.05 points and a single bishop at 3.33 points (using a pawn as 1.0). It gives a 0.5-point bonus for holding the bishop pair. [1, 2, 3]
AlphaZero vs. Human Piece Valuations
While human players traditionally treat the knight and bishop as equal (3 points), AlphaZero’s neural networks discovered that bishops are fundamentally more valuable due to their long-range versatility and board control. [1]
Pawn 1.00 1.0
Knight 3.05 3.0
Bishop 3.33 3.0
Rook 5.63 5.0
Queen 9.50 9.0
The Bishop Pair Advantage
AlphaZero highly values having both bishops over having two knights or one of each. The +0.5 bonus reflects the ability of the bishop pair to control both light and dark squares, coordinate dynamically in open positions, and restrict the opponent's pieces
#4)
Contents of one Google AI answer (plus to a follow-up question, later) to 'what is the fairy-stockfish valuation of a mann on an 8x8 chessboard and does it provide a fighting valuation for a chess king of any kind in terms if fighting value' (note answer may vary per search tried even if with same keywords used)?:
what is the fairy-stockfish valuation of a mann on an 8x8 chessboard and does it provide a fighting valuation for a chess king of any kind in terms if fighting value - Google Search
In Fairy-Stockfish, a mann is valued at approximately 3 points (equivalent to a standard knight or bishop). [1]
Because a mann shares the exact same movement mechanics as a standard king—but lacks the royal property of check or checkmate—this valuation provides a highly accurate 4-point fighting value for the chess king in the endgame, once it becomes an active attacking piece and is no longer at risk of being checkmated. [1, 2]?
what is the estimated chess elo rating of fairy-stockfish
At maximum strength, the estimated maximum Elo rating of Fairy-Stockfish is between 3500 and 3650 Elo, heavily depending on the hardware it runs on, the specific variant being played, and the time control. [1, 2, 3, 4]
Because it is directly branched from the core Stockfish engine—which currently leads the Computer Chess Rating Lists (CCRL) with a peak rating of around 3650 Elo—it inherits virtually all of its parent engine's tactical and computational power. [1, 2, 3, 4]
#1)
Based on contents of one Google search AI answer to query: 'did kaufman ever test his piece valuation methods by using a large database of engine self-play games of chess against itself instead of using a large database of 2300+ vs 2300+ human played games of chess'? (note: AI answer per search tried may vary, even if same keywords used):
did kaufman ever test his piece valuation methods by using a large database of engine self-play games of chess against itself instead of using a large database of 2300+ vs 2300+ human played games of chess - Google Search
No, Larry Kaufman never used a large database of engine self-play games to test or derive his piece valuation methods. [1, 2, 3]
When Kaufman conducted his landmark 1999 and 2008 studies on material imbalances (detailed in articles like The Evaluation of Material Imbalances), his datasets consisted exclusively of hundreds of thousands of human-played games where both players had ratings of FIDE 2300 or higher (Grandmaster and International Master standards). [1, 3]
He utilized human grandmaster data to solve the question of material imbalance precisely because he wanted his conclusions to reflect practical, high-level human play rather than the tactical blind spots and algorithmic biases of earlier chess engines. [1, 2]
#2)
Contents of one answer (including the contents of one answer to a shorter follow-up question) to a Google AI query I made (note answers may vary per search even if same keywords used): 'if kaufman's piece values are made using samples of games that include games between players rated as low as 2300 elo can such methods somehow be soundly and similarly used in the case of evaluating the worth of novel fairy chess pieces if for example very few 2300+ elo chess players have played chess variants that use such a given fairy chess piece type'?:
if kaufman's piece values are made using samples of games that include games between players rated as low as 2300 elo can such methods somehow be soundly and similarly used in the case of evaluating the worth of novel fairy chess pieces if for example very few 2300+ elo chess players have played chess variants that use such a given fairy chess piece type - Google Search
Applying IM Larry Kaufman’s statistical approach to evaluate novel fairy pieces is a sound concept, but it requires adaptation since human data is scarce. Instead of using human tournament games, you can soundly apply his methodology using modern chess engine self-play. [1, 2, 3]
Why Kaufman's Method Works in Theory
Kaufman’s core method from his famous "Evaluation of Material Imbalances" didn't calculate absolute point values. Instead, he analyzed a massive database of human games to determine how many pawns were required to perfectly balance a specific material trade (e.g., how many pawns compensate for giving up a knight). [1]
At its foundation, this is a statistical and game-theoretic approach to expected value. As long as the dataset is large enough and the players are capable enough, the method derives the true exchange worth of a piece based purely on its power and outcomes. [1, 2, 3]
How to Adapt the Method for Fairy Pieces
Because very few high-level players practice variants with novel fairy pieces, you cannot rely on human databases. You can easily adapt the approach using the following methodology: [1]
Why Rating Doesn't Matter for Engine Games
In traditional chess, Kaufman filtered games to players rated \(2300+\) to ensure conclusions were based on the play of experts, avoiding the "noise" of beginner blunders. [1]
You said: is dr muller only qualifies person investigating fairy chess piece values
is dr muller only qualifies person investigating fairy chess piece values
While Dr. H.G. Muller is arguably the most famous pioneer in quantifying fairy chess piece values, he is not the only qualified person, nor do his methodologies exist in a vacuum. His work is brilliant, but other researchers, mathematicians, and programmers have developed equally sound, parallel systems for evaluating novel pieces. [1, 2]
The evaluation of fairy chess pieces is split across distinct experts and methodologies: [1]
1. The Mathematical & Formulaic Pioneers
Dr. H.G. Muller: Famous for creating the "Short-Range Leaper Law", which uses a quadratic formula (\(\text{Value} = \alpha[30N + 0.625N^2]\)) based purely on a piece's static mobility (N, the number of squares it attacks). [1, 2]
Ralph Betza: Long before modern engine testing, Betza developed the "Betza Notation" and a highly logical framework for calculating piece values based on components (e.g., valuing a piece by adding its "step" components, "leap" components, and adjusting for color-blindness penalties). [1]
2. The Statistical & Machine Learning Experts Why You Can Trust Multiple Experts
While Dr. Muller's formulaic laws are excellent for estimating a piece's value out of context, game-data analysis (the method used by Kaufman and modern engine devs) is inherently more accurate for actual gameplay. A static formula cannot perfectly predict contextual mechanics like the "levelling effect" (how a weaker piece can easily harass a stronger fairy piece) or complex piece synergy. [1, 2, 3, 4, 5]
Leverage Engine Self-Play: Use a variant engine (such as Fairy Stockfish) to run hundreds of thousands of self-play games. Set the engine to high depth to ensure high-quality play and eliminate blunders. [1, 2]
Isolate Imbalances: Force the engine to play positions with material imbalances (e.g., starting Black with an Archbishop instead of a Bishop and Knight). [1, 2]
Analyze Outcomes: Calculate the win/loss/draw ratio to find exactly how many pawns (or other standard pieces) are needed to bring the expected win rate to \(50\%\). If giving White an extra pawn still leaves the game at \(45\%\) win-rate for the fairy piece, the piece is worth more than that. [1, 2]
The Engine Equivalent: With engines like Fairy Stockfish, you do not need to filter by rating. Modern engine evaluations—even over short time controls—do not suffer from the strategic blunders found in human lower-rated play. The engine's evaluation perfectly mirrors expert human understanding of tactical consequences, meaning the baseline "play quality" condition of Kaufman's method is inherently met...
Fabien Letouzey (and the Fairy Stockfish contributors): Authors of modern chess variant engines do not guess piece values. They use logistic regression over millions of computer self-play games to observe win/loss ratios. This acts as the exact modern extension of Larry Kaufman's database approach, substituting human GMs with high-elo engines. [1, 2]
Data Scientists / Bloggers (e.g., "ubdip"): Independent researchers have published automated scripts that use machine learning tools to calculate piece values from scratch for any custom rule-set by training a winning-chance predictor against material differences. [1, 2]
#3)
One Google AI answer to query: 'what is alpha zero average value for a single knight and a single bishop and what bonus value on average does alpha zero give if one side holds the bishop pair' (note answer may vary per query even if same search keywords used)?:
what is alpha zero average value for a single knight and a single bishop and what bonus value on average does alpha zero give if one side holds the bishop pair - Google Search
AI Overview
AlphaZero calculates the value of a single knight at 3.05 points and a single bishop at 3.33 points (using a pawn as 1.0). It gives a 0.5-point bonus for holding the bishop pair. [1, 2, 3]
AlphaZero vs. Human Piece Valuations
While human players traditionally treat the knight and bishop as equal (3 points), AlphaZero’s neural networks discovered that bishops are fundamentally more valuable due to their long-range versatility and board control. [1]
Pawn 1.00 1.0
Knight 3.05 3.0
Bishop 3.33 3.0
Rook 5.63 5.0
Queen 9.50 9.0
The Bishop Pair Advantage
AlphaZero highly values having both bishops over having two knights or one of each. The +0.5 bonus reflects the ability of the bishop pair to control both light and dark squares, coordinate dynamically in open positions, and restrict the opponent's pieces
#4)
Contents of one Google AI answer (plus to a follow-up question, later) to 'what is the fairy-stockfish valuation of a mann on an 8x8 chessboard and does it provide a fighting valuation for a chess king of any kind in terms if fighting value' (note answer may vary per search tried even if with same keywords used)?:
what is the fairy-stockfish valuation of a mann on an 8x8 chessboard and does it provide a fighting valuation for a chess king of any kind in terms if fighting value - Google Search
In Fairy-Stockfish, a mann is valued at approximately 3 points (equivalent to a standard knight or bishop). [1]
Because a mann shares the exact same movement mechanics as a standard king—but lacks the royal property of check or checkmate—this valuation provides a highly accurate 4-point fighting value for the chess king in the endgame, once it becomes an active attacking piece and is no longer at risk of being checkmated. [1, 2]?
what is the estimated chess elo rating of fairy-stockfish
At maximum strength, the estimated maximum Elo rating of Fairy-Stockfish is between 3500 and 3650 Elo, heavily depending on the hardware it runs on, the specific variant being played, and the time control. [1, 2, 3, 4]
Because it is directly branched from the core Stockfish engine—which currently leads the Computer Chess Rating Lists (CCRL) with a peak rating of around 3650 Elo—it inherits virtually all of its parent engine's tactical and computational power. [1, 2, 3, 4]