Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On the Effect of Cheating in Chess

Daniel Keren

TL;DR

This paper quantifies how much performance can be gained from limited engine-assisted cheating in chess by evaluating fixed budgets of oracle interventions within engine-vs-engine games. It develops multiple intervention strategies, including fixed-threshold policies and maximal-delta predictors, and introduces an engine-free stochastic model to rapidly tune thresholds using logged data. Results show substantial uplift from even small budgets (e.g., up to approximately $0.91$ with $n=4$ interventions) and demonstrate strong agreement between engine-free optimization and full-engine experiments. The work provides a practical framework for assessing the impact of cheating under constraints and offers insights to inform detection, policy, and fair-play enforcement, including considerations when translating findings to human-vs-human contexts.

Abstract

Cheating in chess, by using advice from powerful software, has become a major problem, reaching the highest levels. As opposed to the large majority of previous work, which concerned {\em detection} of cheating, here we try to evaluate the possible gain in performance, obtained by cheating a limited number of times during a game. Algorithms are developed and tested on a commonly used chess engine (i.e software).\footnote{Needless to say, the goal of this work is not to assist cheaters, but to measure the effectiveness of cheating -- which is crucial as part of the effort to contain and detect it.}

On the Effect of Cheating in Chess

TL;DR

This paper quantifies how much performance can be gained from limited engine-assisted cheating in chess by evaluating fixed budgets of oracle interventions within engine-vs-engine games. It develops multiple intervention strategies, including fixed-threshold policies and maximal-delta predictors, and introduces an engine-free stochastic model to rapidly tune thresholds using logged data. Results show substantial uplift from even small budgets (e.g., up to approximately with interventions) and demonstrate strong agreement between engine-free optimization and full-engine experiments. The work provides a practical framework for assessing the impact of cheating under constraints and offers insights to inform detection, policy, and fair-play enforcement, including considerations when translating findings to human-vs-human contexts.

Abstract

Cheating in chess, by using advice from powerful software, has become a major problem, reaching the highest levels. As opposed to the large majority of previous work, which concerned {\em detection} of cheating, here we try to evaluate the possible gain in performance, obtained by cheating a limited number of times during a game. Algorithms are developed and tested on a commonly used chess engine (i.e software).\footnote{Needless to say, the goal of this work is not to assist cheaters, but to measure the effectiveness of cheating -- which is crucial as part of the effort to contain and detect it.}
Paper Structure (29 sections, 11 equations, 2 figures, 2 tables, 1 algorithm)

This paper contains 29 sections, 11 equations, 2 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Previous work
  3. Chess engines (brief overview)
  4. Previous work on detecting cheating in chess
  5. Engine analysis and its limits.
  6. Statistical/Bayesian decision models.
  7. Transparent/behavioral rules and learning.
  8. Classic man--machine milestones: Junior/Deep Junior and Deep Blue
  9. Modeling human move behavior in chess
  10. Notable cases
  11. Other work on assistance/cheating/intervention, and our contribution
  12. Setting
  13. Modifying the WDL value
  14. Algorithms
  15. Fixed thresholds
  16. ...and 14 more sections

Figures (2)

  • Figure 1: Monotonic fitting vs. engine WDL. Top: isotonic regression. Bottom: monotone neural net. The higher (for $x \geq 0.5$ and lower (for $x \leq 0.5$) for move 30 vis-a-vis move 5 reflect the fact that the outcome is easier to predict when the game is in a more advanced stage.
  • Figure 2: Example of how "strong" moves (by ${\cal \bf{C}}$ improve on "weak" ones (by ${\cal \bf{W}}$): when tuned to ELO 3190, the engine finds a better move than the ELO 1500 engine, as it realizes the danger of Black doubling rooks on the A-file.