Towards the Automatic Synthesis of Interpretable Chess Tactics

Abhijeet Krishnan; Chris Martens

cs.AIOct 6, 2026

Towards the Automatic Synthesis of Interpretable Chess Tactics

Authors: Abhijeet Krishnan, Chris Martens

Organizations: North Carolina State University Venture IV, 1730 Varsity Dr, Raleigh, NC 27606

Abstract

State-of-the-art reinforcement learning agents are capable of outperforming human experts at games like chess, Go and StarCraft II. These agents do not simply take advantage of their digital hardware in being able to react and calculate faster than humans, but employ better strategies that lead to more victories. Interpreting these strategies would give human players valuable insight into how to improve their play. In this preliminary work, we propose a symbolic sub-policy model for playing chess. Inspired by chess tactics, our model attempts to incorporate domain knowledge to improve interpretability. We adapt patterns learned by an inductive logic programming system called PAL to derive our model. We contribute a divergence metric to evaluate our model against a random baseline, and find a set of tactics that is able to suggest moves of similar playing strength to a human beginner. Finally, we propose a computational evaluation scheme for the model by augmenting an off-the-shelf engine with it.

Figures & tables

Pattern	Definition
can_threat	A piece (P1) can threaten another piece (P2) after making a move to (X3,Y3)
can_fork	A piece (P1) can produce a fork to the opponent’s King and piece (P3) after making a move to (X4,Y4)
can_check	A piece (P1) can check the opponent’s King after a moving to (X3,Y3)
discovered_check	A check by piece (P2) can be “discovered” after moving another piece (P1) to (X4,Y4)
discovered_threat	A piece (P1) can threaten an opponent’s piece (P3) after moving another piece (P2) to (X4,Y4)
skewer	A King in check by a piece (P1) “exposes” another piece (P3) when it is moved out of check to (X4,Y4)

Table 1: Patterns learned by the PAL system that are used to create tactics

Tactic	Coverage	Divergence
Tactic	Coverage	SF14	Maia
can_threat	0.96	378.94	9.22
can_check	0.45	549.19	4.02
can_fork	0.32	676.45	4.67
discovered_check	$\approx$ 0	338.55	18.64
discovered_threat	0.96	375.97	1.19

Table 2: Coverage and DCG for each tactic

Figure 4: An example of the limitations of our 1-ply can_fork tactic. White has no immediate forking move here, leading to the tactic not matching. However, if they play 1. Nxb4 , then Black’s best response is 1. … Rxc7 which allows a fork with 2. Nd5+ leading to the capture of the rook.

Explore similar work

Chessformer: A Unified Architecture for Chess Modeling
May 18, 2026Daniel Monroe, George Eilender, Philip Chalmers +2Chess Transformer Encoder
University of Toronto · Williams College
Learning Explainable Representations of Complex Game-playing Strategies
Oct 6, 2026Abhijeet Krishnan, Colin M. Potts, Arnav Jhala +3Self-Play
Venture IV 420, North Carolina State University, Raleigh, NC 27606 USA · Department of Aerospace Engineering, Indian Institute of Technology Bombay, Powai, Mumbai 400076, Maharashtra, India · Tata Research Development and Design Centre, Pune 411013, Maharashtra, India +1
Grounded Chess Reasoning in Language Models via Master Distillation
Mar 20, 2026Zhenwei Tang, Qianfeng Wen, Seth Grief-Albert +4Chess Chain-of-Thought Reasoning
Department of Computer Science, University of Toronto · Queen’s University · Coolwei AI Lab

May 18, 2026, cs.LGJ/K move · Enter open · S save

Chessformer: A Unified Architecture for Chess Modeling

Daniel Monroe, George Eilender, Philip Chalmers, Zhenwei Tang+1

University of Toronto · Williams College

Chess Transformer Encoder Human Motion Prediction

Open paper AskPDF arXiv

Chess has long served as a canonical testbed for artificial intelligence, but modeling approaches for its central tasks have diverged. Maximizing playing strength, predicting human play, and enabling interpretability are typically solved with disparate architectures, and these designs are often misaligned with the geometry of the domain. This raises the natural question of whether these objectives require separate modeling paradigms, or if there exists a single architecture that supports them simultaneously. We introduce Chessformer, a unified architecture that advances the state of the art on all three central goals in chess modeling. Chessformer is an encoder-only transformer that represents board squares as tokens, augments self-attention with a novel dynamic positional encoding called Geometric Attention Bias (GAB) that adapts to domain-specific geometry, and predicts actions with an attention-based source-destination policy head. We evaluate Chessformer on each front. First, we develop \maiathree, a family of models for human move prediction that reaches 57.1% move-matching accuracy, significantly surpassing the previous state of the art with fewer than a quarter of the parameters. Second, we integrate Chessformer into Leela Chess Zero, a leading open-source engine, adding over 100 Elo of playing strength and resulting in tournament victories over Stockfish in major computer chess competitions. Third, we show that Chessformer's square-token design makes attention patterns and activations directly attributable to board squares, enabling granular interpretability analyses that prior architectures do not naturally support. More broadly, our results demonstrate that aligning a model's tokenization, positional encoding, and output design with the underlying structure of a domain can yield simultaneous gains in performance, human compatibility, and interpretability.

Abstract

Explore similar work

Chessformer: A Unified Architecture for Chess Modeling

Learning Explainable Representations of Complex Game-playing Strategies

Grounded Chess Reasoning in Language Models via Master Distillation