The Application of Artificial Intelligence in Chess Game

Chenyu Zhuang

International Digital Economy College, Minjiang University, Fuzhou, Fujian, China

Keywords: Artificial Intelligence, Chess Game, Search Algorithms.

Abstract: Because of its extremely high complexity, the chess game has long been a key field in the development of

artificial intelligence. With the continuous progress of science and technology, computing power is increasing

day by day, and more complex search algorithms and evaluation functions are widely used, which promotes

the continuous improvement of artificial intelligence chess power. There are many kinds of chess games, and

artificial intelligence has shown great strength in many complex chess projects, even surpassing the top human

chess players. This paper focuses on the four kinds of chess, Chinese chess， Black and White chess, Gobang

and Go to carry out in-depth discussion, systematically analyze the application of artificial intelligence in

board games, and discuss the limitations of existing artificial intelligence board games, and then forecast the

development trend of future board games. This paper aims to summarize the current scientific and

technological level of artificial intelligence games and put forward suggestions for the next development.

1 INTRODUCTION

In the course of the development of artificial

intelligence board games, in 1928, the two-person

zero-sum minimax theorem proposed by v. Neumann,

the father of computers, laid the theoretical

cornerstone of board games. In 1950, Claude Ellwood

Shannon proposed a computer scheme for chess(v.

Neumann, 1928; Li & Wu, 1995; Shannon, 1950).

The "Mac Hack VI" chess program, which appeared

in 1967, could beat chess beginners, and by the 1990s

IBM's Deep Blue computer had successfully defeated

world chess champion Garry Kasparov. In 2016,

Google's AlphaGo successfully defeated Go world

champion Lee Sedol, allowing the world to see the

strength of artificial intelligence board games. In

2017, the AlphaGO team launched AlphaGO Zero

without human training, relying only on self-

reinforcement learning, surpassing all previous

versions and pushing the artificial intelligence game

to a new height.

In the field of artificial intelligence board games,

board game provides a valuable experimental scene

for the development of artificial intelligence,

constantly promote the development of artificial

intelligence technology, but also promote the

development of chess sports, provide people with

https://orcid.org/0009-0006-0094-2884

better chess ideas, provide a variety of training tools,

and makes an important contribution to improving the

competitive level of chess players. In artificial

intelligence board games, search algorithms are used

to select strategies, such as depth-first, breadth-first,

iterative deepening search, Monte Carlo tree search,

alpha-beta pruning search, heuristic search, etc. In

terms of evaluation function, the advantages and

disadvantages of chess can be evaluated according to

different rules and characteristics of chess games. It

can also be learned through supervised learning and

using a large number of human chess pieces, or

through reinforcement learning, artificial intelligence

can play chess by itself to continuously improve its

chess ability. Focusing on the above search

algorithms and evaluation functions.

This paper gives a detailed introduction to the four

kinds of chess, Chinese chess ， black and white

chess, Gobang and Go, and artificial intelligence

game methods, as well as the future development of

artificial intelligence games, integrate more artificial

intelligence technology, and look forward to a wider

range of applications, hoping to provide greater value

to the field of artificial intelligence.

586

Zhuang, C.

The Application of Artiﬁcial Intelligence in Chess Game.

DOI: 10.5220/0013702400004670

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on Data Science and Engineering (ICDSE 2025), pages 586-590

ISBN: 978-989-758-765-8

2 METHOD

2.1 The Application of Artificial

Intelligence in the Field of Chinese

Chess

In the field of Chinese chess games, artificial

intelligence has transformed the landscape of this

board game. The chessboard is 9×10 in size, and the

moves are relatively fixed. Its complexity is lower

than that of Go. Traditional search algorithms,

heuristic search algorithms, and evaluation functions

can be utilized to find better solutions. However,

compared to Go, it has a relatively lower dependence

on deep learning. The rules of chess are relatively

clear, and it has more quantifiable evaluation

indicators. For example, different pieces such as the

chariot, the horse, the cannon, and the soldier have

different values, and these values can provide

references for situation assessment. Therefore, the

situation score can be calculated through relatively

clear evaluation functions, which makes the

calculation of chess based on traditional rules easier

to understand and operate.

Based on the analysis of the shortcomings of a

single permutation table, the higher-ups proposed a

permutation table algorithm based on hash

technology. In this algorithm, the first permutation

table adopts a depth-first strategy, while the second

one employs the always-overwriting strategy.

Through this approach, the double permutation table

can effectively improve the hit rate of the system,

avoid the problem that shallow records in the depth-

first table cannot be stored, and solve the defect of the

always-replacing table that ignores the search depth.

Experimental results show that the double

permutation table has obvious advantages in

improving the search efficiency of the Chinese chess

human-machine game system (Gao & Guo, 2008).

Yue et al. conducted research on improving the

efficiency of the α - β pruning algorithm. By

referring to the existing heuristic methods in chess,

they proposed to utilize permutation table heuristics,

static evaluation heuristics, dynamic heuristics (killer

heuristics and historical heuristics), and expand the

window for internal iterations to deepen the

generation of better move arrangement schemes. This

scheme has strong heuristic capabilities. The

experimental results show that it can significantly

improve the efficiency of the α-β pruning algorithm

(Yue & Feng, 2009).

Guo analyzed the roles of the evaluation function

and the auxiliary search mechanism in the chess game

system, proposed the application of the B* algorithm

based on the best-first search strategy, and improved

the evaluation function. This method adopts the

method of returning correction values, which is fed

back from the lower-level nodes to the upper-level

nodes, and modifies the optimistic and pessimistic

values of the upper-level nodes, so as to continuously

shorten the value range and reach the termination

condition of the algorithm, making it easier to find the

best branch. The experimental results show that the

B* algorithm is effective in the chess game and

improves the efficiency of search (Guo, 2010).

2.2 The Application of Artificial

Intelligence in the Field of Black

and White Chess.

In the game of Black and White chess, artificial

intelligence provides more powerful learning and

training tools, enriching the tactical strategies of

black and white chess, and having no small impact on

the development of black and white chess. The Black

and white chess board is 8×8, a total of 64 grids, and

its complexity is far less than Go. When artificial

intelligence is looking for the best solution, more

search algorithms are used to build game trees for

search and evaluation, and less deep learning and

reinforcement learning are used. When evaluating the

situation, Black and White chess mainly calculates

the situation according to the number of pieces,

occupying positions and the number of flipped pieces,

etc., based on relatively clear rules and through clear

evaluation functions.

By combining the estimation process of the

generated game tree nodes with the search process of

the game tree, Du et al. proposed to use Alpha-Beta

pruning and max-min principle methods to search for

the best position, optimize the game tree search and

valuation function from two aspects and change the

search order will improve the efficiency of the

pruning algorithm. When the algorithm gives the

chosen step, do not stop the search immediately, but

search a few steps further on the original estimated

possible path, and check again whether there will be

accidents, and add auxiliary search methods. The

experimental results show the effectiveness of the

algorithm, which improves the original algorithm and

improves the search speed (Du & Cheng, 2007).

Li studied the game tree search algorithm and

proposed a heuristic improvement on the basis of the

game tree search algorithm. Heuristic factors include:

the double substitution table heuristic, shallow

detection heuristic and "killer" heuristic. In node

sorting, the method of dynamic selection of node

The Application of Artiﬁcial Intelligence in Chess Game

587

sorting is used, including controlling a certain sort

depth, pv node sorting, child node sorting of the

"eldest" node and so on. After the above

improvement, the experimental results show that the

improved heuristic game search is better than the

general heuristic search. At the same time, the 0

window heuristic search strategy is used to further

improve the search efficiency (Li, 2010).

Peng studied the application of artificial

intelligence in Black and white chess, introduced the

classic black and white chess algorithm, studied the

Monte Carlo tree search algorithm, Q learning

algorithm and SARSA algorithm, and applied them in

black and white chess. The experiment shows the

effectiveness of the Monte Carlo tree search

algorithm, Q learning algorithm and SARSA

algorithm, which improves the search efficiency

(Peng, 2021).

2.3 The Application of Artificial

Intelligence in the Field of Gobang

In the field of Gobang (Five-in-a-row) games,

artificial intelligence has a significant impact on its

development. The Gobang board is typically 15×15

or 19×19, with fewer moves compared to Go and a

lower level of complexity. In Gobang games based on

artificial intelligence, decisions are often made by

directly matching known winnings or draw patterns.

For more complex situations, artificial intelligence

uses search algorithms to construct game trees for

search and evaluation. Gobang relies far less on deep

learning and reinforcement learning than Go. The

evaluation of Gobang's positions is relatively simple,

and the superiority or inferiority of the game can be

calculated through quantitative analysis of the board

patterns.

Wang conducted in-depth research on the α-β

pruning algorithm and proposed an improved game

tree search algorithm that combines the five-in-a-

row/double-three function, evaluation function, and

α-β pruning algorithm. This algorithm first applies

the five-in-a-row/double-three function and the

evaluation function for judgment. If no suitable chess

point is found, it then uses the α - β pruning

algorithm for search. The experimental results

achieved a human-machine intelligent Go game and

verified the effectiveness of the game tree search,

improving the algorithm efficiency (Wang, 2011).

Cheng et al. improved the Alpha-Beta search

algorithm by studying it and proposed an enhanced

version that incorporates local search, static

evaluation heuristics, and iterative deepening to boost

the efficiency of the Alpha-Beta search algorithm.

This experiment tailored the search algorithm to the

characteristics of Gobang, achieving the goal of

enhancing search efficiency and significantly

improving it (Cheng & Lei, 2012).

Shen proposed a Gobang algorithm based on

Monte Carlo Tree Search and deep neural networks

in response to the study of the branching factors and

complexity of Gobang game situations. The

algorithm simulates Gobang games using Monte

Carlo Tree Search and is fully trained through self-

play reinforcement learning. At the same time, a

Gobang policy evaluation deep neural network

composed of multiple residual blocks is used to

evaluate the move positions and sample movements

in Gobang games. The residual structure forms cross-

layer connections, making the network more stable

and easier to train. Using the policy evaluation deep

neural network with a residual structure to guide the

Monte Carlo Tree Search enhances the intensity of

the tree search, resulting in higher quality moves and

stronger self-play iterations. Experimental results

demonstrated the feasibility of this method and

increased the speed of training the Gobang algorithm

model agent (Shen, 2021).

2.4 The Application of Artificial

Intelligence in the Field of Go.

The introduction of artificial intelligence in the game

of Go has completely transformed the development

landscape of the world's Go. The AI's Go-playing

ability has surpassed that of all Go players around the

world, making it the leader in this field. The Go board

is 19×19, with 361 intersections. Theoretically, there

are 361 possible moves for the first step, and the

complexity of the combinations is extremely high,

making it one of the most complex board games in the

world. Therefore, Go games based on artificial

intelligence, rely more on deep learning and

reinforcement learning for calculation. The

evaluation of the Go board situation is difficult to

precisely quantify with simple rules, and the value of

the pieces is not fixed. AI uses deep learning models

to learn how to evaluate the board situation. This

board situation evaluation is based on learning from a

vast number of games and feature extraction, rather

than using simple rule evaluation and blind

calculation.

Through studying the design and implementation

of AlphaGo's value neural network, Wang created a

Go dataset, reproduced the value neural network

model, optimized the algorithm for the network

training process, and implemented a 36-layer deep

ICDSE 2025 - The International Conference on Data Science and Engineering

588

residual value neural network. Finally, he conducted

large-scale distributed neural network training and

trained a more accurate value neural network. This

experiment designed a more precise value neural

network that can accurately predict the winning rates

of black and white pieces. Compared with AlphaGo's

value neural network, improvements were made in

dataset creation, network model construction, and

large-scale distributed training, enhancing the

valuation accuracy (Wang, 2018).

With the aim of reducing reliance on computing

power and enhancing algorithm performance, this

study focuses on improving the feedback mechanism

and neural network of deep reinforcement learning

algorithms in computer board game applications. A

hybrid deep reinforcement learning model is

proposed, featuring a novel neural network structure

called the "max-average output layer," which replaces

several convolutional layers in the middle of the

CNN. This method is based on deep reinforcement

learning for board games and employs a Q-value

update approach that combines Q-Learning and

Sarsa(λ) in the Q-table. Using Microsoft's .Net

Framework 4.7.2 and the Microsoft Cognitive Toolkit

deep learning library, a Go game-playing program

based on deep reinforcement learning was designed

and implemented. The performance of the network

structure with the "max-average output layer" was

verified, and the algorithm performance was

improved (Lv, 2020).

The AlphaGo series was trained through human

expert supervision learning and self-play

reinforcement learning. The AlphaGo team proposed

an algorithm based solely on reinforcement learning,

without human data guidance, and designed AlphaGo

Zero. This version became its own teacher and

reduced the utilization of tree search. In this

experiment, AlphaGo Zero defeated the previous

most powerful AlphaGo version by a large margin.

The latter was trained from human data through

manual education. However, the former was trained

through self-play reinforcement learning, using 40

residual block neural networks for 40 days, achieving

better performance (Silver et al., 2017).

3 CURRENT LIMITATIONS AND

FUTURE DEVELOPMENT

PROSPECTS

The development of artificial intelligence in board

games has been rapid. Although AI has defeated top

human players in complex games like Go, the

progress in some other board games has been

relatively slow. Moreover, AI's performance may be

affected when facing opponents with different styles

and more complex situations. Therefore, the

adaptability of AI still needs to be improved. When

dealing with complex board game problems, AI

requires a large amount of computing resources and

time for search and decision-making. Even with some

optimization algorithms, it is still difficult to quickly

provide the optimal solution within limited

computing resources, which restricts its application in

scenarios with high real-time requirements. In the

long term, enhancing the generalization ability of AI

is crucial for its application in board games. AI with

strong generalization ability can better adapt to

different playing styles and various complex

situations, rather than being limited to specific

scenarios or patterns. The improvement of the

explainability of AI's decision-making process should

not be ignored either. Explainability enables humans

to understand the decision-making logic of AI more

clearly, which is helpful for further optimizing its

strategies and enhancing the trust and collaboration

between humans and AI in board game interactions.

In the future, it is hoped that multiple AI technologies

can be integrated to increase computing power,

provide the optimal solution more quickly within

limited computing resources, adapt to more complex

situations, and at the same time, strengthen the

generalization ability of AI to handle more complex

scenarios and enhance the explainability of its

decision-making process, so as to promote greater

development and open up broader space for AI in the

field of board games.

4 CONCLUSIONS

This article analyzes the artificial intelligence (AI)

game-playing methods in four types of board games

(Chinese chess, Black and White chess, Gobang, and

Go), delving into the application and development of

AI in board game competitions. Chinese chess and

Gobang to some extent rely on traditional search

algorithms and evaluation functions, through which

effective situation assessment and decision-making

can be achieved. However, due to the highly complex

board and vast computational space of Go, it more

heavily depends on advanced technologies such as

deep learning and reinforcement learning to handle

complex game situations. Through detailed analysis

of these four types of board games, not only

demonstrates how AI has broken through technical

bottlenecks and continuously improved the

The Application of Artiﬁcial Intelligence in Chess Game

589

intelligence level of board game competitions but also

reveals the different challenges and solutions in the

application of AI in different board games. With the

continuous integration and development of multiple

technologies, AI will also play an increasingly

important role in more complex and diverse fields,

promoting innovation and development in various

industries. The application of AI in board game

competitions not only promotes the research of

intelligent systems but also provides valuable

technical accumulation for exploring complex

systems, indicating that AI technology will show a

broader and more profound application prospect in

the future.

REFERENCES

Cheng, Y., & Lei, X. (2012). Research and improvement of

Alpha-Beta search algorithm in Gomoku. Computer

Engineering, 38(17), 186-188.

Du, X., & Cheng, J. (2007). The application of game

algorithms in the game of Othello. Computer

Technology and Development, 17(1), 216-218.

Gao, Q., & Guo, C. (2008). Research on the application of

hash technology in Chinese chess computer game

systems. Science Technology and Engineering, 8(17),

4869-4872.

Guo, X. (2010). Research and improvement of search

algorithms in Chinese chess computer games (Master's

thesis, Hebei University).

Li, X. (2010). Design and implementation of an improved

game tree for Othello (Master's thesis, South China

University of Technology).

Li, X., & Wu, W. (1995). Biographies of world-renowned

mathematicians. Science Press.

Lv, Z. (2020). Research on chess game strategies based on

deep reinforcement learning (Doctoral dissertation,

Central University for Nationalities, Beijing).

Peng, Z. (2021). Application of computer game algorithms

in Othello. Modern Information Technology.

Shannon, C. E. (1950). XXII. Programming a computer for

playing chess. The London, Edinburgh, and Dublin

Philosophical Magazine and Journal of Science,

41(314), 256-275.

Shen, X. (2021). Design and implementation of a Gomoku

algorithm based on Monte Carlo Tree and neural

networks (Master's thesis, Shenyang University of

Chemical Technology).

Silver, D., Schrittwieser, J., Simonyan, K., Antonoglou, I.,

Huang, A., Guez, A., ... & Hassabis, D. (2017).

Mastering the game of Go without human knowledge.

Nature, 550(7676), 354-359.

v. Neumann, J. (1928). Zur Theorie der Gesellschaftsspiele.

Mathematische Annalen, 100(1), 295-320.

Wang, J. (2011). Research on the application of game tree

heuristic search algorithm in the game of Gomoku.

Science and Technology Information Development and

Economy, 21(29), 138-140.

Wang, L. (2018). Optimization research of value neural

networks in computer Go (Master's thesis, Beijing

University of Posts and Telecommunications).

Yue, J., & Feng, S. (2009). The application of game tree

search algorithm in Chinese chess. Computer Systems

Applications, (9).

ICDSE 2025 - The International Conference on Data Science and Engineering

590