How Do Chess Engines Work? (Explained)

Chess engines, the digital brains behind virtual chess games, operate on a foundation of meticulously crafted algorithms.

These algorithms, such as the widely-recognized Minimax, look into the vast tree of possible moves, evaluating each potential future state of the game to determine the optimal move.

The Minimax algorithm, in particular, navigates through all possible future moves, maximizing the player’s advantage while minimizing the opponent’s, hence ensuring a strategic play.

Evaluating Positions: The Essence of Decision Making

A pivotal component in the functionality of chess engines is their ability to evaluate board positions.

The evaluation function assigns a numerical value to each possible position, reflecting its desirability based on various factors like material balance, king safety, and piece activity.

For instance, a position where a player has a material advantage (more or higher-value pieces) might be assigned a higher value than a position with material parity.

The engine then uses these values to determine which move will lead to the most advantageous position.

Depth and Breadth: Exploring Future Possibilities

Chess engines meticulously explore the tree of possible moves to a specified depth, evaluating numerous potential future positions.

The depth of search, often measured in “plies” (one move by one player), is crucial for determining the engine’s strength and playing style.

A deeper search allows the engine to explore more potential outcomes and foresee distant threats and opportunities, albeit at the cost of increased computational resources and time.

Pruning (Alpha-Beta Pruning): The Art of Strategic Elimination

To manage the astronomical number of possible moves in chess, engines employ a technique known as pruning.

Alpha-Beta pruning, a notable method, strategically eliminates branches of the search tree that are unlikely to influence the final decision, thereby optimizing computational efforts.

By setting thresholds (alpha and beta values) and disregarding moves that do not alter these thresholds, the engine efficiently narrows down its search to the most promising options without exploring every possible move.

Endgame Tablebases: Navigating the Final Stages

In the endgame, where the number of pieces is significantly reduced, chess engines often refer to endgame tablebases – precomputed databases that contain exact evaluations and optimal moves for every possible position with a limited number of pieces.

By consulting these tablebases, the engine can play the endgame perfectly, ensuring that it capitalizes on any advantage and avoids pitfalls.

Machine Learning: The Evolutionary Leap

In recent years, machine learning has introduced a paradigm shift in how chess engines operate.

Engines like AlphaZero employ neural networks, training themselves by playing millions of games and learning from the outcomes.

Unlike traditional engines, which rely heavily on predefined evaluation functions and move-generation algorithms, machine learning-based engines create their own understanding of positional values and strategy, often resulting in a more intuitive, human-like style of play.

The Human Element: Tuning and Testing

Despite their computational prowess, chess engines are not devoid of the human touch.

Developers, often expert chess players themselves, meticulously tune evaluation functions and optimize search algorithms to enhance the engine’s performance.

Furthermore, engines are rigorously tested against each other and human players, ensuring their efficacy and reliability in real-world scenarios.

Stockfish vs. AlphaZero

Stockfish and AlphaZero are two of the most powerful chess engines in the world, but they approach the game of chess in fundamentally different ways.

Let’s break down their inner workings:


  1. Traditional Engine: Stockfish is a traditional chess engine, which means it uses a combination of brute force search and heuristics to evaluate positions.
  2. Minimax Algorithm: At its core, Stockfish uses the minimax algorithm to traverse the game tree. This involves looking at all possible moves, then all possible replies to those moves, and so on, to a certain depth.
  3. Alpha-Beta Pruning: To speed up the search, Stockfish uses alpha-beta pruning. This technique avoids analyzing moves that are provably worse than previously examined moves.
  4. Evaluation Function: Once the search reaches a certain depth, Stockfish uses an evaluation function to assign a numerical value to the position. This function considers various factors like material balance, pawn structure, king safety, and more.
  5. Opening Books and Endgame Tablebases: Stockfish uses precomputed databases for the opening phase of the game (opening books) and for endgame positions with a small number of pieces (endgame tablebases).
  6. Iterative Deepening: Stockfish uses iterative deepening, which means it first searches to a shallow depth and gradually increases the depth. This allows it to respond quickly if needed, but also to search deeply if it has more time.


  1. Neural Network-Based: AlphaZero uses a deep neural network to evaluate positions and to guide its search.
  2. Self-Play Learning: Unlike Stockfish, which is hand-tuned by humans, AlphaZero taught itself chess. It started with random play and improved over time by playing millions of games against itself, adjusting its neural network weights using a technique called reinforcement learning.
  3. Monte Carlo Tree Search (MCTS): Instead of the traditional minimax search, AlphaZero uses MCTS. This involves simulating many possible games (called rollouts) from the current position and using the results of those simulations to guide the search.
  4. Position Evaluation: Instead of a handcrafted evaluation function, AlphaZero uses its neural network to evaluate positions. The network outputs both a position evaluation and a probability distribution over moves.
  5. No Opening Books or Endgame Tablebases: AlphaZero doesn’t use any precomputed databases. All its knowledge comes from its neural network, which was trained through self-play.
  6. General Architecture: The same architecture and training methods used for AlphaZero’s chess-playing capabilities were also used to teach it to play other games like Go and Shogi, demonstrating its general applicability.

In short, while Stockfish uses a combination of brute force search and handcrafted heuristics, AlphaZero relies on deep learning and self-play to understand and play chess.

Both approaches have their strengths, and in head-to-head matches, they have shown to be very competitive with each other.

Related: Stockfish vs. Leela

FAQs – How Do Chess Engines Work?

What is a chess engine?

A chess engine is a computer program that analyzes chess positions and makes decisions on the best moves.

These engines use advanced algorithms and computational techniques to simulate millions of positions in a short amount of time, allowing them to play at a level far beyond that of any human.

How do chess engines evaluate positions?

Chess engines evaluate positions using a combination of factors.

These include material balance (the value of pieces on the board), positional factors (like pawn structure, control of key squares, and king safety), and dynamic factors (like threats and attacking potential).

The engine assigns a numerical value to each position, with positive values favoring white and negative values favoring black.

What algorithms do chess engines use to search for the best move?

The primary algorithm used by traditional chess engines is the minimax algorithm with alpha-beta pruning.

This algorithm traverses the game tree by exploring all possible moves, then all possible replies to those moves, and so on, to a certain depth.

Alpha-beta pruning is a technique that skips analyzing moves that are provably worse than previously examined moves, making the search more efficient.

How do chess engines handle the opening phase of the game?

Many chess engines use opening books, which are databases of well-studied opening sequences.

These sequences are derived from historical games and opening theory.

When a position from the opening book is reached, the engine selects a move from the book rather than calculating one from scratch.

This ensures that the engine plays the opening phase quickly and in accordance with established theory.

How do chess engines handle endgame positions?

For endgames, many engines use endgame tablebases.

These are precomputed databases that contain perfect play for positions with a small number of pieces.

When the engine recognizes a position from the tablebase, it can instantly retrieve the best move and the expected result (win, loss, or draw) without further calculation.

What is the difference between traditional and neural network-based chess engines?

Traditional chess engines, like Stockfish, use handcrafted evaluation functions and brute-force search algorithms.

Neural network-based engines, like AlphaZero, use deep learning techniques.

Instead of a handcrafted evaluation, they use a neural network trained through self-play to evaluate positions and guide their search.

This self-learning approach allows them to discover strategies and patterns without human intervention.

How do chess engines use databases or tablebases?

Databases, often referred to as opening books, provide engines with a repertoire of opening moves based on historical games and opening theory.

Tablebases, on the other hand, are used in the endgame phase.

They contain precomputed evaluations for all possible positions with a limited number of pieces, allowing the engine to play these positions perfectly.

What is the role of heuristics in a chess engine’s evaluation?

Heuristics are rules or guidelines that the engine uses to quickly evaluate positions without calculating every possible continuation.

They might include rules about pawn structure, king safety, piece activity, and more.

While they are not always perfect, they allow the engine to make reasonably accurate evaluations quickly.

How do chess engines prioritize which moves to analyze?

Engines use a technique called move ordering to prioritize the most promising moves.

By analyzing likely strong moves first, they can use alpha-beta pruning more effectively, cutting off less promising lines of play earlier in the search.

Common heuristics for move ordering include capturing moves, checks, and moves that bring a piece to a more active position.

How do engines like AlphaZero learn from self-play?

AlphaZero uses a method called reinforcement learning.

It starts with random play and gradually improves by playing millions of games against itself.

After each game, it adjusts the weights of its neural network to favor strategies that led to winning and to penalize strategies that led to losing.

Over time, this iterative process allows AlphaZero to discover strong strategies without any human input.

What is Monte Carlo Tree Search (MCTS) and how is it used in chess engines?

MCTS is a search algorithm that simulates many possible games (called rollouts) from the current position.

Instead of analyzing every possible move to a fixed depth, MCTS randomly explores possible continuations, using the results of these simulations to guide its search.

Over time, the algorithm focuses more on the most promising lines of play.

AlphaZero uses a variant of MCTS, guided by its neural network, to select moves.

How do chess engines handle time management during a game?

Chess engines use algorithms to decide how long to think about each move, based on the total time available and the complexity of the position.

They might spend more time on critical or complex positions and less time on simpler ones.

They also ensure they have enough time left to handle the rest of the game without running out of time.

What are the computational requirements for running a top-tier chess engine?

Top-tier chess engines, especially when competing in tournaments, often run on powerful hardware with multiple CPU cores or even specialized hardware like TPUs for neural network-based engines.

The exact requirements vary, but in general, more computational power allows the engine to analyze more positions in less time, leading to stronger play.

How have chess engines impacted professional chess play?

Chess engines have had a profound impact on professional chess.

Players use them for preparation, analysis, and training.

Engines have expanded our understanding of certain positions and have even led to revisions and improvements in opening theory.

However, they’ve also raised concerns about cheating in online games, as it’s possible for players to consult engines during play.

Can any chess engine play a perfect game of chess?

No chess engine, or even a combination of all existing engines, can play a perfect game of chess due to the immense complexity of the game.

However, top-tier engines play at a level far beyond the best humans and make very few mistakes.

In practical terms, their play is close to perfection, especially in well-understood positions.

The best chess engines play at a 3600+ level, which is well beyond the capabilities of all humans.

Why is Alphazero no longer in development?

AlphaZero, developed by DeepMind, was a research project aimed at demonstrating the capabilities of general reinforcement learning algorithms.

After achieving state-of-the-art performance in chess, Go, and shogi without domain-specific knowledge, the primary research goals were accomplished.

DeepMind then shifted its focus to other challenges in artificial intelligence, rather than further refining a specialized game-playing engine.


Chess engines amalgamate algorithmic precision and, increasingly, machine learning to navigate the complex landscape of the chessboard.

Through strategic evaluation, exhaustive search, and intelligent pruning, they simulate a depth of understanding and strategic foresight that rivals, and often surpasses, human players.

As technology evolves, the synergy between traditional algorithms and machine learning is poised to further elevate the capabilities of chess engines, crafting a future where they continue to be invaluable tools and formidable opponents in the enthralling realm of chess.


Related Posts

Leave a Reply

Your email address will not be published. Required fields are marked *