Can AI Solve Chess? The Truth About the Game’s Complexity

Chess has been called the “game of kings,” a battlefield of logic, psychology, and strategy that has fascinated humanity for centuries. With the explosive rise of artificial intelligence (AI), one pressing question remains: Can AI solve chess? And if so, what does it really mean to “solve” a game as rich and intricate as chess?

In this article, we’ll unpack what it means to “solve” a game, examine chess’s staggering complexity, explore how far AI has come in mastering it, and evaluate whether a true solution is theoretically or practically within reach. Spoiler: the answer is both awe-inspiring and humbling.

What Does It Mean to “Solve” a Game?

Before diving into chess, let’s define what “solving” a game actually entails in computer science and game theory.

Types of Solutions

1. Ultra-weakly Solved
   A game is solved if we can determine whether the starting position is a win, loss, or draw with perfect play from both sides.

2. Weakly Solved
   A game is weakly solved if we know the correct move to make from every position that could realistically occur in gameplay.

3. Strongly Solved
   A game is strongly solved when we know the best move from every possible legal position, no matter how obscure or unlikely.

For example, Tic-Tac-Toe is strongly solved: it’s a draw with best play from both sides. Checkers is weakly solved: the outcome is a draw with perfect play, and a database of optimal moves exists for every position. Chess, however, remains unsolved at any level.

The Complexity of Chess

To understand why, we need to grasp how complex chess truly is.

The Numbers

• Average branching factor (number of legal moves per position): ~35

• Average game length: 40 moves per side (80 plies)

• Estimated total number of legal positions: ~10^43

• Estimated number of possible chess games: >10^120 (Shannon number)

For context, there are only ~10^80 atoms in the observable universe. This puts the total number of possible chess games far beyond what any current or future classical computer could process exhaustively.

Game Tree Complexity

A game tree represents all possible sequences of moves from a starting position. For chess, the game tree complexity exceeds 10^120. Evaluating all nodes in this tree—even with massive parallel computing and distributed systems—would take longer than the lifetime of the universe.

In other words, chess is computationally intractable to fully solve with brute-force methods.

How Close Are We? What AI Has Achieved

Even though chess hasn’t been “solved,” AI has made staggering progress in playing and understanding the game.

1. Stockfish (Traditional Engine + NNUE)

Stockfish is a deterministic, open-source engine using alpha-beta search and a neural network evaluator (NNUE). It calculates millions of positions per second and uses heuristics, endgame tablebases, and deep search to reach near-perfect play.

Key achievements:

• Dominated nearly all major computer chess championships.

• Integrates tablebases that do solve all endgames up to 7 pieces.

• Evaluates positions with astonishing accuracy and speed.

2. AlphaZero (DeepMind)

In 2017, DeepMind’s AlphaZero introduced a radical approach. Instead of being programmed with chess knowledge, it learned the game through self-play using a deep neural network and Monte Carlo Tree Search (MCTS).

Key breakthroughs:

• Defeated Stockfish 8 decisively after only a few hours of training.

• Played in a highly creative, human-like, and strategic style.

• Learned principles like positional sacrifice and pawn structure organically.

Yet, even AlphaZero cannot “solve” chess. It plays exceptionally well but cannot guarantee perfect play.

3. Leela Chess Zero (Lc0)

Inspired by AlphaZero, the open-source project Leela Chess Zero uses a similar architecture and trains itself through millions of self-play games.

What it brings:

• Continually improving through crowd-sourced computation.

• Offers strategic depth often superior to traditional engines.

• Complements brute-force engines with positional intuition.

Still, Leela is approximating optimal play—not solving the game.

What About Endgames?

In chess, some endgame positions have been completely solved.

Tablebases

Tablebases are exhaustive databases that contain perfect information about specific endgame positions (e.g., king + queen vs. king + rook).

• 5-piece tablebases were completed decades ago.

• 6-piece in the early 2000s.

• 7-piece tablebases completed in the 2010s (about 140 TB of data).

In these positions, the best move and guaranteed outcome (win, draw, or loss) are known from every legal configuration. These are strongly solved subdomains of chess.

Why not 8 pieces? The storage and computational requirements explode exponentially—currently beyond practical reach.

Could Chess Ever Be Solved?

Theoretical Possibility

In theory, yes. Chess has a finite number of legal positions and finite game length due to the 50-move rule and threefold repetition rule. Therefore, it is a finite deterministic game of perfect information. It is mathematically solvable.

However, the solution requires:

• Calculating optimal moves for ~10^43 positions.

• Storing the solution compactly (which seems currently impossible).

• Access to computing power that’s far beyond what we possess today.

Quantum and Future Computing

Some argue that future quantum computers could, in principle, perform computations fast enough to “solve” chess or huge portions of it. But as of 2025, this remains speculative. Even then, the complexity of evaluating 10^43 positions is not trivialized by quantum speed-ups alone.

Implications of a Solved Chess

If chess were solved, would it ruin the game?

Not necessarily. Consider these outcomes:

• If the solution revealed that chess is a draw with perfect play (as many suspect), the game would still be rich for human competition.

• Knowledge of the solution would likely remain inaccessible in practical gameplay—only a handful of openings might be mapped fully.

• Chess would still remain fascinating due to its psychological and practical complexity. Humans are not perfect, and practical decisions under time pressure will always matter.

Additionally, even if “solved,” players could switch to more complex variants like Chess960 (Fischer Random) to restore uncertainty.

What Chess AI Tells Us About Human Intelligence

The rise of AI in chess has taught us valuable lessons:

• AI doesn’t think like humans—it finds novel paths to victory.

• AI can enhance human play, offering analysis, inspiration, and training.

• The distinction between calculation and intuition in AI is blurring.

While AI has not “solved” chess, it has mastered it to a level far beyond human capability. World champions like Magnus Carlsen rely on engines for preparation, but during play, they use their human instincts—showing that perfect knowledge isn’t everything.

Conclusion: The Truth About Chess and AI

So, can AI solve chess?

• In theory: Yes, it’s a finite game with a definitive solution.

• In practice: No—not yet, and possibly not for centuries, due to astronomical complexity.

• What AI can do: Play at superhuman levels, train players, analyze mistakes, and advance our understanding of the game.

Rather than rendering chess obsolete, AI has breathed new life into it. Today’s players can access tools and insights unthinkable a generation ago. Chess remains unsolved—but more fascinating than ever.

Whether or not we eventually solve the game, it seems clear that chess will never be “over”—only deeper.