Computer Chess Explained: Its History, Chess Engines and Tablebases

Computer chess covers the machines and programs that play the game and the data they rely on. It runs from hand-built programs of the 1950s through the plastic boards sold in the 1970s to the neural-network engines that anyone can run on a phone. Computer chess history, chess engines and tablebases are the three threads of the field, and this page sets out the main periods and points to the guides on individual machines, programs and file formats.

Computer chess history in brief

The theory came first. Claude Shannon described the principles of a chess-playing program in 1950 and separated brute-force search, which he called Type A, from selective search, Type B. For years selective programs dominated, and the best of them in 1967 played at about the level of an average amateur. The turn came with Northwestern’s Chess 4.0 in 1973, which searched every legal move to a fixed depth. The first World Computer Chess Championship took place in 1974, and in 1977 Chess 4.5 became the first program to win a human tournament.

A famous wager sits in the middle of this computer chess history. In 1968 David Levy bet that no machine would beat him within ten years, and in 1978 he won it against Chess 4.7. The man and the bet are covered in the article on David Levy as player and computer chess pioneer. The machines caught up quickly afterwards. Deep Thought reached a USCF rating of 2551 in 1988, Garry Kasparov lost a single game to Deep Blue in 1996 and still won the match, and the upgraded Deep Blue won the 1997 rematch 3½–2½. Rebel 10 beat Viswanathan Anand 5–3 in 1998, and Deep Fritz beat Vladimir Kramnik 4–2 in 2006.

Dedicated machines, handhelds and the chess computer shop

The first dedicated chess computer was the Chess Challenger 1, designed by Sidney Samole and Ron Nelson and sold by Fidelity Electronics from early 1977. Dedicated machines are built for chess alone, with a few buttons or a sensory board instead of a full screen. They were strongest in the 1980s and early 1990s, and the Mephisto Montreux was the last of the breed at the 1995 world microcomputer championship.

Two guides look at those machines in detail. Fidelity Chess Challenger Models, Manual and History separates the models by year and processor and explains where to find a manual. Best Portable Handheld Chess Computer for Travel covers pocket units from 1980 to 1993 and what to weigh when choosing one. Both are explainers drawn from published collector and museum records, not tested rankings.

Programs, opening books and chess engines

On a PC the same ideas appear as software. Several articles cover the best-known families. Chessmaster 10th Edition and Josh Waitzkin’s Lessons looks at one release of the series, and Chessmaster Sound and Voice follows how the series used speech and audio. On the ChessBase side, ChessBase: History, Products and US Distribution describes the company’s products, and ChessBase Reader and Fritz Trainer Explained explains what the free reader opens.

A program does not search from move one in every game. An opening book is a store of opening positions that it looks up at the start, and once the opponent leaves the book the program returns to normal search. Binary books are keyed by position hash values, and the PolyGlot format uses 16 bytes per item for the hash, move, weight and learning data. The common formats are ABK from Arena, BIN from PolyGlot and CTG from ChessBase, and the article on the CTG file shows how to open and convert one. A smaller technical piece, Unicode Chess Symbols, lists the code points used to print the pieces in text.

Modern chess engines changed again in 2017, when AlphaZero used a Monte Carlo tree search variant, and Leela Chess Zero adopted the same approach in 2018. Stockfish now uses efficiently updatable neural networks, while Leela relies on graphics processors.

Chess engines and tablebases in the endgame

The strongest engines do not calculate every ending. An endgame tablebase is a database of precalculated results for positions with few pieces, and it records whether each one is won, drawn or lost with best play and how many moves that takes. Thomas Ströhlein studied the first endings in a 1970 thesis, and Ken Thompson’s king, queen and king against king and rook table was used in a 1977 match against Walter Browne.

Tablebases grew with storage. The Nalimov set for five pieces or fewer needs 7.05 GB, while its six-piece set needs 1.2 TB. Ronald de Man released the Syzygy tablebases in April 2013, and the six-piece set takes 160.2 GB. The seven-piece Syzygy set takes 18.4 TB. A seven-piece table first finished in 2012 on the Lomonosov supercomputers at Moscow State University, at about 140 TB.

Two measures matter. Depth to mate counts moves to checkmate, and depth to zeroing counts moves to a capture or pawn move that resets the fifty-move counter. Many tablebases ignore the fifty-move rule, so they can mark a position as won that the rule would draw. In correspondence play the ICCF accepted six-man tablebases for such claims in 2014 and seven-man ones in 2020, and the page on correspondence chess covers that setting.

All computer chess articles on this site

Questions