Blogs
Blogs
Evolution of Chess AI: from Turochamp to idChess

Evolution of Chess AI: from Turochamp to idChess

idChess_Club
| 0

In the late 18th century Wolfgang von Kempelen constructed Mechanical Turk, a chess-playing machine, dazzling the court of Empress Maria Theresa of Austria. The automaton was operated by human players, concealed within, but it made people dream of a mechanism challenging them to the game they created.

This article will show how chess technologies evolved from basic chess algorithms confined to singular machines to chess programs that allow to digitize physical games and broadcast those far and wide.

First Chess Algorithms: Turochamp, MANIAC I
Turochamp was created in 1948 by two Englishmen, Alan Turing and David Champernowne. The idea was conceived as they discussed a possibility of a computer performing tasks usually performed by humans.

The algorithm of 1948 proved to be too complex for the machines of the time. One game was recorded, with Turing himself executing the code, playing the part of a machine. However, it lost to his colleague. Adapting the code for Ferranti Mark I failed, yet Turing believed the required technology would come soon.

Sadly, the man wouldn’t live to see that, but he witnessed the machine that would make the revolution – MANIAC-I, built by Nicholas Metropolis. In 1956, Paul Stein and Mark Wells designed simplified chess MANIAC-I could process.

MANIAC-I of Los Alamos National Laboratory

The Los Alamos Chess was played on a 6x6 board, with no bishops, no pawn double-step move, no en passant capture and no castling. Three games of the kind were held, with MANIAC-I outplaying a lab assistant in the final round. It was the first time a computer won against a human in a chess-like game.

Achievements of Automations: Mac Hack, Belle
The next breakthrough came in 1966, as Richard Greenblatt wrote Mac Hack VI. Designed for PDP-6, it was the first chess program to use a transposition table. Superior to its contemporaries, it ruled the late 60s.

The 70s, however, were dominated by Belle. It had dedicated hardware created by Joe Condon, with software designed by Ken Thompson. Belle was a lightning brute-force machine, sorting through all possible moves at highest speeds.

Five times winner of the ACM North American Computer Chess Championship and the winner of the Third World Computer Chess Championship, Belle was unparalleled for its time. Its success became the basis for ChipTest which, in turn, was the precursor to the legendary Deep Blue.

Deep Blue VS Garry Kasparov
In 1980 the three-tiered Fredkin Prize was established to push the limitations of chess engines. Belle received the first-tier prize, being the first machine to reach master-level. Its designers received $5000.

Five Carnegie Mellon graduate students designed Deep Thought in 1985. It was the first computer to beat a grandmaster in a regular tournament game. The team was awarded the second-tier prize of $10,000 and got the attention of IBM, who were eager to perfect the system.

The opening book of IBM Deep Blue, the perfected computer, incorporated more than 4,000 positions and 700,000 grandmaster games. The original machine of 1995 could search 200 million positions per second.

Garry Kasparov Vs. Deep Blue, 1996

In 1996, the reigning World Champion Garry Kasparov managed to win against Deep Blue by the skin of his teeth. However, in the 1997 rematch the grandmaster lost to its upgraded version. This triumph marked a milestone for artificial intelligence as a whole.

That year the Deep Blue team received the third-tier Fredkin Prize of $100,000.

Chess AI of the 21st Century: Stockfish, Leela Chess Zero
Since Deep Blue defeated the reigning World Champion, direct confrontation of man against machine had become obsolete. Instead, developers would challenge their dedicated chess creations against others’.

The conventional approach to chess engine design is now represented by the Stockfish engine that dominates various championships. In 2020 it incorporated a new technology that has proven its worth: neural network.

Neural networks allow an engine to learn chess by figuring out the rules of the game. The first program of the sort, AlphaZero, was trained by playing against itself. It could defeat Stockfish without listing through 60 million positions, only needing 60 thousand.

The greatest challenge to Stockfish was Leela Chess Zero, first released in 2018. It had been consistently challenging Stockfish; in 2019 and 2020 it managed to win in the Top Chess Engine Championship.

Human and Machine, Training Together
Time shows that chess engines can be used for various purposes. One of the major ways they’re used is assisting human players as they perfect their chess skills.

One such example is the way Stockfish is incorporated into Chess.com and the Lichess service to assist the analysis of games its users have played, highlighting their moves in chess notation. Not only that, but Stockfish also provides bots to train against.

Technologies such as the Square Off automated chess boards allow one to play against a computer on a physical board, adding to one’s immersion and preparing to challenge a human player.

Digitizing and Broadcasting Chess Games
Modern technologies assist our chess games not only online but also in real chess boards. As such, demonstration boards have become obsolete.

For example we know DGT-boards that record the moves taken as a chess diagram. They are used by chess federations of the highest caliber, but are generally seen as too expensive for most clubs and chess schools, leaving them with traditional hand-written notations.

The idChess mobile app, first released in 2019, provides a simpler digital notation solution for a wider audience of chess players.

idChess Application, 2019

It recognizes chess moves during a game, records them in the form of chess notation and saves them in a smartphone. idChess digitizes games in real time, so it can be used to broadcast games to a wide audience. Also, fans can get a link to the broadcast and open it on their phone.

idChess uses computer vision and machine learning technologies. idChess algorithms are able to take into account  many complex factors that may arise during the game, e.g., the fall of a chess piece, players’ hands  blocking the board, or shadows from the hands.

idChess became the first mass solution that allowed organizing broadcasts even at small children's tournaments in chess schools.

Chess AI has gone a long way. From the first algorithm only playing one game and losing, to Deep Blue overcoming the World Champion, to Stockfish playing chess on a level beyond what humans can achieve on their own and special AI-products that can digitize and broadcast chess games.

A man will hardly beat a machine now, but machines not only continue to improve, showing the brilliance of their creators, but can also assist chess players, with the machine learning technologies and computer vision leading the man to become the best versions of themselves.