Tuesday, July 14, 2015

Maniac I, an Early Chess Playing Computer

Los Alamos scientists Paul Stein and Nick Metropolis with Maniac I

     In the January 1957 issue of Chess Review Magazine there was an article titled “Experiments in Chess On Electronic Computing Machines.” It was written by mathematician and mathematical physicist Stan Ulam and his colleague P. Stein of the Massachusetts Institute of Technology. Professor Ulam had done work with atomic energy and 'computing machines.' He normally worked at the Los Alamos Scientific Laboratory but at the time was serving as a visiting professor at MIT. 
     The experiment they ran was the first time that a computer played an entire game based on instructions of a strictly general nature. Previous attempts had never concerned themselves with more than either short mating combinations where the computer would find the solution by running through a series of legal moves or elementary ending where simple instructions would enable the machine to avoid blunders or where routine ways of forcing mate were included in the instructions move by move. The method was similar to the way a human would consult a book like Fine's Basic Chess Endings
     Previously Norbert Weiner had written a book, Cybernetics, where he speculated that it might be possible to construct a computer that 'might be as good a player as the majority of mankind.' and Claude Shannon had actually constructed a simple home-made computer that was able to play a few simple positions. Shannon's computer used a starting position with a reduced number of pieces so as to avoid over-taxing the computer's memory. 
     When Prof. Ulam had discussed the possibility of making a chess-playing machine many of his colleagues flatly refused to believe that it was possible to construct a machine that played on its own. Ulam described how many of them 'violently rejected the whole idea of a machine being able to do something akin to thinking.' Ulam admitted that because men's thinking process was not fully understood, it would be impossible to design a machine that thinks the way humans do, but he believed that if it was given a set of instructions, it could arrive at results that a human could by 'thinking.' 
     In the article Edward Lasker, also a strong checker player, described how he had recently played a game of checkers against one of the most advanced 'electronic computers' designed by IBM. The designer, Dr. Arthur Samuel, had programmed the computer to calculate three moves ahead and it had defeated Lasker with some rather brilliant play, so Lasker thought it might be possible to design a computer to play a decent game of chess. 
     A group of scientists, including chess players, at Los Alamos decided to construct a code based on only two factors, material and mobility. Shannon had described such a possibility earlier and Alan Turing had previously written a code that allowed his machine to see one move ahead, but it did not play very good chess and was easily beaten. 
     In order to test their ideas, Ulam and Stein used a 6x6 board and the machine was designed to look two moves ahead for each side (known today as 4 plies) so that it could play a game in a reasonable amount of time, about 10 minutes per move. The computer they used, Maniac I, was an older, slower computer in their laboratory and the estimated that it could perform about 10,000 elementary operations per second. 
     Here is a game played by their machine and analyzed using Houdini 1.5. I set the position up on Fritz using the 'Position Setup' feature in order to restrict it to the 6x6 board. Analysis proved to be kind of tricky because a) the position got complicated and b) the engine was unaware that because of the restricted size of the board, a lot of better moves were not possible!
 

No comments:

Post a Comment