COKO III: The Cooper-Kozdrowicki chess program

Abstract

The performance capabilities of the best computer chess programs are compared with their human counterparts with emphasis being placed on machine behavior limits. A grandmaster usually spends a lifetime collecting knowledge or information about the game. Some of this knowledge is given to COKO in the form of a 12 000-line FORTRAN program. Using this knowledge COKO plays very poorly but at the super rate of approximately one move/see. The use of a brute-force selective tree-searching procedure yields an order of magnitude improvement in performance at the standard rate of 3 min/move. Perhaps three orders of magnitude additional improvement is needed to defeat the world champion, a gap which must be bridged, if ever, by programming more chess knowledge into the machine. This paper discusses the “tree-searching catastrophe” as a natural phenomenon that plagues selective tree searching for both man and machine. In addition so-called “interminimal-game communication” is considered as a natural, powerful procedure frequently used by humans to guide their selective search and as a point of emphasis for future development. It is concluded that COKO's development is just beginning, with no immediate barriers to progress, and no lack of ideas for improvement. At present COKO combines brilliant solutions to individual board position puzzles with unimaginable blunders.