Excerpt from Danny Kopec'sRecent developments in computer chess[6]:
Dartmouth's program was the first program to challenge Northwestern's ascendancy in the ACM tournaments. In the ACM U.S. Computer Chess Championship, held in Atlanta, Northwestern's program was lucky to draw against Dartmouth. It only did so because the latter had no repetition check. Particularly encouraging was the fact that in a game of more than 50 moves, Dartmouth was never in a losing position.
The program is divided into two major evaluation functions, ĥ and đ. ĥ is concerned with the "soft", positional features of a given board position, while đ is concerned with the "hard" tactical features of a position. The specific chess concepts which comprise đ and ĥ are called "Detectors". A set of related detectors are assigned various values (weights) and are put into a table. ĥ includes tables such as Centre Control, Piece Mobility, Pawn Structure, King Safety, etc., while đ includes tables such as Pins, Forks, Discovered Attacks, Levers. The program is also divided into modules (Opening, Middle Game, and Endings) which allow greater flexibility in the assignment of weights. For example, in the opening, Piece-development, Centre control and King safety are stressed. A persistent problem which many programs still have is the too early development of the queen, because of its tremendous square control, mobility, and ability to produce threats. By assigning a value of -300 (where 100 = pawn) to every minor piece (B or N) still on the back rank, piece development is given prominence, since the program tries to get rid of these initial negative values. Other examples of tables which employ modular flexibility are Occupation of the Centre, and Rook on 7th. Greater weights are assigned to these in the middle game and ending than in the opening, to avoid moving the same pieces too often, before others have moved at all.
An idea which was never fully implemented was that of an "Attack-Defence Ratio". This is a measure of the difference between the sum of the forces attacking the quarter of the board where the enemy king is located and the sum of those forces which defend the same squares. If this difference in force is greater than a certain threshold value, an "alarm" is set off which results in a higher đ value and an increase in the depth of search. In this manner, long sacrificial variations are more carefully investigated. A benchmark of sacrificial positions would be a good test for its effectiveness.
Dartmouth's most "informed" table was the one on pawn formations, called "PFORM". Among its standard detectors were Isolated Pawns, Backward Pawns, Doubled Pawns, Passed Pawns, and Duos. Detectors such as Chains, Mini-chains, Shielded Backward Pawns, Potential Passed Pawns, and the table, "Levers", were among the more esoteric concepts which were added later. Many of these definitions were taken directly from Hans Kmoch's classic work Pawn Power[8]. The concept, Levers, using a modified definition of my own "pawn mover wich improve our formation and hurt our opponent's" - proved useful in the recognition of critical pawn moves. In addition, the levers concept helps to guide the placement of pieces especially in the opening and middle game. It could also help toward plan formation. Some further pawn formational concepts from Pawn Power which were never programmed were Outpost and Weak Square Complexes. The Dartmouth program is probably, in theory, capable of more sophisticated pawn formational evaluations than any other program; however the implementation is rudimentary. The program had at one time approximately 50 detectors in various tables and many others were planned.
Larry Harris (1974). Heuristic Search under Conditions of Error. Artificial Intelligence, Vol. 5, No. 3, pp. 217-234. ISSN 0004-3702. Also published (1977) under the title: The heuristic search: An alternative to the alpha-beta minimax procedure.Chess Skill in Man and Machine (ed. Peter W. Frey), pp. 157-166
a chess program developed in the early 70s at Dartmouth College by a team lead by professor Larry Harris. Primary author was Warren Montgomery [1], supported by co-authors Danny Kopec, Hal Terrie, David Levner and others [2]. Dartmouth CP was written in GCOS assembly, the language for the GE-635, to run under the Dartmouth Time Sharing System.
In his 1973 paper on Bandwidth Search [3], which was applied successfully to chess and therefor likely to Dartmouth CP, Larry Harris acknowledged professor Steve Garland and the undergraduates Warren Montgomery, Dave Chenerow, Dexter Kozen, and Steve Poulsen for their work on that topic [4].
Table of Contents
Challenging Chess
Excerpt from Danny Kopec's Recent developments in computer chess [6]:Evaluation
Danny Kopec in Recent developments in computer chess on the Evaluation of Dartmouth CP [7]:An idea which was never fully implemented was that of an "Attack-Defence Ratio". This is a measure of the difference between the sum of the forces attacking the quarter of the board where the enemy king is located and the sum of those forces which defend the same squares. If this difference in force is greater than a certain threshold value, an "alarm" is set off which results in a higher đ value and an increase in the depth of search. In this manner, long sacrificial variations are more carefully investigated. A benchmark of sacrificial positions would be a good test for its effectiveness.
Dartmouth's most "informed" table was the one on pawn formations, called "PFORM". Among its standard detectors were Isolated Pawns, Backward Pawns, Doubled Pawns, Passed Pawns, and Duos. Detectors such as Chains, Mini-chains, Shielded Backward Pawns, Potential Passed Pawns, and the table, "Levers", were among the more esoteric concepts which were added later. Many of these definitions were taken directly from Hans Kmoch's classic work Pawn Power [8]. The concept, Levers, using a modified definition of my own "pawn mover wich improve our formation and hurt our opponent's" - proved useful in the recognition of critical pawn moves. In addition, the levers concept helps to guide the placement of pieces especially in the opening and middle game. It could also help toward plan formation. Some further pawn formational concepts from Pawn Power which were never programmed were Outpost and Weak Square Complexes. The Dartmouth program is probably, in theory, capable of more sophisticated pawn formational evaluations than any other program; however the implementation is rudimentary. The program had at one time approximately 50 detectors in various tables and many others were planned.
Publications
External Links
References
What links here?
Up one Level