AI Chess, (A.I. Chess)
a chess program by Marty Hirsch and predecessor of MChess. AI Chess was written in 8086 assembly language to ran on an IBM PC under DOS . It played the ACM 1988 in Orlando, the WCCC 1989 in Edmonton and the WMCCC 1989 in Portorož .
A.I. Chess uses a fairly complicated algorithm combining full-width search, selective search, and a "layered" quiescence search which behaves differently at different levels in the search tree. The program performs an iterative full-width search using a modified form of the Principal-Variation-Search (PVS) algorithm. On top of this, it does a combined selective/quiscence analysis. A.I. Chess has the unusual feature of sometimes re-searching a "quiscence node" with a full-width investigation.
The quiescence search incorporates a detailed "threat analysis" and therefore, the program spots may combinations long before a contrasting "brute force" approach would find them. The gain (from needing less full-width plies) seems to exceed the loss in speed by a significant amount.
Position evaluation starts by considering if the side to move is threatened with pawn promotion, check, or double attack, or has trapped, pinned, or skewered pieces. Penalties similar to swap-off scores are imposed if the position is too deep to merit a re-search. Scores are then added for other tactical patterns, pressure on pieces and pawns, development, King safety, passed pawns, pawn structure, outposts, and mobility.
Some types of endgame positions are scored differently, by pattern recognition processing. The program is alert to simplifications, and to tactics involving passed pawns.
- Monty Newborn, Danny Kopec (1989). Results of The Nineteenth ACM North American Computer Chess Championship. Communications of the ACM, Vol. 32, No. 10, pdf
- AI Chess' ICGA Tournaments
- Kings Move - Welcome to the 1989 AGT World Computer Chess Championship. Edmonton, Alberta, Canada, Courtesy of Peter Jennings, from The Computer History Museum, pdf