BGonline.org Forums

Exploiting a weak opponent in chess

Posted By: Timothy Chow
Date: Thursday, 28 September 2023, at 3:32 a.m.

I just finished reading One Jump Ahead: Challenging Human Supremacy in Checkers, by Jonathan Schaeffer. Written in the late 1990s, it recounts Schaeffer's quest to create a computer program, dubbed "Chinook," that would be able to beat the greatest 8x8 checker player of all time, Marion Tinsley. It's a fascinating story, and Schaeffer tells it very well. If you don't want to slog through the whole 450-page book, then you can check out the short version by Jim Propp.

However, the main reason I'm mentioning Schaeffer's book is that a footnote on page 342 caught my eye. It mentions Peter Jansen's 1992 Ph.D. thesis, Using Knowledge about the Opponent in Game-Tree Search. The last section of the thesis is particularly interesting. It discusses the possibility of playing sub-optimal moves in order to increase the chances of a mistake by your opponent.

Specifically, Jansen studies the chess endgame of King and Queen against King and Rook (KQKR), which is almost always a theoretical win for the player with the Queen. However, if the player with the Queen plays inaccurately, and takes too long to clinch victory, then the player with the Rook might be able to claim a draw by virtue of the fifty-move rule. Jansen's computer program tries to confront the stronger side with "difficult" decisions, where "difficulty" is measured in various ways (e.g., a difficult move might go against a heuristic that humans often use, or it might be a needle in a haystack of many wrong moves). In some cases, Jansen was able to make the case for making a "bad" move (one that would shorten the game against a perfect player), because it would increase the chances that a fallible opponent would make a mistake.

I had never heard of this thesis before. Google Scholar lists many papers that cite it, which I might try to look at more carefully. Many of the papers concern games like poker, which have imperfect information, but some of them are about games of perfect information. Unfortunately, I didn't see any papers on backgammon. It's not clear to me how many of Jansen's ideas would carry over from chess to backgammon.

Messages In This Thread

• Exploiting a weak opponent in chess
  Timothy Chow -- Thursday, 28 September 2023, at 3:32 a.m.
  • Exploiting a weak opponent in chess
    Frank Berger -- Thursday, 28 September 2023, at 2:19 p.m.
    • Scrabble™
      Timothy Chow -- Sunday, 1 October 2023, at 2:15 p.m.
      • Scrabble™
        Nack Ballard -- Sunday, 1 October 2023, at 3:18 p.m.