Computer chess notes
Back to the Computers and Chess page.
- Botvinnik's "mathematical map of the position" included information
on how long it would take each piece to get to each square
- In "Machine Intelligence 2", Jack Good discusses "turbulence"
(the opposite of "quiescence") suggesting that it could be defined with respect to goals other than material balance. He also discusses "agitation"
which also takes into account whether the position is obviously won/drawn/lost.
"stability of evaluation" is a theme which runs through both planning and
annotating tasks. Planning depends on there being some kind of stability
in the position - a weak, isolated pawn is likely to remain so for a
number of moves. If the situation is unstable (e.g. a 7-ply search
and a 9-ply search come to very different conclusions) then planning
won't be so useful, and (from the Annotations point of view) there will
be more "interesting" or "critical" moves. I think Kasparov aims for
unstable positions, but can such positions be characterised simply by looking
at the evaluation tree in more detail?
- DarkThought Goes Deep "even at high search
depths of 11-14 plies modern chess programs steadily discover new best moves in still 16% of all searches on average." (by Ernst A. Heinz in: ICCA Journal, Vol. 21(4), pp. 228-244, Dec. 1998).
- From "A Computer Chess Tutorial" by Norman Whaland (BYTE October 1978)
- Chess games between computers are often dull because the programs don't
follow any plan ... Here are some of the types of specific goals
that occur frequently: Get control of a key square; Attack an area of the
board where the opponent is weak; Free an immobile piece;
- Most chess programs spend almost all of their time considering silly moves.
There are two main types of silly moves: moves irrelevant to the important
goals of the position, and sacrifices that gain nothing
- From "Using Patterns and Plans in Chess", David Wilkins, Artificial Intelligence, V14, 1980
- The purpose of this research is to investigate the issues involved in
expressing and using pattern-orientated knowledge to analyse a position, to
provide direction for the search, and to communicate useful results from
- In a knowledge based program, the cost of processing the knowledge
should be offset by a significant smaller branching factor in the tree search.
- It is very important to get the correct level of detail in a plan.
The plan should handle as many replies as possible without causing a re-analysis, but it should avoid suggesting poor moves.
- By communicating plans down the tree, PARADISE [PAttern Recognition
Applied to DIrecting SEarch] can understand new positions on the basis of
its analysis of previous positions.
- .bishop computer chess publications
- "Chess Metaphors", Diego Rasskin-Gutman (MIT, 2009) has quite a lot of
biology in it, and history of chess and AI. Its "Brute Force or Heuristic?"
section is useful - some of the older heuristic approaches might be useful
again. I noticed some typos - on p.45 I think there should be a pawn on e5; where the book uses "recursion" it means
"iteration"; on p.142 his use of "AND" needs clarifying
Updated: Dec 2009