# Computer Chess Club Archives

## Messages

### Subject: Re: The Limits of Positional Knowledge

Date: 14:46:36 11/18/99

Go up one level in this thread

```On November 18, 1999 at 16:38:41, Bella Freud wrote:

[snip]
>
>
>I agree that with a sequence
>
>a x b y c z d
>
>that if d is a bad move that only manifests itself over the horizon, then it is
>dangerous to have backed that best line and score back to the root.
>
>If d loses then there is a danger that a loses.
>
>The shorter the line the greater the danger and the longer the line the less the
>danger, since side 'a' can deviate from b,c,d,e,f,g,h in the move sequence.
>
>Or, that the deeper the line, the less chance there is that the position at d is
>so important, thus causing a to lose.
>
>Depth therefore helps bad evaluation functions to play better chess, in the
>sense that search applied at the next few moves allows for escapes from the
>consequent trouble. Such a system is just playing obstructionist chess.
>
>
>Bella
>

Yes, but wouldn't that imply that all programs play obstructionist chess until
we get to the level that chess is solved (or up to the point that they hit the
tablebases)?

In other words, trouble for most programs can come in one of three basic ways:
loss of game, loss of material, loss of positional factor (regardless of whether
the program is able to determine that). So, no matter how far you search and no
matter how good your evaluation is, there will always be a horizon effect. This
in turn implies that any program within any given game could always be walking
towards a cliff and not know it, just due to the peculiarities of the given
game/position and although a program (just like a human) might be capable of
avoiding a given cliff once it comes within view, even the act of avoiding it
may and most likely will be sending the game towards another cliff (i.e. the
positional/material factors of the game may have a great likelihood of resulting
in many cliffs or landmines in a given direction, not just one).