It must be in the positional area, it can't be otherwise.
So in the case of Rybka it seems to be the focus of chess knowledge that has changed where priority has been given to strong pawn positions, particularly advanced and supported pawns not only centrally but also on the wings.
In some games it becomes clear that while opponent goes off on some wild goose chase king attack, Rybka is only interested in advancing and supporting pawns. Sometimes that is its achilles heel but more often than not is the key to Rybka's superiority.
How many Grandmasters try to mate their opponent as initial goal? Most look for positional advantages that may accumulate to a pawn advantage or pawn position advantage that may be enough to win. What Rybka's inbuilt knowledge seems to have done is clarify the positional moves at some expense of tactical or combinative knowledge. However, in doing so it has improved the evaluation to dispense with less promising moves quicker and hence its displayed search depth usually shows deeper than opponents in same position on equal hardware.
The main question then is how has it harnessed this knowledge in a way that makes it so competitive compared to other engines?
There are very high evaluations for passed pawns and advanced pawns that are clearly in error when it comes to opposite colour bishop endings so there is strong evaluation emphasis on where a pawn is on the board and whether it is blocked by opponents pawn without fully analysing if the pawn can actually be promoted. This suggests a number of evaluation short cuts that speed up the calculations and work in most cases but as mentioned above can get caught out with grossly erroneous evaluations. The major and minor pieces also work to support the pawns, On many occasions Rybka can give the impression of strong tactical awareness but on closer analysis it came about because of the strength of or threat from pawns. In many games, Rybka will identify winning pawn advances several moves before opponent engine finds the danger, all too often too late.
In terms of Rybka's evaluation values there may be a misconception that the values should be interpreted by normal convention. However a while back, there was some very interesting analysis in Eric Hallsworth's Selective Search magazine that showed that Rybka's evaluation process seemed very different to other engines. There is an assunption too that the true evaluation is also not hidden from view!
PeterG
