Interesting thoughts by many on this topic, but I'd like to add a comment to that of <CIO>. A defintion of 'strategy' might be "a plan of action designed to achieve a particular goal." (http://en.wikipedia.org/wiki/Strategy) Using this definition, an engine does not use strategy, unlike a human player. Instead, from a given position, an engine generates candidate moves and evaluates the resulting positions. The process is repeated iteratively for each half move. I think the engine is looking for the next best move based on its evaluation function, and not to implement a strategy.

and here is some information about Pandix (along with other engines participating in the 10th Int'l tournament the above game comes from)http://www.csvn.nl/index.php?option...

<zanshin>, lol, that's exactly what I did: attribute a long term strategy to Rybka, where there was only good move followed by good move.

Best move versus strategical planning - it's hard to reconcile how the moves can fit together harmoniously if there is no overall guiding plan.

So Rybka simply sees a little further and a little faster into the future than other engines and by doing so sets a flexible plan into motion, move by move, that it doesn't actually envision?

It's hard for me to believe that Rybka would typically beat a super GM, and do it without a plan. But then, I've seen the weak moves (according to engines) those same GM's make game after game, and can see how engines would be able to capitalize on them.

Therefore, consistently strong moves trump a long term strategic plan executed with some strong and some weak moves. That makes sense.

I think the answer, at least at the moment, is probably no. But there are circumstances where a computer can appear to have a long term strategy in mind.

Let's start with a definition. I would define a strategy as a long-term plan which determines a number of moves. So a strategy might be, for example, to gang up on a weak enemy pawn, to exchange pieces when ahead in material or to push a queenside pawn majority.

As far as I am aware, most software engines evaluate each position afresh. They do not usually carry information from one position to the similar position that arises the next move on. In this respect, computers are like hyper-smart goldfish. They are incredibly intelligent, but can only remember something for a short while. That is why the poor goldfish thinks he is swimming in the ocean rather than a tiny glass bowl. It all looks new to him.

So computers don't think in terms of long term strategies.

But most computer engines are programmed to take positional and strategical considerations into account when evaluating a position. So they value central pawns more than wing pawns in the opening and middlegame. A centre pawn might be valued at 1.02 and a wing pawn as 0.98, for example. Using similar evalution tricks, engines can also be programmed to put rooks on open and half open files, to centralise pieces, find knight outposts, castle and so on.

This means that computers will play moves that are strategically sound. And this can develop into a consistent approach over several moves. An engine that is programmed to exploit weak pawns will keep banging away at a backward enemy pawn, as if it was pursuing a deliberate strategy. It doesn't consitute a plan, although it might appear to do so. Each move is calculated on its own merits.

No, Rybka is the fish ;-)

<that's exactly what I did: attribute a long term strategy to Rybka, where there was only good move followed by good move.>

Because you think like a human <CIO>!

<Best move versus strategical planning - it's hard to reconcile how the moves can fit together harmoniously if there is no overall guiding plan.>

The only 'plan' of the engine is to seek the line yielding the highest numerical evaluation, asssuming best play from both sides. You can see the effect of lacking a strategy when you have quiet, non-tactical positions. Typically, an engine will start shuffling pieces (often Rooks) from one square to another. This is a dead giveaway that the engine 'does not know what to do'. But come to think of it, you could say the same for a human player!

<So Rybka simply sees a little further and a little faster into the future than other engines>

I think she sees more effectively rather than faster or deeper. It's really hard to get an idea of how fast she is evaluating or how deep the plies are running because Vas Rajlich (her programmer), is famous for trying to mislead competitors by under-reporting node counts and ply depths.

<consistently strong moves trump a long term strategic plan executed with some strong and some weak moves.>

<Once> Actually, they do. Numerical evaluations of positions are stored in the hash (or transposition) tables. This can save time because a position that has arisen previously, e.g. by transposition, does not have to be evaluated again.

More importantly, evaluations obtained further down a line can significantly affect analysis of an earlier move. In fact, that's the reason auto-analysis modes start at the end of a game, and move towards the start. The engine analyses after having seen the outcome of the game! This technique is also useful in analyzing a strategy. You can effectively 'convince' an engine that a move is better by showing its outcome.

I may be wrong, but I think Rybka may only flush branches of its tree once a contrary move is made. Even then, it seems to keep some calculations around.

And I must admit that I have bought quite a few as Bill Gates continues his bid for world domination by making me buy a new computer every few years.

But I think the point still stands. The more recent engines may store some variations and evaluations, but I don't think they plan in the way that we would define it.

<Once> On this main point, we agree. Engines do not plan strategy in the way humans do. Humans think of an objective, then try to find the moves that might achieve it. Engines look for the next best move - then the next best reply, then the next move move ... etc.

"Positional Chess" means thinking in abstract terms; a powerful way to tame the vastness of chess, but inherently imperfect.

"Tactical Chess" means concrete, certain calculation; it is potentially perfect, but practically limited because chess is so vast.

Computers' tactical power are now reaching a level where in many positions they can all but tame the jungle of variations, but... not quite. And so the Centaur remains the strongest beast in the jungle.

But if you could build a computer of infinite calculating power, let's all it "Laplace's Demon", it would perfectly calculate chess down to a forced draw or win, and "Positional Chess" would be shown to be a series of educated guesses, mostly right, sometimes wrong, with no over-arching pattern that would be meaningful to us brainy apes.

And it would no longer be a game at all.