POLL: What is more important?

[Uly]

You claim that when chess is perfected all engines will be the same, because on a given position all of them will always play the perfect game, that will end in draw. That's equivalent to saying that they will offer draw in move 1 (because they assume they're playing someone that also has 32men TB) which is equivalent to saying that they'd resign if the position was lost.

The truth is that if the position is lost they're still going to try to win, just like they'd do if the position is drawn. There are many ways to try to win a drawn position, thus, all engines will play different moves to try to achieve victory.

[xshat]

You claim that when chess is perfected all engines will be the same, because on a given position all of them will always play the perfect game, that will end in draw. That's equivalent to saying that they will offer draw in move 1 (because they assume they're playing someone that also has 32men TB) which is equivalent to saying that they'd resign if the position was lost.

The truth is that if the position is lost they're still going to try to win, just like they'd do if the position is drawn. There are many ways to try to win a drawn position, thus, all engines will play different moves to try to achieve victory.

It may or may not end in a draw, it is not yet known whether chess is solved as a draw or forced win. It is NOT equivilent to saying the engine will offer a draw on the first move because it is highly unlikely the engines would be programmed to give up on the first move even if it was positioned as a loss, considering they ought to be programmed not to do so since they would be made for versing humans (at that point engine vs. engine games would be negligable).

Yes, they are still going to try to win as long as they are programmed not to forfeit on the first move. There are many ways and when chess is solved it will outline one or more ways that engines will always adhere to first.

[Uly]

So finally you accept that when chess is solved engines will play differently and thus won't be clones of each other?

[xshat]

That is not what I said. They will be clones in the respect that they follow the outlines for perfect play to always win in specific situations.

[Uly]

So you say they will be clones only on the winning positions? What about drawn and lost ones? Please reread what I've been saying, for the engines to try to win from such situations, they have to play different from each other.

[Marek]

It was never my idea that all engines with 32TBs will play the same moves (unless there's only one winning move); it was never my idea that a perfect engine would immediately resign a lost position either.

Yes, they were xshat's ideas (the former, the latter is just an extension), I agree with your arguments, it's him who seems to be stuck on a loop. Like in his last post, where he says the TB's are going to tell the engine whether to play e5 or c5, which is impossible if both moves lead to draw, the engine has to decide which one to play in some other manner.

Sorry Ovyron! Regard it as friendly fire.

[Marek]

32TB's will tell it what to do and it will not be asking human questions such s whether to play c5 or e5. The tablebases will say. If programmed correctly, it will have solved chess from move 1.

No. The tablebases have insufficient information. Let's say 1.e4 wins. The perfect engine is Black. What move does it play? All 20 legal moves lose according to 32TBs, but that doesn't mean that the perfect engine will lose against all opposition. What is the right move for Black now? In practice, the 20 legal moves will perform differently - this is not contained in the tablebases.

I could program my perfect engine to favour gambits when in a lost or drawn position, in the hope that the opposition goes wrong in the complications. Another programmer could favour imbalanced endgames. Another programmer could have a third strategy or a refinement of mine. And so on.

Now stick us all in a grand tourney of many thousand games. Indeed, have many thousands of such tourneys - which is like the present testing scene with its various rating lists. Let Rybka 4 be one of the participants, but without the 32TBs that some of the others will have. Rybka is imperfect - we know that already - but it is quite strong. It will win some won positions and/or it will draw some drawn positions, however few. It could even draw some won positions, which against perfect opposition counts as a success. It doesn't matter if it misses a mate in 40 but delivers a mate in 70 instead; it doesn't matter if it struggles to draw providing it still draws.

Who will win the series of grand tourneys? Obviously it won't be an engine that resigns on move one. It will be an engine that scores best against the likes of Rybka. It will be the programmer of a perfect engine who, in a necessary addition to the 32TBs, has happened on the best algorithm to play a lost or drawn position in these circumstances. And in order to further improve his engine's performance, he can try to refine his algorithm. All the other programmers will try to refine theirs in preparation for the next round of tourneys. Full tablebases will not put a stop to competitive chess engine programming. I guess it would dampen the enthusiasm though.

Some further points... If we're allowing perfect engines to possess 32TBs, which would be a hardware miracle, what hardware should we allow the imperfect Rybka 4? I could imagine competitions to program engines without 32TBs (because of arbitrary hardware restrictions) that, starting from a won position, could still occasionally take points off perfect engines. Finally, remember that engines are also used for analysis. Whatever the true assessment of the starting position in chess, a perfect engine may still get asked to play a drawn or losing position - and then, as I've said, 32TBs won't provide all the answers.

[Marek]

That is not what I said. They will be clones in the respect that they follow the outlines for perfect play to always win in specific situations.

There are no "outlines for perfect play." In a lost position, 32TBs will tell you that every move of yours loses. Where is your outline now?

If we had an outline for perfect play we wouldn't need tablebases.

[xshat]

So you say they will be clones only on the winning positions? What about drawn and lost ones? Please reread what I've been saying, for the engines to try to win from such situations, they have to play different from each other.

If chess is solved then they will play the same.

[xshat]

32TB's will tell it what to do and it will not be asking human questions such s whether to play c5 or e5. The tablebases will say. If programmed correctly, it will have solved chess from move 1.

No. The tablebases have insufficient information. Let's say 1.e4 wins. The perfect engine is Black. What move does it play? All 20 legal moves lose according to 32TBs, but that doesn't mean that the perfect engine will lose against all opposition. What is the right move for Black now? In practice, the 20 legal moves will perform differently - this is not contained in the tablebases.

I could program my perfect engine to favour gambits when in a lost or drawn position, in the hope that the opposition goes wrong in the complications. Another programmer could favour imbalanced endgames. Another programmer could have a third strategy or a refinement of mine. And so on.

Now stick us all in a grand tourney of many thousand games. Indeed, have many thousands of such tourneys - which is like the present testing scene with its various rating lists. Let Rybka 4 be one of the participants, but without the 32TBs that some of the others will have. Rybka is imperfect - we know that already - but it is quite strong. It will win some won positions and/or it will draw some drawn positions, however few. It could even draw some won positions, which against perfect opposition counts as a success. It doesn't matter if it misses a mate in 40 but delivers a mate in 70 instead; it doesn't matter if it struggles to draw providing it still draws.

Who will win the series of grand tourneys? Obviously it won't be an engine that resigns on move one. It will be an engine that scores best against the likes of Rybka. It will be the programmer of a perfect engine who, in a necessary addition to the 32TBs, has happened on the best algorithm to play a lost or drawn position in these circumstances. And in order to further improve his engine's performance, he can try to refine his algorithm. All the other programmers will try to refine theirs in preparation for the next round of tourneys. Full tablebases will not put a stop to competitive chess engine programming. I guess it would dampen the enthusiasm though.

Some further points... If we're allowing perfect engines to possess 32TBs, which would be a hardware miracle, what hardware should we allow the imperfect Rybka 4? I could imagine competitions to program engines without 32TBs (because of arbitrary hardware restrictions) that, starting from a won position, could still occasionally take points off perfect engines. Finally, remember that engines are also used for analysis. Whatever the true assessment of the starting position in chess, a perfect engine may still get asked to play a drawn or losing position - and then, as I've said, 32TBs won't provide all the answers.

If chess is solved then it is impossible to have "insufficient information". And Like i said it would not give up after the first move even if in a losing position.

32men tablebases would not work with r4 because it is not programmed to have solved chess, which includes coding outside of the 32men bases.