POLL: What is more important?

[xshat]

I was saying that chess engines (that are programmed to play perfect moves & have solved chess), will essentially play the same "perfect outline" hence being a clone.

The outside coding I was talking about is just that. The 32TB plus the perfect move /solved chess programming. It does not exist yet but could in the future. If chess is solved then why would a player be asking what move to make in a certain situation, since it is already outlined in 32men tablebases (assuming they are playing with that kind of assistance)?

[Marek]

I was saying that chess engines (that are programmed to play perfect moves & have solved chess), will essentially play the same "perfect outline" hence being a clone.

The outside coding I was talking about is just that. The 32TB plus the perfect move /solved chess programming. It does not exist yet but could in the future. If chess is solved then why would a player be asking what move to make in a certain situation, since it is already outlined in 32men tablebases (assuming they are playing with that kind of assistance)?

In a lost position all moves lose. The player or engine can look up the position in the 32TBs and every move will have the same assessment. So, what does he play? You think that there can be a perfect strategy in a lost position against an unknown opponent? It's impossible. Here's why. The player has no way of predicting the replies to the moves chosen by his "perfect" strategy. If he's lucky, somewhere down the line his opponent will make a mistake. But the world is populated by all kinds of chessplayers, some of them very strong while still liable to occasional mistakes. One sometimes moves his knights in the middlegame a little less than he should, another one over-values passed pawns, another one sometimes plays weak moves right after a time control, and so on. There is no one strategy that will maximise the points tally against all these and all the rest, if only because their errant characteristics can't be known in advance. A perfect strategy is a logical impossibility.

What there could be is a strategy (alongside 32TBs) that works in a particular tournament. It might do well in the next tournament too. But then another programmer could refine it, and maybe in a third tournament exceed the points of the winner of the first two tournaments. So "when the day comes that chess is solved, all chess engines will be clones of each other" won't be right.

[xshat]

I was saying that chess engines (that are programmed to play perfect moves & have solved chess), will essentially play the same "perfect outline" hence being a clone.

The outside coding I was talking about is just that. The 32TB plus the perfect move /solved chess programming. It does not exist yet but could in the future. If chess is solved then why would a player be asking what move to make in a certain situation, since it is already outlined in 32men tablebases (assuming they are playing with that kind of assistance)?

In a lost position all moves lose. The player or engine can look up the position in the 32TBs and every move will have the same assessment. So, what does he play? You think that there can be a perfect strategy in a lost position against an unknown opponent? It's impossible. Here's why. The player has no way of predicting the replies to the moves chosen by his "perfect" strategy. If he's lucky, somewhere down the line his opponent will make a mistake. But the world is populated by all kinds of chessplayers, some of them very strong while still liable to occasional mistakes. One sometimes moves his knights in the middlegame a little less than he should, another one over-values passed pawns, another one sometimes plays weak moves right after a time control, and so on. There is no one strategy that will maximise the points tally against all these and all the rest, if only because their errant characteristics can't be known in advance. A perfect strategy is a logical impossibility.

What there could be is a strategy (alongside 32TBs) that works in a particular tournament. It might do well in the next tournament too. But then another programmer could refine it, and maybe in a third tournament exceed the points of the winner of the first two tournaments. So "when the day comes that chess is solved, all chess engines will be clones of each other" won't be right.

In a lost position all moves lose, but if it's programmed with the intentions of versing imperfect moving humans then it will not give up after the first move.

At that point there is no perfect strategy, and since it can't win, it will play the moves which delay checkmate for as long as possible. There will certain positions in 32men tablebases which will take many moves.

There could be a single, or multiple strategy in this situation. A perfect strategy is logically possible only if chess is solved to be white forcing a win. In a losing position, there is only the most optimal strategy(s), which is delaying checkmate as long as possible. When the day comes that chess is solved, only the perfect moves that force checkmate will be the "cloned" part within the engines, the "optimal delay" features could be different.

[Marek]

I was saying that chess engines (that are programmed to play perfect moves & have solved chess), will essentially play the same "perfect outline" hence being a clone.

The outside coding I was talking about is just that. The 32TB plus the perfect move /solved chess programming. It does not exist yet but could in the future. If chess is solved then why would a player be asking what move to make in a certain situation, since it is already outlined in 32men tablebases (assuming they are playing with that kind of assistance)?

In a lost position all moves lose. The player or engine can look up the position in the 32TBs and every move will have the same assessment. So, what does he play? You think that there can be a perfect strategy in a lost position against an unknown opponent? It's impossible. Here's why. The player has no way of predicting the replies to the moves chosen by his "perfect" strategy. If he's lucky, somewhere down the line his opponent will make a mistake. But the world is populated by all kinds of chessplayers, some of them very strong while still liable to occasional mistakes. One sometimes moves his knights in the middlegame a little less than he should, another one over-values passed pawns, another one sometimes plays weak moves right after a time control, and so on. There is no one strategy that will maximise the points tally against all these and all the rest, if only because their errant characteristics can't be known in advance. A perfect strategy is a logical impossibility.

What there could be is a strategy (alongside 32TBs) that works in a particular tournament. It might do well in the next tournament too. But then another programmer could refine it, and maybe in a third tournament exceed the points of the winner of the first two tournaments. So "when the day comes that chess is solved, all chess engines will be clones of each other" won't be right.

In a lost position all moves lose, but if it's programmed with the intentions of versing imperfect moving humans then it will not give up after the first move.

At that point there is no perfect strategy, and since it can't win, it will play the moves which delay checkmate for as long as possible. There will certain positions in 32men tablebases which will take many moves.

There could be a single, or multiple strategy in this situation. A perfect strategy is logically possible only if chess is solved to be white forcing a win. In a losing position, there is only the most optimal strategy(s), which is delaying checkmate as long as possible. When the day comes that chess is solved, only the perfect moves that force checkmate will be the "cloned" part within the engines, the "optimal delay" features could be different.

We're nearly in agreement!

Delaying checkmate (in a lost position) seems a pretty good strategy. But is it the most optimal? What about avoiding symmetrical pawn structures? Or avoiding queen swaps? Etc. So the optimal strategy features could indeed be different and not cloned. Note that we're no longer talking about a perfect strategy.

[xshat]

I was saying that chess engines (that are programmed to play perfect moves & have solved chess), will essentially play the same "perfect outline" hence being a clone.

The outside coding I was talking about is just that. The 32TB plus the perfect move /solved chess programming. It does not exist yet but could in the future. If chess is solved then why would a player be asking what move to make in a certain situation, since it is already outlined in 32men tablebases (assuming they are playing with that kind of assistance)?

In a lost position all moves lose. The player or engine can look up the position in the 32TBs and every move will have the same assessment. So, what does he play? You think that there can be a perfect strategy in a lost position against an unknown opponent? It's impossible. Here's why. The player has no way of predicting the replies to the moves chosen by his "perfect" strategy. If he's lucky, somewhere down the line his opponent will make a mistake. But the world is populated by all kinds of chessplayers, some of them very strong while still liable to occasional mistakes. One sometimes moves his knights in the middlegame a little less than he should, another one over-values passed pawns, another one sometimes plays weak moves right after a time control, and so on. There is no one strategy that will maximise the points tally against all these and all the rest, if only because their errant characteristics can't be known in advance. A perfect strategy is a logical impossibility.

What there could be is a strategy (alongside 32TBs) that works in a particular tournament. It might do well in the next tournament too. But then another programmer could refine it, and maybe in a third tournament exceed the points of the winner of the first two tournaments. So "when the day comes that chess is solved, all chess engines will be clones of each other" won't be right.

In a lost position all moves lose, but if it's programmed with the intentions of versing imperfect moving humans then it will not give up after the first move.

At that point there is no perfect strategy, and since it can't win, it will play the moves which delay checkmate for as long as possible. There will certain positions in 32men tablebases which will take many moves.

There could be a single, or multiple strategy in this situation. A perfect strategy is logically possible only if chess is solved to be white forcing a win. In a losing position, there is only the most optimal strategy(s), which is delaying checkmate as long as possible. When the day comes that chess is solved, only the perfect moves that force checkmate will be the "cloned" part within the engines, the "optimal delay" features could be different.

We're nearly in agreement!

Delaying checkmate (in a lost position) seems a pretty good strategy. But is it the most optimal? What about avoiding symmetrical pawn structures? Or avoiding queen swaps? Etc. So the optimal strategy features could indeed be different and not cloned. Note that we're no longer talking about a perfect strategy.

I guess we won't know the specifics of what is most optimal until chess is solved, because that will provide the greatest insight. The perfect strategy would be cloned, while the optimal features used in losing positions may vary.

[hyatt]

I was saying that chess engines (that are programmed to play perfect moves & have solved chess), will essentially play the same "perfect outline" hence being a clone.

The outside coding I was talking about is just that. The 32TB plus the perfect move /solved chess programming. It does not exist yet but could in the future. If chess is solved then why would a player be asking what move to make in a certain situation, since it is already outlined in 32men tablebases (assuming they are playing with that kind of assistance)?

In a lost position all moves lose. The player or engine can look up the position in the 32TBs and every move will have the same assessment. So, what does he play? You think that there can be a perfect strategy in a lost position against an unknown opponent? It's impossible. Here's why. The player has no way of predicting the replies to the moves chosen by his "perfect" strategy. If he's lucky, somewhere down the line his opponent will make a mistake. But the world is populated by all kinds of chessplayers, some of them very strong while still liable to occasional mistakes. One sometimes moves his knights in the middlegame a little less than he should, another one over-values passed pawns, another one sometimes plays weak moves right after a time control, and so on. There is no one strategy that will maximise the points tally against all these and all the rest, if only because their errant characteristics can't be known in advance. A perfect strategy is a logical impossibility.

What there could be is a strategy (alongside 32TBs) that works in a particular tournament. It might do well in the next tournament too. But then another programmer could refine it, and maybe in a third tournament exceed the points of the winner of the first two tournaments. So "when the day comes that chess is solved, all chess engines will be clones of each other" won't be right.

In a lost position all moves lose, but if it's programmed with the intentions of versing imperfect moving humans then it will not give up after the first move.

At that point there is no perfect strategy, and since it can't win, it will play the moves which delay checkmate for as long as possible. There will certain positions in 32men tablebases which will take many moves.

There could be a single, or multiple strategy in this situation. A perfect strategy is logically possible only if chess is solved to be white forcing a win. In a losing position, there is only the most optimal strategy(s), which is delaying checkmate as long as possible. When the day comes that chess is solved, only the perfect moves that force checkmate will be the "cloned" part within the engines, the "optimal delay" features could be different.

We're nearly in agreement!

Delaying checkmate (in a lost position) seems a pretty good strategy. But is it the most optimal? What about avoiding symmetrical pawn structures? Or avoiding queen swaps? Etc. So the optimal strategy features could indeed be different and not cloned. Note that we're no longer talking about a perfect strategy.

This is an easy question to answer. Delaying mate as long as possible is _not_ the best strategy against an imperfect player. Often that would involve sacrificing your pieces which can make the process simpler. I'd much rather play in a lost KRRB vs KRB position as opposed to KR vs K. For draws it is even more complex and that's the reason I did the "swindle mode" in crafty. The best EGTB move is difficult to define when almost all moves lead to a draw. I try to pick the one that hangs on to material to give the opponent a chance to covert this draw into a loss...

[Marek]

I guess we won't know the specifics of what is most optimal until chess is solved, because that will provide the greatest insight. The perfect strategy would be cloned, while the optimal features used in losing positions may vary.

Yes, "the optimal features used in losing positions may vary." But then you regress by talking about a perfect strategy again. When chess is "solved" in the game theory meaning of "solved," i.e. when there are 32TBs available, there is still no perfect strategy (unless you call looking up a TB entry a strategy, which anyway only applies to singleton solutions in won or drawn positions).

Of course, just because there are 32TBs available, won't mean that any two engines would access them exactly as quickly.

[xshat]

I guess we won't know the specifics of what is most optimal until chess is solved, because that will provide the greatest insight. The perfect strategy would be cloned, while the optimal features used in losing positions may vary.

Yes, "the optimal features used in losing positions may vary." But then you regress by talking about a perfect strategy again. When chess is "solved" in the game theory meaning of "solved," i.e. when there are 32TBs available, there is still no perfect strategy (unless you call looking up a TB entry a strategy, which anyway only applies to singleton solutions in won or drawn positions).

Of course, just because there are 32TBs available, won't mean that any two engines would access them exactly as quickly.

There could be a perfect strategy, that is not regression, that is a possible future for when chess is solved. Until then it seems to be a balance of material and position.

[Marek]

Delaying checkmate (in a lost position) seems a pretty good strategy. But is it the most optimal? What about avoiding symmetrical pawn structures? Or avoiding queen swaps? Etc. So the optimal strategy features could indeed be different and not cloned. Note that we're no longer talking about a perfect strategy.

This is an easy question to answer. Delaying mate as long as possible is _not_ the best strategy against an imperfect player. Often that would involve sacrificing your pieces which can make the process simpler. I'd much rather play in a lost KRRB vs KRB position as opposed to KR vs K. For draws it is even more complex and that's the reason I did the "swindle mode" in crafty. The best EGTB move is difficult to define when almost all moves lead to a draw. I try to pick the one that hangs on to material to give the opponent a chance to covert this draw into a loss...

It's easy to answer that delaying mate is not the best strategy, but it's somewhat harder to prove it. Nevertheless, I would be inclined to agree with you - except for one thing. Because we're discussing what's best, not what's good or practical, we expose ourselves to an academic quibble. We can say what's best (or not best) only if we make assumptions about our opponents. It is of course entirely reasonable for you to make assumptions, based on your considerable experience, about Crafty's (likely) opponents in the present or imminent chess landscape.

[Marek]

There could be a perfect strategy, that is not regression, that is a possible future for when chess is solved. Until then it seems to be a balance of material and position.

How could there be a perfect strategy in a drawn or lost position against unknown opponents, including imperfect opponents? An opponent could be any opponent. In theory, an opponent could play any game (sequence of positions) in the 32TBs. There is no perfect strategy to cope with that. Remember that a "perfect" strategy (alongside 32TBs) may sometimes get a draw from a lost position (or a win from a drawn position), but sometimes it won't. What's to stop an opposing programmer noting where the strategy failed thereby improving his own imperfect program relative to the one with "perfect" strategy? The perfect strategy itself can't be improved; for how is it possible to improve on perfection? Unless it was never perfect in the first place.