True or false? Chess will never be solved! why?

           Sucking up to the boss isn't gonna help. Try sending him some money$$   Out of respect for the other members and the OP of THIS thread, I'll obey the rules and leave. Why don't you take up my invitation and come express your anger in MY THREAD??  Its dedecated to YOU. 

Thought experiment:

Imagine that humans don't tell computers that a position with two kings and a bishop is a draw. How long does it take to find this out with the method engines use and with the method tablebases use?

The only way an engine can find out that such a position is a draw is by the 50 move rule. It is safe to say that there are at least 10 legal moves on average in such a position ( depending on which side is to move and how the pieces are placed the actual figure will change move by move ). Under the fifty move rule each side has to make 50 moves without capture, pawn advance or promotion ( that is 50 full moves, 100 halfmoves ) before a draw can be claimed. This means 10^100 possible games, which is astronomic.

Now, let's look at how a tablebase would find out that the position is a draw. A tablebase would first generate all positions with two kings and a bishop ( or less - that is two kings ). There are roughly 10^5 such positions ( a hundred thousand ). Generating 10^5 positions is nuts for today's computers. In the second step the tablebase generating software will look at all those positions to find out whether they are mate. This can also be done fairly quickly. Since it will soon become evident that no position involving two kings and a bishop ( or two kings ) is a mate, the tablebase will proudly announce after examining 10^5 positions that the position is a draw. At that time the engine will be still be struggling at depth 5 ( since at depth 5 there are already 10^5 possible games ) with its tree search ...

You have to note that not everything that is possible to compute mathematically is possible to compute in a physical sense as well. There are physical limits to what can be computed in what time and space. There are computational problems which you can't solve even if you turn the entire observable universe into a computer. It is safe to say that those problems are unsolvable.

While an engine and a tablebase may be equivalent in the final analysis in a purely mathematical sense, they are not equivalent in a physical sense. As far as we are beings limited by physics, we can safely say that they are not equivalent.

White to move:

Engines think that black is winning here. After all black has a rook and a pawn against a single knight. However it turns out that this position is a fortress and white can draw. Engines don't see this because draw by the 50 move rule is beyond their search horizon. It is very difficult for an engine to establish a draw by the 50 move rule ( if not impossible ). If you lower the limit of the 50 move rule and make it a 10 move rule ( it is possible to configure Houdini like this ) than the draw is found almost immediately.

This position is from a real game succesfully drawn as white by the great Lasker. What Lasker as a human could find out, engines can't find out. This real life example yet again highlights a major weakness of engines which can be easily overcome by tablebases ( since it is only a 5 men position ).

I was always poor in Math; but excelled in English and history.  A close friend (who plays chess) learns much faster than I do, simply from playing against a computer, while I toil away at books, books, and more books.  

He appears "talented" in chess, while I feel 'short-changed' and limited!  

thanks for the reply.  

Huh?

It seems more like a concession that you don't know how engines work...

What do you mean solved?

Let's say a computer program is eventually invented that achieves a rating of 3500. And no human can ever beat it (I would say if it plays 1,000 games against the top 10 GMs and loses none as black or white, that is proof of its unbeatableness, some may disagree with the 1,000 figure but that's up to an individual: there is a number that realistically constitutes proof).

Then you set it to play itself, and over 1,000 games, it's record is W0, L0, D1000

Is that solved?

I'm rapidly coming to the conclusion as a consequence of this thread that engines are not the right tools to solve chess. In fact they seem to be useless when it comes to solving chess. Engines are all about generating lines of play from a given position, while solving chess is all about systematic generation of positions in the form of tablebases. Engines and tablebases attack the problem from very different directions and solving chess is more amenable to the approach that tablebases take. Engine evaluations necessarily reflect human opinions about the merits of chess positions while solutions to chess may completely disregard 'common sense' as understood by humans. Common sense works in 99 % of the cases but the remaining 1 % is the realm of bizarre many hundred move counter intuitive forcing lines.

Am I right in assuming that any engine that prunes lines (which is just about all engines these days) is by definition not solving chess? And any engine that doesn't prune lines would take almost forever to solve chess.

I think this sums up my opinion better than I could do it.

@watcha: interesting post about the way a chess engine and the way a tablebase would solve the problem of a certain position.

Your basic argument is correct imo, but a chess engine might use different ways to evaluate a position:

to deliver mate for instance you must have the capacity to restrict the movements of the opponent. This idea reminded me of the statistical notion of degrees of freedom. That is not possible with only one bishop. The degrees of freedom of the king of the opponent can never be brought back to zero by one bishop only.

My point is that a chess engine is not restricted to calculate lines to reach an evaluation.

Chess engines - as we know them - calculate lines. This can not be otherwise because we expect an engine to come up with a reasonable move ( which is not necessarily the perfect move ) within a few seconds or minutes. Any program which involves features beyond calculating lines from a given position, for example one that generates all possible positions for a certain set of pieces first, and then tries to find forcing ways backwards from mates to the position it was presented with, does not fit this definition because it will have a very long think ( days or months possibly ) without giving any clue as to the best move, but after that it will present the solution with 100% certainty. Such a program is not an engine any more. It is a tablebase.

Apart from that I have not seen any method that is useful for finding chess moves other than tree search ( whether it is done in the way engines do it, or the reverse way tablebases do it ). Such a method would only work if chess turned out to be computationally reducible ( that is: formulas exists which can tell for certain the outcome of positions without tree search ). So far there is no evidence for this and there is a lot of evidence to the contrary.

Do you realize this position is still a draw even after Kd3-d4? Unless black loses a knight or allows white's knight sacrifice, black will simply claim the draw at move 50+. There is no forced mate at all, because the 50 move rule applies first. Databases don't take it into consideration.

Ok let's make it a chess composition. White to play and win.

No way under actual rules, that is the point.

I have two comments:

First, the 50 move rule is a rather arbitrary, ad hoc kind of rule which does not follow in any organic way from the essence of the game and can be subject to change at any time in the future.

However this is the rule, so we have to take it seriously.

Now, it is true that this particular position having only 4 non king pieces, none of them being pawns, offers little room for captures and no room for pawn moves, rendering the position a draw under the 50 move rule. The reason I had to bring up this position, because proofs of very long mating sequences only exist for these kind of few men positions in the current tablebases. However based on this simple example you can easily imagine a more complex position with many pieces and pawns, which is also a mate in 261, but there are captures and pawn moves in the sequence to reset the 50 move count, making it a legal win. Such a position would be even more hopeless for an engine to solve.

Already sortof has, it is a forced draw after 75 moves currently. (Which proves your point). So the position with win in 200+ moves can never be legally won. 

@NotAllowedTo, I think 8.Kxd7 , leading to the K+B+N vs K endgame, wins for white.

It might be solved in that maybe the software will find in any position a move that doesn't change the evaluation. But you have no proof that it is solved.

Such a computer is essentially an oracle - you never caught it wrong, but how do you know it will never be?

I have never read the Wikipedia article on solving chess. However I have looked at many other sources, in particular those concerning the physical limits of computation. If you read my posts in this thread you can see that I have come to conclusions which are almost identical to that of the Wikipedia article.

Unless there is some unexpected turn in our understanding of physics ( I mean a dramatic development, for example the possibility of hypercomputation ) or we can come up with some magic formula which can predict the outcome of complex and general chess positions with certainty ( none of the above is likely ), the solution remains a tough problem.

"a tough problem"--this year's understatement superiore