Originally posted by vivify
Something I've been thinking about more and more as I play, is ,that chess is all just mathematics. There exists a mathematically perfect counter to every move; it's just a matter of time before a computer figures it all out. kinda makes it a little disapointing to me.
Chess has the illustion that it takes brilliance and strategy to win at; but in reali uit whining. But I'm sure even this, is something that no human can beat a computer at.
I feel there is some truth in what you are saying, but there's more to it than just calculation (i go here, he goes there, etc...)
Computers themselves are starting to play differently. Take this game i posted the other day...
(straight cut and paste)
Black (Houdini chess engine) Is evaluating the activity
of the pieces. It calculates that it can get away with sacrificing all this material, because the activity of it's pieces means it can win back the material. But it plays on, down in material, increasing it's activity and hindering the activity of it's opponent (Rybka, 4 times Word champion at the time). Rather than calculating every
move available, it selects a few core variations (after a short evaluation of it's options) and puts all it's processing power into evaluating these specific lines to a much higher depth. This allows it to avoid horizon issues that most engines suffer from, because it is calculating lines sometimes up to 40 ply (20 moves per side). What's interesting, is that when a computers horizon is increased like this, the style it plays in is quite different. When Tal made these seemingly illogical sacrifices, he was evaluating piece activity. His games are spectacular because the outcome looks unclear. Houdini is capable of finding such moves, because it can actually calculate the outcome.
When Kasparov played Deep blue (which there is no current version of btw, it was dismantled straight after the match, to great controversy) he was playing a purely calculating machine. Traditionally chess computers evaluate a position purely in tactical grounds, it seeks only to win material or deliver mate (which is why spotting them on correspondence sites is quite easy, in quiet or purely positional positions engines tend to make strange manoeuvres that are hard to explain without introducing non-chess words like psychology..)
Now where am i going with all this? Well, the thing is this. When Kasparov played Deep blue (or Adams against Hydra) it spelled the end of humans beating engines. Fare enough. But are computers moving closer to solving the game? I think this game above points towards a resounding no! On the one hand you have an engine (white) which is a brute forces material hungry monster. White losses because it seeks a material advantage
Let's just stop here and think about this, because this is quite profound.
You can develop new engines that play in different styles. Nothing is invincible, unless it has every possible position pre-evaluated to a win before the match. This, to me, is realistically impossible.
The only way chess will be solved, in my opinion, is when end game table bases work their way right back to the starting position. But when you really think about how much information that represents, the possibility that any human could memorise enough to become unbeatable is just absurd.
What does this mean for human chess?
Well, i think Houdini plays quite like Tal/Kasparov in this game. Human players are capable of making these huge leaps of imagination that allow them to sacrifice material like this. They only need to see certain key moves and can develop outrageous plans as a result. That is not mathematics to me. Yes, when they calculate a line to make sure it works, this could be compared to maths. But the reasoning that goes into what they should calculate
is much more akin to fuzzy logic. "If my bishop was on f7, i'd be able to do 'x' 'y' 'z', can i get my bishop there? ...calculate... yes."
Computers need to use a similar approach. Houdini gets it's strength because of it's initial evaluation. It has to choose which lines look promising, then it calculates. What is a promising line? How many promising lines does it miss as a result of an earlier shallow evaluation? There is so much room for improvement, perhaps in time computers will discover a completely different style of play altogether?? What about Petrosians style, do we have an engine that plays like him? If someone developed such an engine (which put removing risk from a position as its first priority, for example) would that counter the houdini style? The same arms race that takes place between humans will inevitably take place between engines. This is really just the beginning, i'm not sure there will be an end..