Anyone watch this match on ESPN? I saw it via the webcast and it was pretty exciting (if you call waiting 3 minutes for each move to be exciting -- but still more fast paced than baseball). What I don't understand is how the two players were compared. A lot of comparisons stated that Fritz could analyze 800,000 per second while Kasparov can only analyze 3. Well, if that's the case, then why isn't Kasparov getting his ass kicked every time? I don't like the original analogy that IBM used when Kasparov played Deep Blue. Apparently, Kasparov's brain is analyzing thousands of positions per second. He may not be consciously aware of it, but instinctively he knows that certain moves do nothing for the game. There has to be some serious computation going on in the recesses of his mind.
A while ago the "BEST" was betten by a machine a few times in a row . What is this guy trying to prove? I hope it dosen't turn out like last time with the guy freaking out and leaving "half way" thru a game because he new he was allready in a corner.:eek:
you can run through the games at the x3d site: http://www.x3dchess.com/news/analysis/kasparovx3dfritzgame3.htm
Madison, Thanks alot for the link, cool site Aphexcoil and Madison, who do u think will win the 5th game?
I predict neither because the match was only 4 games. The entire match now stands at a draw but apparently there will be a rematch. Fritz ran on a quad Xeon 2.8ghz with 4 gigs of ram.
The idea is that raw search doesn't help in certain games. For example, it is estimated that the possible number of _legal_ chess games is on the order of 10^120. That is more than all the atoms in the entire (observable) Universe. Now clearly, there are many many less games than 10^120 that are of a quality level, but even that number is enourmous. Basically, even if you had a computer that consisted of computational units the size of an atom, with each atom capable of doing supercomputer like searches for the entire age of the Universe, even with this monstrosity you could not exhaust all of these possibilities. However, computers use all kinds of tricks (alpha-beta pruning) to pivot search trees so that they do not go down a node in the search-tree that is less likely to produce a "score" less than the current score. This is where the static evaluation function comes in. The role of the static evaluation function is to assign a value to the current position. Older programs used to use very simple scoring functions, e.g., add up the values of the pieces held by both sides, which side has more "space", who controls the center, etc. After Deep Blue/Thought, that changed. IBM hired several Grandmasters to enhance the computers "understanding" of positional chess. They incorporated that knowledge into it's scoring function. Positional chess is what humans do. It is a slippery term that is used to describe what constitutues good positions and is the hallmark of the very best players. There is a huge volume of chess theory that talks about it... When Gary played DB, there was one game where the machine made a move that simply could not have been seen by a search. It was a purely positional move that showed a VERY deep understanding of the position. Gary was so taken aback by this game that he was sure that there was some cheating going on, that there must actually be humans involved in helping DB, that he demanded of the IBM team that he be given the "sheets" at the current position when the move was made by DB. It is my opinion that after this game, Gary has never recovered psychologically and seems to play "nervously" against the computer. Answering your original question is very difficult - perhaps the most difficult question in all of Science. No one really knows how the human brain works at this level. Penrose has a theory, but it is not known if it is true, and most think it is a scam. IMHO, it is the first real attempt at an "explanation" that I have ever seen... Most people think we lost something as a race and a people when the World Chess Champion lost to a chess playing computer. I instead marveled that "we" are able to stand to-to-toe and compete with something as monstrous as Deep Blue. It is as close to being Neo and doing the thing that he does in the Matrix against machines... nitro
That's a great post, Nitro. Thanks. I've heard that number before (10^120). What is interesting is that there is a very refined and large opening book, and another very large endgame database that has the permutations for any 7 pieces anywhere on the board -- but everything in between is simply too much for a database to hold. That's where the computation comes in. Now apparently at some points, X3D Fritz was going as deep as 14-15 ply. The funny thing about that is, in a closed game (pawns closing off the board), the computer simply can't see far enough to develop any sense of strategy. That is why Kasparov cleaned up easily during the third game. The computer was so stupid that it moved its Bishop and then later moved it right back where it was -- giving Kasp two free moves. Hypothetically, let's assume that a computer had a database with every conceivable game that could be played with chess. Now, would it simply be a matter of who was white, if two equally powerful computers paired off, to determine who would be the winner? Is there some proof that exists somewhere to show that if white begins and knows all possible permutations, if black could still have a chance to draw the game?
No one knows if perfect moves by both sides would lead to a draw or a win for white. I doubt anyone will come up with a mathematical proof of that hypotheses for a very long time. It was interesting that at one point in the third game, Fritz was going as high as 19 ply. It is funny, but a human could have looked at that position and known _instantly_ what the correct plan was for the black pieces (push the f-pawn and create counterplay on the king side.) Garry, knowing this, overprotected f2. This is beginning to show that humans are adapting to the way computers form plans. Garry, by overprotecting f2, steered the computation away from that plan. The evaluation function of the machine, the one that directs searches, was intentionally fooled. Garry was playing his opponent, not the position. Garry, knowing that was the only plan for black, gave the computer no clear choice. The result is that the machine spent 20 moves moving its pieces back and forth to no advantage while Garry penetrated decisively with a Queen side attack. nitro
Do you think that the computer would have played better if it were able to have searched 25-30 ply or do you think the problem would have still existed because of the way the computer was programmed to search for moves? In other words, would thinking "deeper and further" ahead have made a difference or would a more efficient "do this, if that" method would have been more effective?