Slashdot Mirror
Kasparov Wins Game 3 Against X3D Fritz
Vulcao writes "Garry Kasparov just brilliantly won game 3 in the Kasparov vs. X3D Fritz chess match, which pits man against machine. Kasparov created a positional advantage on the queen side with a very strong pawn structure to which Fritz didn't have an answer. The result is now 1.5 - 1.5, and the last game will be this Tuesday, Nov. 18."
For those who receive ESPN/ESPN2, the sports network has televised all three matches and will televise the fourth on Tuesday at 1:00pm. I've watched all three games on there, and it's actually very entertaining, if only for the humor of seeing history's greatest chess player in action and wearing those stupid X3D goggles. I just hope Garry can pull off Game 4 with a win.
porp
This gives a nice introduction to Go and AI and why it is so hard to play for a computer.
To quote (from memory) the online commentator Mig Greengard:
"If X3D Fritz lacks a clear target it plays like a braindamaged lemur"
As Fritz moved its pieces back and forth throughout the game, Kasparov could make several free moves. That isn't brilliant, that's just making use of the other guys mistakes. Kasparov dominated the whole game, while Fritz had no clue at all what to do. According to one of its makers, X3D Fritz reached a new record of reading deeply (19 ply if I'm not mistaken) since the number of possible moves was so small in the cramped space they were building up their positions. This, however, didn't help a bit and I had a few giggles over bishops and knights moving away and then back again to the very same place they were coming from.
Only at the very end did Fritz realize it was losing, throughout the whole game it couldn't see what was glaringly obvious to the audience.
I've been told that this was proper anti-computer chess. The cramped position makes it tremendously difficult for a computer program to play properly while a human can easily see what's to be done.
All in all, it wasn't brilliant, Fritz just didn't have a clue
What am I discussing all this chess for? Let me get back to KGS...
"We live in our minds, and existance is the attempt to bring that life into physical reality" Ayn Rand
Its not so much a matter of computers doing things one way and humans doing things another way, but a matter of both being bound to the rules of computational theory. There is a large solution space, and both humans and computers have to decide what parts of the solution space to search and what parts to ignore. They have to do this, even if they do not do this in the same way. That's all that pruning the search space really means.
Now, just because you don't do it conciously doesn't mean you don't do it. Your brain does an incredible amount of processing behind your back. Think about visual processing or auditory processing. All of that goes on completely outside of your concious thought.
A deep unwavering belief is a sure sign you're missing something...
When humans play they rely primarily on pattern matching rather than searching a tree (unless they suck). Computers tend to be very poor at pattern matching, and humans tend to be extremely good-that's why a small child can look at photographs and categorize them instantly, but the most advanced computers have great difficulty with that kind of task.
Most skilled human chess players apply pattern matching by looking at the board and identifying interesting things. They start with nothing and add new options to the list until they feel they have a sufficiently comprehensive understanding of the situation.
Search tree pruning, by contrast, starts by including the entire space of potential moves and identifying courses of action that can be eliminated. Alpha-beta pruning is a particularly poor example here since it has the useful property that all a-b pruned subtrees are guaranteed not to be optimal. However, humans often ignore superior courses of action and choose suboptimal ones that "feel right" or match prior experience.
There have been various experiments on the limits of raw pattern matching ability with chess pieces. An interesting one involved asking participants to memorize an arrangement of pieces and reconstruct it a minute after the arrangement is removed. Participants included people with little or no chess experience and masters.
Those without experience memorized it as raw data, and did as well as they would have if asked to memorize random numbers instead of chess arrangements. The masters were more interesting. They did about the same as the beginners on random arrangements that could never actually happen in a game, but they were infinitely better at reconstructing realistic arrangements that often show up in games. The experiments proved that masters can recognize groups of pieces and evaluate them collectively.
In a game situation this means the master looks at the board, and certain parts of it just stand out. The master will pay no attention to areas that don't grab his attention, and doesn't need to evaluate whether any of those individual pieces are worth moving. Interestingly, this means that playing with nonstandard rules (such as changing piece movements) will likely devastate a master's ability while only slightly reducing an amateur's skill level and leaving the computer's ability unchanged.
Even though I think the parent is a troll, here's an academic article detailing some other experiments on the topic.