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 interested in AI game programming without the insane complexities of chess, Nine Men's Morris is fun. Also a frequently researched topic in AI.
Play here.
sig
Correct me if I'm wrong - but isn't Go proven to be EXPSPACE-Complete?
Meaning that no computer anywhere is going to be decent at beating a human on a relatively large board?
I play a little chess. When I was younger I had a 1600 rating. I wanted to play because I was humilated at getting beat by the chessmaster on Nintendo. So I practiced and finally became good enough to beat the computer (albiet only a Nintendo) What I learned then (and seems to be common knowledge among chess players) is that when playing a computer, you stand a much better chance if you keep all your pawns on the board and manouver your pieces behind them. Computers think about the game in a very different manner, and I think eight pawn chess highlights where their weakness lies. They do not have a plan. They do not start the game with a long term plan to the ending. I believe that in the past, Garry was a true sportsman and did not play eight pawn chess against the strongest computers. He played real chess. He played what he would play against another Grandmaster. I really think he could probably beat the computer almost all of the time by playing eight pawn chess.
I just got back from the 15th Mid-Atlantic GO championship about 30 minutes ago where we discussed this. (Images on my site). Apparently the largest GO board that's 'solved' is 5x5. One could imagine building a quantum computer that could solve a larger board using 3 state q-bits (q-trits?) -1 = black stone, 0 = empty, 1 = white stone. Then a simple 361 'q-trit' system could represent a 19x19 board. (And be in a superposition of all states at once). Just a thought, though the quantum computing guys I work with would probably have a fit about what I just said.
http://www.glue.umd.edu/~cornick/
because they bring out so many people who bitterly complain and make excuses and want to challenge Fritz to a game of poker or something because it would give the human the advantage.
This is far from the end of our species, chill out. Even if we are worse at chess than the computers, it doesn't make the experience of being human meaningless. It doesn't mean we will be welcoming our new robot overlords any time soon.
Anyway, would it really be so bad, if AIs started getting better than humans at a lot of things? I think that in the end, we could take our greatest joy as a species in knowing that we created something better than ourselves.
Of course, that is an issue so seperated from computer chess, that many of you are probably complaining to yourselves.
That's how I feel when I read the excuse making and naysaying.
====
Crudely Drawn Games
Does Fritz learn from today's defeat... or could Kasparov repeat today's win simply by repeating today's move sequence on Tuesday?
Let me rephrase what McDermott said many years ago when people brought up the same argument. If a computer can beat us by calculating all the moves, so what?
Just because an airplane does not flap its wings to fly like a bird does, is it really not flying? On the contrary, airplanes are better fliers than birds.
AI isn't about emulating humans but about matching humans in mental capacity. How it will accomplish that is up to the researchers.
EvilCON - Made Famous by
Yeah, it's funny that when the human wins it is seen a victory for humanity. It could just as easily be seen as a loss for humanity, since the humans weren't able to build a computer that could beat a person at chess!
...or does that not get any press?
I play chess...since 3rd grade but I don't follow tournament play. Does he get more money to play the computers?
======== In the future, everything will be artificial. ========
The computer monitor alternately displays complete left-eye and right-eye images in time (temporally), each time the monitor refreshes. X3D glasses have high-speed electronic shutters (made with Liquid Crystal material), which open and close at the same refresh rate as the monitor, in sync with the page flipped images.
When the left image is on the screen, the left shutter is open and the right shutter is closed, so that the image is viewed only by your left eye. When the right image is on the screen, the right shutter is open and the left shutter is closed, so that the image is viewed only by your right eye. This happens so quickly that the brain perceives 3D depth.
Something as intense as calculations for chess really much push the brain to its "limits." The X3D technology fools the brain into seeing 3D.
I wonder how much this 50% lack of visual stimuli changes the way the chessmaster's brain works here.
I don't know if this would help or hurt the human mind trying to perform these calculations... however, I think it can be said that this is not exactly like playing a person face-to-face.
Our brains do such amazing things... I just wish I could remember where I placed my keys...
Davak
A lot of people are trying to make of this a kind of John Henry against the steam drill contest. Here is my take on it.
Some while ago someone told me that computer programmers "break things", and I never quite understood what they meant. Some while later, a "competitor" was demoing a Windows version of a type of program for which I had put a great deal of effort into a DOS version. The program had a lot of graphical and interactive displays of scientific plots and other data, and I knew enough about Windows and all the stuff you had to do (WM_SIZE, WM_PAINT) to make it look right, and I suspected my acquaintence was "first to market" by taking a lot of short cuts on his UI. He let all the scientists in the room play with his program, but he was very reluctant to let me near the thing -- because the first thing I was going to do was try and break it to find out how much work he had yet to do.
The only way Kasparov is going to beat that chess program is if he uncovers some limitation or shortcoming -- in other words to break it, and once broken I bet he could beat the thing at will. Last time around the cheat was a team of programmers hanging around trying to patch the program as soon as Kasparov latched on to such a weakness.
The chess program will have reached true AI (in a limited problem set) once Kasparov is able to find a weakness, beat if for several games straight and for the program to somehow learn from what is going on and "close the hole", and if the program can withstand other such attacks from other chess grand masters and likewise "close the hole" without going unstable (one of the problems with learning algorithms is that can overadapt and go into limit cycles). That would have far reaching implications in terms of computer security, spam prevention, 24-7 uptime, and automated bug correction -- a program capable of fixing itself would be an advance indeed.
After studying AI and Cognitive Science at University, the best definition of Intelligence that I heard was:
;)
"Only intelligent beings can make stupid decisions"
It kind of encapsulates the problem with AI really nicely; whenever you try and define it, all you are really doing is pushing the definition requirement into another area.
People have been arguing this since Plato, and IMHO have not made much headway since. If anything computer models have only confused the issue. Until this problem is solved, you cant create AI? How can you create something that you cannot adequately define?
Having said that, neural networks and emergent behaviours are cool
Invoicing, Time Tracking, Reporting
Sure, but:
The X3D technology fools the brain into seeing 3D. I wonder how much this 50% lack of visual stimuli changes the way the chessmaster's brain works here.
No effect whatsoever, because there is no "50% lack of visual stimuli". Anything that flickers fast enough is perceived by the retina itself as a solid unchanging image.
The "critical flicker fusion rate" that determines "what is fast enough" varies from about 40 frames per second to about 80 frames per second, depending on image brightness, ambient illumination, the particular individual viewing it, etc.
TV in the US, for instance, flickers at about 60 hertz, but in non-flourescent ambient illumination, most people don't notice. In Europe the rate is 50 hertz, and people frequently do notice. And some people get headaches from computer monitors that flicker even at 72 hertz, especially under flourescent lights.
I used to regularly get annoyed at PC monitors in conference rooms flickering at 60 hertz, when others didn't notice -- so I'd bring up display preferences and set it at the highest refresh rate. Until it occured to me that I was sabotaging people who needed to interface it to the overhead projector at 60 hertz. Oops! :-) But I digress.
Movies are displayed at 48 hertz (although only 24 unique frames per second; they are "double-shuttered" to double the frame rate). Cartoons sometimes have as few as 6 unique frames per second (although they are displayed at movie or tv flicker rates) because that's about the threshold for perceiving continuous motion. Lots of issues, lots of thresholds.
But even if the 3D viewing shows perceptible flicker, there isn't any issue of "50% lack of visual stimuli". Both eyes are constantly receiving information.
I could imagine that any number of things about this 3D gadget could distract a chess player -- but so does cigar smoke (a trick used to advantage in chess matches early in the 20th century).
Years ago I used to use similar 3D goggles to play first person shooters like Quake, and it was great. It helped my game. Quake isn't chess, but 3D goggles aren't rocket science.
You can assume that, if he agreed to use this 3D setup, he was confident it wouldn't throw off his game. He does care, after all.
Professional Wild-Eyed Visionary
I wonder how much this 50% lack of visual stimuli changes the way the chessmaster's brain works here.
Besides which, why don't they just have a real chess board with a guy sitting there moving the pieces as the computer directs? Maybe I'm missing something, but it seems like that would be more like playing a person face-to-face.
A computer will be intelligent when it is capable of learning things which are entirely new, by applying its knowledge, trying things, general reasoning, and developing new representations. Chess is fine, go is fine, but I'll be impressed when a chess program goes on the internet and learns to play go.
There have been a series of problems that people have posed with the idea that they couldn't be solved without giving the program the ability to figure out the problem. Each time, however, people figure out how to solve the problem without giving the computer general intelligence. But the problems weren't definitions of intelligence ("Intelligence is what is necessary to play chess"), but rather people thought (incorrectly) that the problems couldn't be solved without intelligence.
On the other hand, there is a cluster of problems which have not been solved, despite many attempts, which are not intended to be difficult (which can be overcome eventually with clever coding and fast hardware), but rather which explicitly require undirected learning and the ability to generalize to new applications.
Of course, many of the problems originally posed as measures of intelligence were, in fact, problems that people wanted solutions to, because they were difficult for people. So the effort put into solving these problems without trying to use general intelligence is actually worthwhile.
The position after 29. a6 was indicative of how paranoid Kasparov was about the computer's tactical capabilities. In addition to the pawn blockade stretching diagonally from f2 to b6, he had marched his king all the way from e1 to b2 and protected it behind a wall of pieces. The king's bunker looked like this:
BN N
K R Q
As chess positions go, that one cracked me up.
The trouble with practical jokes is that very often they get elected. -- Will Rogers
On one hand, a victory for the computer means a victory for everything we've been working at for a long time. It means that computers are getting smarter, and smarter, and smarter.
Call me a hypocrite, call me sentimental, but I desperately want Kasparov to win. I want us to still be better than computers at this game. It's highly mathematical, but there's always been a level of flare, panache, and style to the game. Even though 'Knight to King 4' may not sound particularly interesting, it could have been something intrinsically bold and audacious when done by a human player. When the same move is made by a computer it becomes purely calculated.
I want Kasparov to win because I feel like it'd be a blow to the game to let an algorithm (albeit a brilliant one with an unbelievable amount of brute force behind it) beat something feeling.
while (!sleep){
sheep++;
}
Does Kasparov play human beings anymore? or is he too good for us?
Ponomariov held out for more money (non-existant) in a sheduled match with Kasparov that would have led to a championship match between either Kramnik or Leko. Neither match ever happened so Kasparov headed back to New York for another payday with Fritz. The problem is not Kasparov playing other humans but other humans having the guts to play Kasparov. Kramnik has not defended his title in 3 years. The FIDE stripped Fischer's title after that long.
an ill wind that blows no good
How can you create something that you cannot adequately define?
Evolve it.
grib
maybe
I have to say that I still think that humans and computers use essentially the same algorithm. Even the very best chess players search and prune a tree of moves.
Of course, grandmasters don't think of this as searching and pruning a tree, but that's what they are doing subconciously. If they didn't, they'd get obliterated in tactical exchanges.
The main difference between the human and computer tree searches is simply that the computers are far better (faster) at searching while the humans are far better at pruning and evaluation (thanks to the pattern matching you mention).
Humans simply ignore (prune) the vast majority of possible moves, so their "search tree" has a very low branching factor. I wouldn't be surprised if the average was less than 2 for top players, so they can look very deeply into a position without considering many alternative lines.
The other advantage that humans have is that they are much better at considering "strategic" aspects when evaluationg a position. It's very hard to get computers to recognise these. But in effect, this just means that the human has a much better evaluation function, so this is still consistent with the fundamental tree search approach. Like the parent post says, this ability doubtless involves considering groups of pieces using pattern matching etc.
The other interesting thing that humans do is remember aspects of the position which they subconciously make use of while doing move search and position evaluation, e.g. the existence of a pin or potential fork threat. My own style of play uses this quite a bit - I like to look for threats and weaknesses then try to find lines that will help me exploit them.
In this way, humans seem to be doing a kind of "directed search" which is theoretically possible in a computer's tree search but I've never seen actually implemented in computer chess. Anyone seen anything like this in a tree search algorithm? I think there could be potential for research in this area.