integer

December 5, 2006 4:20 PM

Machine beaths Man. Deep Fritz (.pdf) has beaten world chess champion Vladimir Kramnikin in Bonn.
posted by four panels (68 comments total)

Thay it isn't tho!
posted by Astro Zombie at 4:21 PM on December 5, 2006 [4 favorites]

In 2002, Mr Kramnik held Deep Fritz to a draw after eight games, but the chess software has since been updated, calculating millions of positions every second.

Meh. Keep on calculating, you computers. This is just one of the many, many things human brains can do and its not even something they were designed to do!
posted by vacapinta at 4:27 PM on December 5, 2006

Wait... this computer, it plays chess? Preposterous!
posted by Ricky_gr10 at 4:33 PM on December 5, 2006

thith ith thocking!
posted by mrnutty at 4:34 PM on December 5, 2006

Yes, but Mr. Kramnikin can now go out and get drunk and fuck a hooker. Beat us at that, you metal bastards!
posted by bob sarabia at 4:35 PM on December 5, 2006 [4 favorites]

Shhh! Don't give them any ideas!
posted by brundlefly at 4:36 PM on December 5, 2006

In related news: Chess champion beats computer, programmer with hammer.
posted by papakwanz at 4:38 PM on December 5, 2006

Wow, the human didn't win any games at all. Ouch.
posted by Malor at 4:42 PM on December 5, 2006

in the photo of Kasparov vs deep blue in the pdf link, why is deep blue's flag a Nazi flag?
posted by Dr. Twist at 4:43 PM on December 5, 2006

That's really weird. Looks like it's been altered. Here's the original.
posted by bob sarabia at 4:50 PM on December 5, 2006

Wait... this computer, it plays chess? Preposterous!

For those not inclined to dig through the pdf:

In the past, building a personal computer equivalent to Deep Blue was not a realistic
goal. IBM had spent millions on Deep Blue (the cost of the Deep Blue project from 1985
to 1997 is estimated to have been over $100 million), which was a massively-parallel
RS/6000 SP based computer with 32 processors that could evaluate 200 million chess
positions per second. Setting aside the multi-million dollar price tag, Deep Blue
consisted of a pair of 6 foot 5 inch black towers weighing 1.4 tons.

Deep Fritz is neat because it runs on a PC.
posted by juv3nal at 4:56 PM on December 5, 2006

So it wins by brute force calculation of moves? Am I reading that right? The compy don't think creatively? Or is that how chessmasters operate: they think of every possible move and select one? What's strategery is involved on the human and machine sides? I read Brute Force is superior algorithm wise than the as yet unknown algorithms that go on in the human mind.

I will content myself with the fact that robotics has not produced a humaniform robot yet. Merge one with a chess playing brain and the world of chess will be in an uproar!
posted by Mister Cheese at 4:58 PM on December 5, 2006

Well, there's still a long way to go with computer Go.
posted by MetaMonkey at 5:04 PM on December 5, 2006

The nazi flag thing is bizzare.
posted by phrontist at 5:15 PM on December 5, 2006

Mr. Cheese: Or is that how chessmasters operate: they think of every possible move and select one?

The great Capablanca is supposed to have said, "I see only one move ahead, but it is always the right one." Probably been linked here before but this Scientific American article on cognitive research into how chess grandmasters think is an interesting read.
posted by otio at 5:17 PM on December 5, 2006

Chess? Whatever? Can it beat a any dan level go players? Now THAT would be something.
posted by redbeard at 5:30 PM on December 5, 2006

That nazi flag thing is disturbing. Obviously someone is still trying to give IBM grief about their nazi links .
posted by wilful at 5:35 PM on December 5, 2006

After the game, Mr Kramnik said he was "a bit disappointed" but hoped a rematch could be arranged in a year or two.

"With more time to prepare, I still have a chance," he said.

Uh, no, sorry Vlad. No human -- you included -- will ever have as good a chance as you just did ever again.
posted by Flunkie at 5:37 PM on December 5, 2006 [1 favorite]

boy howdy, I knew my chess software was good, but I had no idea it was that good.
posted by carsonb at 5:39 PM on December 5, 2006

Call me when a computer, armed only with a knife, (say, six or eight inches long) can be programmed to consistently "win" fights with a grizzly bear. Assume no element of surprise.
posted by It's Raining Florence Henderson at 5:42 PM on December 5, 2006

Chess? Whatever? Can it beat a any dan level go players? Now THAT would be something.

That's right, humans, keep moving the goalposts if you want to keep feeling special.

Twenty years ago, this was unthinkable.

Ten years ago, it happened - and a large portion of the world reacted by making excuses for why Kasparov was the better player even though he lost.

Now, it's "Well duh, whatever, booooooring".
posted by Flunkie at 5:44 PM on December 5, 2006 [1 favorite]

You may not be aware that a computer has, in fact, been running MeFi for the last eighteen months while Matt's been sitting back and watching the money roll in. Yup, that's right, the snarky deletion reasons, the occasional posts, the MeTa comments—all computer-generated. The winners of the December contest will be selected by the same algorithm used to choose a President... but I've said too much.
posted by languagehat at 5:53 PM on December 5, 2006

That machines getting a polonium enema...
posted by Artw at 5:54 PM on December 5, 2006

GIS for kasparov+deep.blue doesn't turn up the Nazi version at all, let alone as a common photoshop that could have been mistaken for the original. It may be a subtle swipe at IBM dealings with Nazi Germany.

So it wins by brute force calculation of moves? Am I reading that right?

That's the usual way computers operate. Lots of AI research into neural nets and other rules-based logic systems has found very limited applications. We're past 2001, but we don't have HAL.

redbeard: What's remarkable here is the replication of a $100 million mainframe system with a couple of PCs, not the playing of chess specifically. It should be obvious that Go is combinatorially more complex, so that isn't exactly a criticism.
posted by dhartung at 5:56 PM on December 5, 2006

Spelling beaths four panels.

This is the pile-on thread, isn't it?
posted by mr_crash_davis at 6:03 PM on December 5, 2006

Actually, Deep Fritz (the computer that just beat Kramnik) is significantly more heuristic-based (and less brute force-based) than Deep Blue (the computer that beat Kasparov ten years or so ago).

Blue was running on a supercomputer.

Fritz was running on a PC (a pretty nice PC, but a regular old PC nonetheless).

Blue, ten years ago, evaluated twenty-five times more moves per second than Fritz just whupped Kramnik with.
posted by Flunkie at 6:04 PM on December 5, 2006 [1 favorite]

Call me when a computer, armed only with a knife, (say, six or eight inches long) can be programmed to consistently "win" fights with a grizzly bear.

The computer doesn't need to beat the bear. It just needs to outrun the human.
posted by eriko at 6:14 PM on December 5, 2006 [2 favorites]

In college, we invented a form of drunken chess. You set up and play like normal, but each player may choose, at one point in the game, to flick a pawn toward the other side with the forefinger. Any pieces that roll off the board are taken. Any pieces that fall but remain on the board are set back up on the square where their base comes to rest. The exceptions are the king and queen, which are either stood up where they fall or are put back in their original positions if they roll off the board.

Someday, Deep Fritz, wasted on too many margaritas, will extend his metallic forefinger-shaped digit and flick a pawn into outer space.
posted by I Am Not a Lobster at 6:15 PM on December 5, 2006

Languagehat: Ahem...
posted by Steven C. Den Beste at 6:17 PM on December 5, 2006

Nah, the computer didn't whup Kramnik at all. Kramnik had the better chances in many games, but it was never enough to win. The computer just defends too well. Not only that, but the game of chess has been solved from the final mating positions, backwards to 6 pieces. The computer plays the end of the game without even calculating; it plays by lookup tables. Nor does it play the opening. There's still a lot of room for AI improvement for those who are interested, seeing as the computers aren't even playing the ending or openings anymore.

And even then, the computer doesn't necessarily play better chess, but it certainly plays tougher, more consistant chess.

Kramnik lost one game by blundering a mate in one. Compare that to Kasparov resigning against Deep Blue in a drawn position. Kramnik lost the second game in the final game, when he was in a must-win scenario and had to risk it.

Most of the games between these two will end up being draws. Kramnik will lose occasionally because he's human and cannot play his best all the time.

And in any case, if you want to see what a supercomputer can do now, and want to see a real machine whup a GM, take a look at the Hydra-Adams match from this last year. 5.5-0.5 in favor of the computer. (Though Adams is an attacking player and ill suited for playing a computer, it was still an awful showing)
posted by cotterpin at 6:25 PM on December 5, 2006

I wish my computer would beath me every now and then.
posted by Schlimmbesserung at 6:29 PM on December 5, 2006

Nah, the computer didn't whup Kramnik at all. Kramnik had the better chances in many games, but it was never enough to win. The computer just defends too well. Not only that, but the game of chess has been solved from the final mating positions, backwards to 6 pieces. The computer plays the end of the game without even calculating; it plays by lookup tables. Nor does it play the opening. There's still a lot of room for AI improvement for those who are interested, seeing as the computers aren't even playing the ending or openings anymore.

All the excuses in the world will not change the fact that only one of the two players won any games at all, and that player was not Kramnik.

And even then, the computer doesn't necessarily play better chess

Riiiiiiiight... it doesn't play better chess. It merely plays the kind of chess that wins more games.
posted by Flunkie at 6:32 PM on December 5, 2006

That's exactly what I was thinking, Flunkie.

Winning more games == playing better chess.
posted by teece at 6:46 PM on December 5, 2006

There's nothing wrong with re-defining the rules so that we still "win." In fact, its exactly what makes us superior to computers.
posted by vacapinta at 6:53 PM on December 5, 2006

Riiiiiiiight... it doesn't play better chess. It merely plays the kind of chess that wins more games.

I don't agree with the idea that the player who plays better chess always wins. It's not true in other games. The better playing football team is not always the one who leads on the scoreboard at the end of the match. Often, the winner is not the player who plays better, but the player who errs less.

Now, I know you can say that "better" play is the play with fewer errors, but that just isn't what most people think of when they think "better". Quality != consistancy. The player who shows a deeper understanding, plays with more purpose, and controls the game most of the time can be said to play better even if he doesn't win. To say otherwise would be like saying that the most intelligent orator is the one who makes the fewest grammatical mistakes. It's not so and the reason is that humans understand and evaluate ideas.

The computers are relying on brute force more than ideas on the board. And the endgame databases, ugh ... a 30GB external database is "better" chess? If this is what makes the computer "better", then maybe it's time to insist that the computer actually compute the moves it plays.
posted by cotterpin at 7:00 PM on December 5, 2006

Pfft.... Fritz doesn't even have arms... I could take 'im.

Let's go, Fritz. You. Me. 3:00. The schoolyard. BRING IT.
posted by Greg Nog at 7:06 PM on December 5, 2006

There's nothing wrong with re-defining the rules so that we still "win." In fact, its exactly what makes us superior to computers.

Kirk: What's on your mind, Lieutenant?
Saavik: The Kobayashi Maru, sir.
Kirk: Are you asking me if we're playing out that scenario now?
Saavik: On the test, sir. Will you tell me what you did? I would really like to know.
McCoy: Lieutenant, you are looking at the only Starfleet cadet who ever beat the no-win scenario.
Saavik: How?
Kirk: I reprogrammed the simulation so it was possible to rescue the ship.
Saavik: What?
David Marcus: He cheated.
Kirk: I changed the conditions of the test. I got a commendation for original thinking. I don't like to lose.
Saavik: Then you never faced that situation. Faced death.
Kirk: I don't believe in the no-win scenario.
posted by papakwanz at 7:09 PM on December 5, 2006

But can Deep Fritz beat a human at enjoying the game? If you play a game and don't enjoy it then there is no point in playing.
posted by Vindaloo at 7:36 PM on December 5, 2006

Now, I know you can say that "better" play is the play with fewer errors, but that just isn't what most people think of when they think "better".

That's a strawman. I don't say that the "better" play is the play with fewer errors. I say that the "better" play is the play that wins more often.

And if people want to pretend to disagree with that, well, um, allrightee, fine with me.
posted by Flunkie at 7:44 PM on December 5, 2006

If you play a game and don't enjoy it then there is no point in playing.

I bet there are a significant number of grand masters who enjoy it less than I do. Sure, they all obsess on it more, but that's not even remotely the same thing.

So, I propose that, in the spirit of your new explanation of why Fritz beating Kramnik is irrelevant, that I should be the new undisputed world chess champion.
posted by Flunkie at 7:47 PM on December 5, 2006

cotterpin:
And the endgame databases, ugh ... a 30GB external database is "better" chess? If this is what makes the computer "better", then maybe it's time to insist that the computer actually compute the moves it plays.

What's good for the goose is good for the gander - a massive database is exactly how human chess masters do it. In fact chess is often described as a game played from memory.

If you want to limit the methods to "compute the moves you play", the computers would have wiped the floor with the humans long ago - humans playing by memory instead of by computation has given humans a huge advantage over computer players.
posted by -harlequin- at 7:50 PM on December 5, 2006

That's a strawman. I don't say that the "better" play is the play with fewer errors. I say that the "better" play is the play that wins more often.

It's all qualitative, but I think I can present a clearer argument: better play, in this human-centric sense, is play that advances the meta-state of the game. For chess, a game of perfect information at a gamestate-by-gamestate level, better play is inventive play, counter-intuitive, outside-the-box, guts play that escapes the traps of local minima and maxima that define the rote brute-force method. Better play is play that teaches us something new about chess.

In that sense, talented humans play better chess than computers; and, from what folks have said in the thread, Fritz plays better chess than Deep Blue—more heuristic, less brute-force for the same results. Working smarter, not harder.

By the same token, consider game theory or economic theory—is the better economic model one which "wins" more individual trials, or one which produces stabler, healthier results over the long term? If the best theory is the one that wins the player the game most consistently while, as a result, destroying the economy in the long run, what, exactly, makes it better?

A win is a hard thing to define, because scope matters.
posted by cortex at 8:00 PM on December 5, 2006

That's right, humans, keep moving the goalposts if you want to keep feeling special.

You can see this as humans not wanting to be special, but about exploring what is indeed special about the way a human mind can selectively process a situation.

Right now it's hard to imagine a computer that could beat a dan ranked player in Go; they can't wastefully calculate their way to a victory and, sci-fi quantum computing aside, I'm not sure a brute force machine is physically possible (feel free to correct me with any facts I'm sorely lacking).

To have a computer beat a human at Go is a drive to make computer intelligence as special as possible.
posted by Chuckly at 8:05 PM on December 5, 2006

The nazi flag thing is very strange, and I think it's more of a swipe at the US, rather than IBM, which is ironic considering that the nation of the other flag, Russia, was for many decades far more like Nazi Germany than the United States ever was.

And it's a silly, stupid, childish thing to put it in a paper on computer science.
posted by Pastabagel at 8:15 PM on December 5, 2006

Chuckly:
Go is as brute-forceable as chess, which is to say, not very much with current technology. These computer chess players do not do it just through brute force - that just doesn't work well enough with current computing power to win games at this level. Brute force is their central pillar because calculations-per-second is their strongest point, it's their forte, and they put a lot of weight on that because they have to, but on it's own is inadequate for the task.

If the same resources and creativity that went into creating these chess machines, were put into a Go machine, I don't imagine the results would be any different. Is there any reason to think otherwise? These things are not simple algorithms and optimising heuristics, they are repositories of human strategies, libraries of the experiments of masters, and so on.
posted by -harlequin- at 8:23 PM on December 5, 2006

That's a strawman. I don't say that the "better" play is the play with fewer errors. I say that the "better" play is the play that wins more often.

It's all qualitative

No, actually, wins, losses, and draws are quantitative.
posted by Flunkie at 8:26 PM on December 5, 2006

(ie, Go may be tricky in software, trickier than chess, but I think the lack of Deep-blue resources applied to the problem is probably the biggest reason lower high level Go players remain able to beat computers. If I'm wrong about the time, money, and expertise invested in computer go, vs chess, then I'll happily withdraw the assertion :)
posted by -harlequin- at 8:29 PM on December 5, 2006

No, actually, wins, losses, and draws are quantitative.

Ye christ. And lord knows, win/loss records are the only meaningful aspect of AI research, games research, and the pursuit of understanding of human intelligence as a whole.
posted by cortex at 9:10 PM on December 5, 2006

Out of curiousity--will Deep Fritz get better at chess as it plays more games? Does it learn what tricks its opponents use?
posted by Citizen Premier at 9:17 PM on December 5, 2006

If the same resources and creativity that went into creating these chess machines, were put into a Go machine, I don't imagine the results would be any different. Is there any reason to think otherwise? These things are not simple algorithms and optimising heuristics, they are repositories of human strategies, libraries of the experiments of masters, and so on.

The reason to think otherwise is the sheer branching factors present in a Go game. I hear what you're saying—the measure of true success is not the high-water capacity for brute force speed but the elegance and efficiency of reductive heuristics—but I still doubt (as an armchair observer, granted) the easy reduction of Go-level open-endedness to the Chess-level.

I can't provide you with any rebutting citations regarding Go funding, but I expect it has been, if not comparable to Chess funding, still very much non-trivial over the last twenty years. And, notably, the development of heuristic insights, as opposed to bleeding-edge computational brute force achievements, can be done by lone, poorly funded genuises.

And actually, I hope your suspicion is correct—I think major breakthrougs in computation Go strategy would be a fantastic discovery. The challenges seem different from those of chess—it seems like the keys would lie in creating a scalable set of models of play, from the very local tactical to the larger board-wide strategy, spacial in focus, that differ signficantly from the more constrained and temporal combinatorics of chess.
posted by cortex at 9:20 PM on December 5, 2006

But can it beat an extremely lucky idiot?
posted by tehloki at 9:53 PM on December 5, 2006

The Wikipedia article about computer Go linked above contains an excellent discussion of why it is that Go is not readily susceptible to the kinds of approaches which have been used so successfully for Chess.

The branching problem is part of it, but not all. Fact is, they can't even write a program now that can score a completed game between two humans. Computer programs can't reliably determine which groups are alive and which are dead.

Computer programs often do pretty well at Go tactics, but they don't handle Go strategy well. Go tactics are computationally similar to the kind of thing that's done in Chess, but Chess has no equivalent of Go strategy.

I think the biggest difference between the two games, in terms of affecting the computing problem, is the victory condition. In Chess your entire goal is to get the enemy king and prevent the enemy from getting your king. In Go, your goal is "more territory". But that's a lot more nebulous, and it doesn't matter where your territory is, as long as you have more of it.

People who have not played Go heavily can't really understand how fundamentally different it is from Chess, and how much more challenging it is for computer algorithms. Even coming up with an evaluation algorithm to compare two board positions and decide which one is preferable is nontrivial.

[When I was at my best I was 9 kyu. I had a chance to play a 4 Dan one time. He gave me 9 stones, and creamed me.]
posted by Steven C. Den Beste at 10:02 PM on December 5, 2006

If you play a game and don't enjoy it then there is no point in playing
I'm sure Fritz would rather be searching for Sarah Connor.
posted by Joeforking at 1:48 AM on December 6, 2006

It's all qualitative, but I think I can present a clearer argument: better play, in this human-centric sense, is play that advances the meta-state of the game.... Better play is play that teaches us something new about chess.

No, what you're describing is not chess. Some kind of meta-chess, if you will, or the study of chess, or "chessology," but not chess. Perhaps one can "win" at chessology by "advancing the meta-state of the game" and learning something new about chess, but that may or may not actually win at chess.

We do not confuse geology--the study of the earth--with the earth itself, and likewise we should not confuse the study of chess with chess itself. As a proof of concept, it is possible to advance human knowledge of chess on one's own, without an opponent. Playing chess, by definition, requires an opponent. The two are not the same.

A win is a hard thing to define, because scope matters.

A win may be hard to define in chessology. It's well-defined in chess.

And lord knows, win/loss records are the only meaningful aspect of AI research, games research, and the pursuit of understanding of human intelligence as a whole.

Nice try at a strawman, but I don't see anyone here actually saying that. What people are saying is that win/draw/loss records are the only meaningful measure of chess skill. AI research, no. Games research, no. Understanding human intelligence, no. Skill at chess, yes.
posted by DevilsAdvocate at 5:02 AM on December 6, 2006

Ye christ. And lord knows, win/loss records are the only meaningful aspect of AI research, games research, and the pursuit of understanding of human intelligence as a whole.

After you tell me exactly which part of any of my posts led you to believe that I was discussing whether or not there was anything meaningful left to research in AI, games, or human intelligence, we can continue this conversation.
posted by Flunkie at 5:07 AM on December 6, 2006

I don't agree with the idea that the player who plays better chess always wins.

Always, no. Usually, yes. Sure, on any given day, blah blah blah, but if you play a series of games and lose them all, you weren't playing better than your opponent, and your opponent is very likely simply better at it than you are. Now, if you want to talk about advancing the meta-whatever, that's all well and good, but the original question was whether the computer is better at chess, and at this point the answer is a pretty unequivocal yes. ("Sure, the Cougars have kicked our asses every time we've played them, coach, but we have higher-level inspirations while our asses are getting kicked! We may not be able to score, but our hearts are pure and our thoughts are lofty!")
posted by languagehat at 5:55 AM on December 6, 2006

What's good for the goose is good for the gander - a massive database is exactly how human chess masters do it. In fact chess is often described as a game played from memory.

This is not true. Masters do not remember individual positions and individual moves that go with each position.

Masters remember classes of positions and the ideas that go with them, but over the board they must decide which class of positions are applicable and how to use their knowledge. This is applying past experience; this is human intelligence.

Machines are using a database of billions of positions, each position individually, and stored with each position is a single move pointing to some other position in the database. This memory is nothing at all like how humans play.
posted by cotterpin at 6:00 AM on December 6, 2006

People who have not played Go heavily can't really understand how fundamentally different it is from Chess, and how much more challenging it is for computer algorithms.

And humans. Part of the reason the Chess problem was preeminent is that it was a far more tractable problem than go.

The other reason? Well, the gang at SAIL and MIT-AI didn't know about go -- it only started to appear in the US in the late sixties, and by the time it became well known, the Chess problem was well established as "very hard, indeed", and even glancing at the Go problem showed you just how much harder it was. (It wasn't until 2000 that a American, Michael Redmond, achieved 9 Dan professionally, the highest rank in Go.

Another issue -- Go is a huge pattern recognition problem, which is a hard issue for computers.

I'd say if we get a computer to 1 Dan (the pro version, aka 1 Ping) , the first of the pro ranks) then we're either talking to a sentient AI or won't be able to tell the difference.
posted by eriko at 6:04 AM on December 6, 2006

>>And lord knows, win/loss records are the only meaningful aspect of AI research, games research, and the pursuit of understanding of human intelligence as a whole.

Nice try at a strawman, but I don't see anyone here actually saying that. What people are saying is that win/draw/loss records are the only meaningful measure of chess skill. AI research, no. Games research, no. Understanding human intelligence, no. Skill at chess, yes.

But we're discussing this in the context of what can only be seen as "chessology". This conversation is centered around a recent development in chess research.

I will not dispute it: the clearest measure of per-game success in a series of chess game is the win/loss/draw record. Okay? Is it clear that I grasp that little tidbit? Yes. A program or person who consistently wins more games is, in terms of winningness and therefore in the most obvious sense the better player. Great.

But where the hell did Fritz come from? Why do we distinguish it from Deep Blue, or Chessmaster 3000, or pulling moves out of a hat? Why do we describe human grandmasters in any other terms than "better" and "worse"? Why doesn't chess criticism and analysis just plot ever notable player on a single 1-D continuum of talent and simply leave off?

Because there are greater qualitative aspects of how the game is played that we associate inescapably with the game. Chessology, as you have it, is the only thing that makes chess interesting.

I'm not trying to present a strawman—I'm trying to state, in as obvious terms as I can, why folks would make the apparently contentious statement that "better" might not reduced simply to "wins more". You were the respondent in this:

>>Now, I know you can say that "better" play is the play with fewer errors, but that just isn't what most people think of when they think "better".

That's a strawman. I don't say that the "better" play is the play with fewer errors. I say that the "better" play is the play that wins more often.

And if people want to pretend to disagree with that, well, um, allrightee, fine with me.

I was responding to that. I'm neither pretending to disagree with your cut-and-dry argument, nor pretending not to see how limited a view of what we can mean by "better" that is. I think it's an interesting question, and I think a repeated "win = better, nuff said!" is missing the point. You're welcome to say that a more complex sense of "better" doesn't apply to chess, only to chessology, and I'll grant that in the strictest, narrowest, dullest sense you're right.
posted by cortex at 6:23 AM on December 6, 2006

As we've moved away from Chess, to Go...
In a previous life I worked with Mark Boon, writer of the computer Go program "Goliath". From what I heard, it was pretty good for its time, at both playing the game, and helping him score cute Japanese women ;)
posted by lowlife at 6:32 AM on December 6, 2006

-harlequin- writes Go is as brute-forceable as chess, which is to say, not very much with current technology

I'm not sure I'd agree with that... a chess board has an average branching factor of 35 or so, and a really good program would probably need to look 7 or 8 moves ahead to be playing at a high level. That's about 65,000,000,000 moves to calculate, which sounds enormous until you figure that Deep Blue could do 200,000,000 a second--5 minutes to calculate a move, if you explore the whole tree with a naive algorithm. And those numbers have scaled with processing power gains. Go, on the other hand, seems to have a branching factor at least an order of magnitude larger (I haven't looked into it deeply, but ye gods, a given board position has a lot of possible moves), which makes it completely intractable for a brute-force approach.
posted by Mayor West at 7:19 AM on December 6, 2006

You're welcome to say that a more complex sense of "better" doesn't apply to chess, only to chessology, and I'll grant that in the strictest, narrowest, dullest sense you're right.

So you're saying I am technically correct--the best kind of correct.
posted by DevilsAdvocate at 7:26 AM on December 6, 2006

Heh. In the strictest, narrowest, dullest sense. :)
posted by cortex at 7:38 AM on December 6, 2006

Masters do not remember individual positions and individual moves that go with each position.

Masters don't only remember individual moves, but they certainly remember some. With openings, for example, you sometimes see move per move regurgitation for the first handful of moves.

Machines are using a database of billions of positions, each position individually, and stored with each position is a single move pointing to some other position in the database. This memory is nothing at all like how humans play.

This is how a program relying only on its brute force + database might go about it, but these days there are heuristics involved. It's still not how humans play, maybe, but it's closer than you suggest.
posted by juv3nal at 9:57 AM on December 6, 2006

I agree with juv3nal. The human brain is just several orders of magnitude more complex, having trillions of trillions of logic gates instead of billions of billions. If you split an operation into enough tiny steps, it appears to be a fluid process with "heuristics", but in reality, it is still just a massive array of computations, comparisons, and boolean expressions.
posted by tehloki at 10:00 AM on December 6, 2006

Machines are using a database of billions of positions, each position individually, and stored with each position is a single move pointing to some other position in the database. This memory is nothing at all like how humans play.

This is how a program relying only on its brute force + database might go about it, but these days there are heuristics involved. It's still not how humans play, maybe, but it's closer than you suggest.

This is true in the middlegame, but not in the endgame.

Programs in the last several years have started using what are called "tablebases" for endgame positions. The computer really does have a database for when there are fewer than 6 pieces left on the board. They database will say, "White wins in X moves with move Y". The computer just needs to hash the position, make the move, and the next position will say "White wins in X-1 moves ... ". Repeat until you win.

Computers used to have difficulty playing endgames -- the positions required long-term planning which the computers were bad at. Instead of solving that with software, computer chess makers have precomputed the solutions and incorporated the solution into the machine. This was exactly the right thing for them to do, as it is perfectly effective. It's just not AI.
posted by cotterpin at 10:26 AM on December 6, 2006

Re: winning = better. To my mind, as demonstrated by DE tournaments, a win doesn't indicate the better player - an overall weak player may be strong in a limited area that an overall strong player is weak in, so when they happen to play each other, their distinct styles mesh such that the weaker player usually wins. But put those two player in a round-robin tounament, where they each play against many different styles, and the true strength of the stronger player emerges, likewise the weakness of the weaker player.

By these conditions, a computer player does indeed play strong chess with meaningful wins, and as such, I would suggest that the assumption that a computer win has little or less to offer in the way of advancing the meta-state of the game, is probably quite false. Not that I am in any position of authority to say so. But for example, against Deep Blue, Garry Kasparov commented along the lines that some of the computer moves were things a human would not have done - fresh and counter-intuitive, like playing an eccentric (I don't remember his words, just the sentiment, and slightly out of context, as it was a run-up to a point that a few moves later, the goal became apparent). Computers have influenced the way chess is learned and played. I think it's begging the question to point to the understandability of the computer system as precluding the operations of that system from significantly advancing the meta-state.
posted by -harlequin- at 3:05 PM on December 6, 2006

as demonstrated by DE tournaments

- read that as: as demonstrated by the poor accuracy of DE tournaments in selecting the best player as winner (it does so with the least information, ie the fewest matches), compared to round-robin (which uses the most information - all possible matches)

Unfortunately, pretty much all world-level sports and games are DE these days because it's so much cheaper to run as few matches as possible. The DE system makes no difference in say, a race, but it really bugs me to see tactical/strategic sports have their placings determined on inadequate information.
posted by -harlequin- at 3:11 PM on December 6, 2006