Man-Machine Beef

Mig:

We sit around and wait for a blunder in human vs. human games, too. I don't see how its different.

I don't think human wins against the machines are ugly. That Kasparov's game vs. Fritz was beautiful. The game just before that where Kasparov lost because he blundered a pawn was much uglier, IMO.

But anyway, these humans vs. machine games are a lot of fun. They give us a way to objectively measure how these completely different chess playing entities match up in the game we love. What can be better than that?

Your comment about the different hardware is an important observation. Hardware was way overestimated in Man vs. Machine events recently. It was ok in the times of single-digit clock rates, but not now when we have 1.000+ MHz. I guess, as a rough estimation, the hardware gives ~40% of the performance and the program's intelligence (+ openings + tablebases) make ~60%.

Btw. I think the human master's success depends at least as much on his opening knowledge as the programs depend on their opening databases. In a FRC match, where both side have no such variant based knowledge, GM Rogozenko lost clearly against Chess Tiger (single cpu).

Actually I assume a loss of opening theory knowledge would be the bigger disadvantage for the humans. With effective 12+ plies of calculation depths, the progs will never blunder in the opening (and hardly ever play weak move).

Regards, Mike Scheidl

Looking at the crosstable, it's funny that Fritz on the notebook made "1st place". The sentence "The Bilbao event saw the humans winning only a single game against the silicon, and Fritz was running on a laptop" sounds like Fritz lost the only game, but it was Deep Junior. I know it was not meant that way. I agree that all the top engines are more or less playing at the same strength. Nice to know that you don't necessarily need some monster hardware to get more or less reasonable evaluations.

Mig, I guess I agree with you, in that I no longer have interest in man-machine matches, or even follow the games. Some of the comp-comp games feature incredible combinations that make their way into the _Informator_ combinations section, and that's all the comp chess I look at any more.

More and more, I'm moving toward a Vassily Ivanchuk point of view (iirc, he won't play comps, or participate in events which include them). I've also read that Mikhail Tal was opposed to playing comps...

Grandmasters play better Chess than computers some of the time. Computers play better Chess than Grandmasters some of the time.

What is really interesting, surely, is to watch a struggle in which one player is not afraid and the other player cannot afford to be afraid.

Everyone likes different things about this game. I think you will find that USCF regular memberships were highest around 1996 and 1997. I woudl find it hard to believe the Deep blue hype had nothign to do with this. It certainly pushed me into chess. As someone who knew the moves and little else, I found it amazing that any human could even come close to palyign a computer that can calculate 100-200 million "moves" (I later learned it was actually plies) per second! Up until then I really just thought Chess was just a matter of patience and takign the time to say ok bishop takes pawn Knight takes bishop etc. I admit I was very ignorrant in my views then but certainly not more ignorrant than most people who may have some seed of interest in this game. There was definitely something mysterious as to how Kasparov could do this and so my first interest was chess and computers. (I knew very little about etiher at the time.)

I know youll say so what? Becasue I admit I was a patzer before and yes I'm still a patzer. So nothing really happend that was good for chess. Just another patzer. Well that may depend on what different people think is good for chess. And thats a whole other topic.

Suffice it to say these man versus machine matches are interesting for a number of reasons wich include:
1) How can anyone beat a computer going at 3 million moves/nodes per second? I must say the numbers are so large it is still hard to get a handle on this.
2) Sure the best in the world can play eachother and beat eachother, by making certain moves I don't understand in the least, but show me someone who can beat my desktop monster and I'm interested. Sure I can take your word for it that these various Eastern Europeans would beat me just as much if not more than my desktop. However, after actually gettign crushed over and over it is like a magic show watchign people beat these monsters.
3) Games are just as intersting as those at the top level tournaments for most chess players. Smirin v. Tiger was a great game. So was the bishop sac in Junior v. Kasparov. The knight sac in Kramnik v. fritz etc. If Kasparov would have played on I think he would have won the game but still it was a fantastic bluff by the Deep Junior team.

4)Even if my point in number three simply reveals my ignornce of the game, there are far more chessplayers at my level than those who can really prefer the actual chess of a top tournament over the chess of Human v. computers.

Which raises the big question what are we trying to accomplish in having chess events of any kind?

In my view these matches are great. When asked how strong the computers are we still have so few games the answers are vague and involve huge ranges. I think one of the best things that could happen to chess is for some human to come back and win one of these hyped up matches. I would go so far as to say I think it would be good for chess if one of the top 10 players simply dedicated thier career at being able to beat these machines even though it woudl mean not being as good agaisnt other humans.

These matches are only interesting for the computer-crowd or the so-out-of-computer-crowd that they think a little magic-man lives in each CPU. On the level of an actual game they most resemble Super-GM vs Super-GM matchups and are thus prime examples of Stock-Exchange Chess accentuated by the undue influence of the opening book (say! can I use a book at my next tournament?) Being a patzer myself - what I love about chess is the cut and thrust of the clash of ideas and were I fortunate enough to ascend to the heights of Super-GMism I suspect that other than the money, fame, easy women, and so on, I would find it extremely difficult to actually enjoy the game. I might have to resort to micro-analyzing every aspect of the opening to attempt to remove all unclear evaluations from the books - all for the sake of finding some obscure position that I and my fellow gods could disagree about.

Being part of the computer-crowd myself I always describe them to the non-initiated as a "box-of-sticks" because in themselves they have no value. If you interact with a computer you are really interacting with a programmer's ideas. My brother was a big fan of the flight-sim, but after watching him fly a mission I realized that he had merely learned to parrot the correct (programmer-supplied) series of actions. Thus the chess program forces the human to play in a different way without offering any real ideas in return - to me a dismal fight.

Mig,

I couldnt agree more with your original post. What really made me laugh was the article by David Levy on Chessbase saying how much more interesting these comp-human matches are. I am a guy who really believes that each is entitled to his opinion, but this remark got my back up a bit: "My own view is that the games in Bilbao are far more interesting for the chess public than are those played in Brissago.". Probably he should have dropped the "for the Chess public". I am a member of the public, and I find these a huge yawn. I watch Sports and games because of the competition and drama between humans. A Chess playing computer is a good engineering accomplishment but no more and no less. I have a PhD in Electronics, and I know the tradeoffs in H/W S/W codesign, different algorithms, different technologies quite well. There is absolutely nothing amazing in an unbeatable Chess computer which most will soon be. If there is some novel algorithm maybe its worth a paper in a journal, but so what? Personally I find a Ferrari formula 1 car a better Engineering accomplishment, but that's just me. I guess sometime in the future there will also be a Snooker playing robot who gets break after break of 147, but I would rather watch Ronnie O'Sullivan play Paul Hunter. The same way I would rather watch Kramnik vs Leko than Frtznikhydra version 1 million 7 hundred thousand vs some poor GM hoping for some lunch money.

After the novelty wore off of having computer programs that could play chess on an (almost?) equal level with top players, I lost most of my interest in computer vs. human games. I would, for example, much rather see Kasparov or Kramnik (or Leko or Anand or any other top player) play a match against another top player than play a match against the latest incarnation of Fritz or Junior.

One of the problems with computer vs. human games is that they are mostly only instructive re: how to (or how not to) play against computers, and I'm just not that interested in playing against computers. Humans generally do not (and probably should not) play the same way against humans that the play against computers. The fact that computer vs. human games are "abnormal" chess in some sense and not particularly instructive removes much of their value, at least from my perspective.

Moreover, at least to some extent, I somewhat blame strong commercial chess engines for the modern emphasis on exhaustive opening preparation largely based on computer-generated (perhaps computer + human generated would be more accurate) analysis. Fritz's creators have a lot to answer for in my book, as they have contributed (and profited from) a process which perhaps is not ruining chess but is certainly detracting from my enjoyment of it. So without actually yet being a Luddite as far as chess engines are concerned (I do, after all, own Deep Fritz 7 and Shredder 8 myself), I am beginning to sense the appeal of that philosophy. Perhaps this explains my lack of interest in how these machines do against actual breathing humans.

I do not really understand Mr. Levy's suggestion that a chess engine's lack of fear is interesting in any way, nor does he explain while it should be. It seems to me that a lot of attempts by computer scientists to generate public interest in their products make use of a personification of computers which is, at its best, merely false, and at its worst, rather disingenuous. Of course Mr. Levy is correct that chess engines do not fear anything or anyone (O brave engines!), but they also do not tire, or hope, or dream, or become inspired, or feel triumph or despair, or do any of the other things that make humans so interesting. Nor can I take any pleasure from the possibility that human chessplayers might fear these machines (and need to control that fear), as Mr. Levy seems to suggest I do, as this would require a bit more of a sadistic mindset than I can conveniently muster. (I sometimes think that much of the enjoyment some people seem to get from watching computers beating GMs is in the form of a kind of vicarious victory over superior chess players. "I can't do that myself, but I can by a program for $49.99 plus tax that can," Joe Blow says to himself.) I can only really view Mr. Levy's post (and his companion piece at the ChessBase Online site) as a bit of harmless (well, mostly harmless) promotion of a field (computer chess, and in particular computer vs. human chess) in which he has a certain vested interest. I personally will continue to believe that human vs. human matches are more instructive and have twice the human interest of a human vs. machine match.

- Geof Strayer

Heres my spin. I enjoy the computer-computer matches, as well as the computer-human matches. If not only to prove the premise that good tactics will win over positional play most of the time. Re: How i gained 400 points (chesscafe.com)Michael de la Maza, and his book that followed based on that argument, Rapid Chess Improvement. The computers are just stronger tactically, period. Just like a GM is stronger than a master. Heres a case in point a recent game played by Nakamura against Karklins is an example of someone just getting outplayed tactically, too bad Nakamura was crushed but Karklins couldnt see the winning moves.

It's pretty obvious that computers play better chess than the overwhelming majority of grandmasters. The results tend to support this observation :-)

And, as David Levy so brilliantly put it, the matches between humans and computers have the same allure of Fischer's games: one of the two players comes to win at all costs, every move.

That's more than enough for me.

One idea that might help to uphold the human colors. Fielding a team of young high-ELO professionals who have World Champipnship aspirations was a recipe for disaster. These guys tried to outdo each other in balls-out attitude only to blunder miserably.
Next time let's sign up some old wise-ass players like Seirawan, Khalifman, Serper or Huzman, who would have no problem swallowing a little pride when it comes to exploiting computer weaknesses.
In general we need a team of anti-computer GM's working full time to turn this thing around. Anybody willing to underwrite it?
Strategy:
The human player should be merciless like Bareev was against his computer a year or two ago. Exchange pieces and it first sight of trouble pitch a pawn to escape into a rook ending. The machine will never refuse the bait!
Be cynical. Make ugly draws and wear the computer down! Sooner or later some quirky opposite-color bishop ending will appear and we'll see how the tin can handles it.

Why doesn't anyone use the infamous anti-computer systems that claim to win all the time, such as Eduard Nemeth's

http://www.xs4all.nl/~timkr/chess2/honor.htm

or from the book by Pecci?

http://www.amazon.com/exec/obidos/tg/detail/-/1929331045/qid=1097617385/sr=8-2/ref=sr_8_xs_ap_i2_xgl14/103-5397297-9745465?v=glance&s=books&n=507846

Are these systems obsolete against today's engines/hardware?

Because they only work if you get 50 takebacks during the game and then only publish the ones you win. Let's see Nemeth and Pecci put this stuff into practice in a live game with spectators. That doesn't mean their constructed wins don't prove a point. Those moves in that exact situation (clock, time control, CPU, settings) DID work to crush the machine. But doing it live is another story and they would lose 99% in practice.

Like most competent anti-computer players I was using the Stonewall and variants thereof 20 years ago; Pecci hardly created it. Kramnik used it to demolish Junior awhile back. These closed positions can still work but are harder to achieve every generation. For weaker players such systems are the only realistic hope of competing with top programs consistently. I say consistently because if you get the right opening you can swap and draw an endgame without too much difficulty, but most of the time you will lose material.

For GMs, such turgid system play is both boring and antithetical to what they are best at. Look at Kasparov against machines. What is he supposed to do, avoid tactics and initiative play when that is the only type of chess he can play? A GM who truly mastered such systems would be a computer-killer, perhaps scoring 50%, but it might have a deleterious effect on his chess against humans.

Apparently the new top-dog in computer chess is Hydra? Line up Kasparov to play it :)

"Such system play is turgid for GMs.." Exceping Vlad.

Bilbao's main conclusions are:

1. When comps win against GMs, it's NOT just because they have such large opening databases. Hydra comes with an opening book philosophy of limited depth (unlike other chess progs, so far), max. 10 moves, designed by GM Lutz. In one game against Topalov, Hydra was out of book after the 2nd move!
http://www.chessninja.com/cgi-bin/ultimatebb.cgi?ubb=get_topic;f=10;t=000298;p=1#000006
GMs and their fans have to search other exuses.

2. When comps win against GMs, it's NOT because of gigantic superhardware. Fritz scored equal to Hydra, running on a standard Centrino notebook.
GMs and their fans have to search other exuses.

3. GMs should apply anticomputer strategies. No chess moves, ideas, strategies which are in compliance with the chess rules, can be antiethical (?!). It is a game.

Regards,
Mike Scheidl

Bilbao's main conclusions are:

1. When comps win against GMs, it's NOT just because they have such large opening databases. Hydra comes with an opening book philosophy of limited depth (unlike other chess progs, so far), max. 10 moves, designed by GM Lutz. In one game against Topalov, Hydra was out of book after the 2nd move!
http://www.chessninja.com/cgi-bin/ultimatebb.cgi?ubb=get_topic;f=10;t=000298;p=1#000006
GMs and their fans have to search other exuses.

2. When comps win against GMs, it's NOT because of gigantic superhardware. Fritz scored equal to Hydra, running on a standard Centrino notebook.
GMs and their fans have to search other exuses.

3. GMs should apply anticomputer strategies. No chess moves, ideas, strategies which are in compliance with the chess rules, can ever be antiethical (?!). It is a game.

Regards,
Mike Scheidl

There are some interesting comments in the recent posts here. One view is that a game cannot be interesting per se if one or both of the participants is/are computers. This is akin to saying that the drama is there only if both participants are human, which is a perfectly reasonable point of view (but not one with which I would agree). To my mind the drama in a game of chess owes a great deal to the struggle between the players, no matter whether they are made of flesh or of silicon.

As to the question as why a lack of fear by programs is important in the struggle - the reason is that the GM stops thinking about draw offers and knows that his opponent is trying to win. This leaves the human having to fight even to achieve a draw, and I believe that a fight is what the chess public wants to see.

Mr. Levy makes an interesting point, as the high "contempt factor" displayed by computer programs does tend to avoid short draws. But maybe a better solution would be to find a way to increase human players' "contempt factor"? Would that Kramnik or Leko (and even Kasparov, these days) had a pull down menu that could be used to set their "contempt factor" to high. :)

-Geof

I just got an idea: "contempt factor" was always money. So perhaps price money should be splited according to the points... perhaps some of the players would fill like loosing money when they accept a draw? Something like 1000 for a win and 250 for a draw...

Regarding money "incentives":

It's quite ironic, but I feel the reason we see many draws is that top players don't want to risk bad results that could diminsh their earning potential. All this appears logical, until you think a little deeper and begin to realize that players are victims or their own "invention": by playing dull, ultra-cautios chess, they are guaranteeing that the public remains largely uninterested. That's bad for their pocket.

Sooner or later, the average grandmaster (the guy playing the swiss tournament for $200 first prize and generally struggling to make a living as a chess pro - what accounts to over 80% of them) will realize that shorter time controls and general riskier play will attract more people to the game. As it is now, chess is not self-sustaining and is dependent on the whims of a few "chess patrons". Hardly a dignified position for a professional in any field...including chess.

What ever happened to the idea of "Advanced Chess"? This idea is conceptually interesting although implementation into an attractive event is not obvious.

I also think that humans have less and less motivation each time to try harder and beat the machines. New formats need to be devised.
I do not know! A new technology is born and usually a new sport appears with it. We have to be creative. Teams of humans? One takeback to humans? Well, this actually sounds like a circus, so a balance will need to be found but I really think we need to find a place for the chess computer. It has already modified the practice of the game: no adjournments, an even greater emphasis in opening preparation.

Regards,

Mig,

I agree. I want to know who the best chess player in the world is, and knowing that a machine can beat him means about as much to me as the fact that it can memorize a phonebook better than him. It was interesting watching the progress of machines' skills to this point but now that they are here, let's turn them off as competitors.

Mig says August 17, 2004: If they actually started working on opening algorithms and tested them against each other, comps might have an impact on the openings like they have had on the endgame. (I doubt that, but the idea is there) ...
Okay, then maybe you aren't playing good chess, figure it out! Give'em a Reuben Fine book and teach them to play the openings.
Steve replies August 17, 2004: They CAN'T play good openings.

Mig says October 10, 2004: (And they are still overly dependent on their human-designed opening books, but don't get me started on that right now.)
Steve replies: I stand by my previous statement.

Your statement is meaningless unless you explain why or provide proof. We've seen programs play quite competently in the opening without books. They won't reinvent the King's Indian, but they don't have to do that to do well.

So if they can't play good openings they shouldn't have to? They play bad endgames too, should we abolish that for them?

Mig says: Your statement is meaningless unless you explain why or provide proof. We've seen programs play quite competently in the opening without books. They won't reinvent the King's Indian, but they don't have to do that to do well.

So if they can't play good openings they shouldn't have to? They play bad endgames too, should we abolish that for them?

Steve: My point is that they can't play the openings well. Competently? Maybe. But at a 2700+ level? I don't see it. You seem to believe they could, but the programmers just haven't chosen to.
The next time a computer match involving Kasparov comes along, do you really think the computer could compete without its opening books? I don't think the computer would have a chance. And I think it's a matter of inability, not a lack of effort by the programmers.

In 1960, it was a matter of inability that chess programs will ever reach a club player's strength, not a lack of effort by the programmers.

In 1980, it was a matter of inability that chess programs will ever reach a Grandmaster's strength, not a lack of effort by the programmers.

(etc.)

Mig wrote:
"It's true that the comps get a little better every generation, but as game three of Kasparov - X3D Fritz showed, machines can still play like brain-damaged gerbils in the wrong positions."

But this is a total misunderstanding of what happened. The computer was following its book -
Reshevsky,S-Keres,P/NLD/URS 1948 - which was a win for black. One might also add that black had/has a good reputation as a player so it is not a "mistake" per se to follow his play.

However, in this instance Black is just lost on move 12 and there was nothing the computer could do to overcome that. It had played a busted opening. In a lost position everyone/thing looks bad: think Kramnik in Kramnik-Leko game 8. It means nothing. Kasparov got very lucky or was it preparation?

To judge the current level of computer chess based on one faulty line in an opening book created by humans is asinine.

It's not when people post wrong information or are mistaken that bothers me. It's that they can be so aggressive and insulting at the same time. I guess volume is supposed to compensate for facts?

I was one of the commentators on that match. I spoke with the Fritz team, including book expert Alex Kure, about that game. Not only is Black not lost (unless Black is a computer), but it didn't follow its book to get there.

Fritz was out of its book after Kasparov's ninth move and its position is not that bad. But for Fritz, entirely the point of what I said about them playing like gerbils in the wrong positions, it was dead lost. It had no idea what was going on and thought the position was about equal until after move 40! It had been dead lost for at least 15 moves by that point, but didn't understand it at all.

Black could have mounted counterplay as late as move 22, but didn't. Faulty opening line FOR A COMPUTER, not for Keres or any other human.

My annotatations of that game for the official site: http://x3dchess.com/news/analysisgame3.htm

Mig,

Re X3dFritz-Kasparov 3rd game:
You argued both that computers don’t understand the position and that we should accept their evaluation of that position they don’t understand out to move 40. Hmmm. A contradiction.

Maybe it is better to actually look at it.

On the queenside, White is going to win the a-pawn because it can only be defended by a rook and it can be attacked with a knight, bishop and/or queen. Furthermore White cannot be stopped afterward from pushing his unopposed a-pawn to a6 when three things happen: 1) White gets a protected passed pawn on the sixth rank 2) White gets the only open file on the board and 3) White has a nice target, the weak c6 pawn. So White is totally winning by force and quickly on the queenside.

What is happening in the center? Nothing. It is blocked. In the 1948 stem game Sammy blundered with Nxe4 allowing Keres to put his knight on e4 and exert pressure on the kingside. But no club player would think of that blunder, so nothing is happening.

Of course Black will, in desperation, throw his pawns and pieces at White’s kingside but he has four problems: 1) Black is cramped and his pieces are misplaced (especially Nf6 and his K) so it will take too long 2) If he diverts power to the kingside, that will accelerate White’s breakthrough on the queenside and Black will quickly be down massive material that will be mating the abandoned black King 3) White naturally will be eating Black’s pawn chain in the center so that the possibility of defending White’s kingside from the center will exist and finally 4) If White is in an impish mood, he can simply move his King to the middle or queenside since he has so much time. He can afford to abandon all his kingside pawns.

The objective verdict – Black is just lost on move 12.

The proof is in the eating so let me make two suggestions. Play the white side of this with your friends and you will be shocked and amazed that you always win once I have explained it to you. Secondly, as to your assertion or main point that computers don’t understand this position, wrong-o, buddy. Let any engine have the white side of this position and they will beat you like a drum, smash your head in like an eggshell every time. They understand this position very, very well. They also understand there is nothing for Black to do on the kingside.

Amy (“Grandmaster” to you) Sommers

Amy, what are you saying is that the Engine understood the position for White but not for Black. The Engine is suppossed to understand both sides to win a game. But in this particular game, the Engine fell dead even before it understood its own position. How can you imagine that it understood the enemy's position? I advise you to study Sun Tzu's Art of War!

Actually, in the line given by Mig in his notes to 14...Bd6?!, it seems as if Black is getting a lot of compensation for the a-pawn (possibly even too much compensation, although this is not entirely clear) after a quick f4. Miss Sommers' comments do not take into account that it actually costs White some tempi to grab the a-pawn and then extract his pieces, and meanwhile Black can force open lines to White's king. Of course it would be nice if White could (i) take the a-pawn, (ii) extract his pieces from the pin on the a-file; and (iii) run his king to the queenside, all without giving Black any play in the center, but unfortunately Black is allowed to make moves also.

I would think the position after the pawn storm line with 16...g5 as given by Mig is most objectively decribed as "unclear" or possibly as "Black has (excellent) compensation for the coming loss of the Pa5." Computers might not see the compensation, because the fruits of such an attack are many ply away. But most strong humans would see it, and this is the kind of position in which a strong human player with Black might well beat a strong computer.

Just my opinion.

-Geof