To kick off some technical discussions

[fserrano]

I must admit I have follow this thread with great interest. Let me point same ideas I have:

a) Current Chess IA is nearly dead. 99% people here apply known algorithms and the only research done is about changing what I call "inputs in the formula". We research if 325 as value for knight is good or if it is better 300, or if LMR must be doing from 4th move or from 3th. For me this research has nothing to contribute to chess computer.

b) Derived from a), I would like to see people appling human imitated actions for chess. I mean: OTB has shown that been the world champion can be done from different styles. Why dont researchers trying to imitate this? In robotic, a robot which can play the guitar is a great advance. We can not measure its success, but there is IA in it because it is emulating humman behavior (is it passing Turing test?)

c) In this sense I have the idea that changing the way current engines search can be a great field to research. If we program the search to do an eval before each node and not only at qsearch, and based our decision in what we have found, I thing we would see a lots of new and fresh ideas, and they are not of the type "change 325 for 300". This, I thing, is the way Rebel and very few other do, and it is the demostration that this can work.

In fact, I have been thinking for a long time now about take that direction myself with Rodin....
conclusion: I think that in current chess computing we are not making new "IA".

Regards,
Fermin

[Fernando]

Fermin:
Great idea you have to try something else. At the very least you will entertained trying it instead of changing values of the rook .
If you ever begin the work and you believe a poor amateur can be of some help here and there, let me know.

Fernando

[Chris Whittington]

Here's a question: How do you _KNOW_ it is unsound? What if a 40+ ply search proves that it wins, even though it just looks like it tosses the queen out a window? To be certain, you have to be able to search deep enough to see the win, loss or draw. Anything less and it is only speculation.

well, discounting obviously unsound sacrifices (such as Ed's Qxh7 in the board above your post) and combinational sacrifices that aren't really sacrifices at all because the win back is forced and easy to see, we have unclear sacrifices left which are the type I concentrate on.

the effect in a computer game of such a sac is zero compared to the effect in a human game - the opponent may well be unsettled by the sac. Even if later you can prove the sac unsound, it was sound in the sense of the game it was played in. It might well remain sound until someone finds a refutation, maybe the refutation is good, maybe someone else will unrefute it.

Where is your WIN-obsession in all this? Suddenly not terribly important compared to the fun, excitement, understanding, gain in knowledge, effect on others and so on. All lost in computer chess development concentrating on statistical win/loss/draw/ELO. Shock effect on the opponent - lost to computer chess too.

btw, your 40 ply search may well prove nothing, even if it claims a positive score, we already know that huge branches of the tree, specifically those with low material scores get pruned away - you only search your conical section, as the other machine opponents search their conical sections, Thorsten will remember the number of times CSTal would sac against some beancounter, Fritz say, where Fritz would immediately go positive, and then suddenly, several moves later, see the loss. Not only had the beancounter pruned away the sac, it pruned away the continuations from it as well - as it would, it searchs a materially biased subsection of the tree only - as does Crafty. Some sacs can remain very uncertain for a very long time, for some the jury is still out.

Answering your question again, in another way, you know quickly for some sacs, slowly for others, and perhaps "never" for others. Within all that there are games to be played as the knowledge of the soundness gets more or less slowly or never determined.

[Chris Whittington]

I'm still waiting for objective measurements. Looking for material down won games is subjective. Which is better, to be a queen down for 10 moves and then win, or to be a rook down for 24 moves and then win? Which is better, to be a queen down and almost win, but ultimately lose, or being a queen up and almost losing but ultimately win?

True science requires accurate measurement. Elo gave us an accurate measuring tool. Since there is no other measurement around, discarding the only objective one we have seems a bit rash.

Now if one wants to simply rank players on a "Tal-ish" scale (or a Karpov-ish scale for boring) there's nothing wrong with doing that. And if one wants to move from the boring side to the Tal side, without hurting overall performance, there's nothing wrong with doing that. However, the flashiest sword-fighter in the world still has the same problem against a gunfighter. He ends up dead, no matter how dashing his personality is.

Chess is a game for humans to enjoy. They enjoy playing and watching exciting games. Exciting games very often involve sacrifices. So, measuring programs on a scale which assesses excitability (with various fuzzy parameters to account for your above and other stuff) seems eminently senseful to me. And quite easy to do.

Maybe you are just worried Crafty is not a very exciting opponent, after all, as you keep telling us, it is optimised for boring old win/loss/draw/ELO performance?

btw, the swordfighter lives again to fight another day (nobody is killed in chess) but he does get the newspaper coverage, the articles and almost certainly the girls, not to mention the wine. The boring gunfighter goes home to loneliness and his Wii .....

The flat earth is also more romantic. But people who cling to it are called stupid. Science banishes darkness. You may think chess is only a game so we can cling to darkness there in safety. But this is harboring a viral infection in our brain that can only end in religion. Be careful of protecting treasured feelings at the expense of science. A joy like that must be dropped like a bad habit.

There are more assertive conjectures in the above than I care to argue about, but, for starters, your initial assumption that computer chess as it currently is performed is science is staggeringly unsound. Computer chess may have once held out some sort of hope for usefulness or even as a hope as a platform on which science can be done, but the science is long gone, the development goal of ELO, ELo and more ELO is a) going nowhere and b) utterly obsessive and the whole process is about as unscientific as you can get.

Gimme fun every time instead.

I hope you will not take offense if I observe that for a statement about science this seems very subjective. To say the science is gone and that Elo is going nowhere is backed up by nothing but the air it takes to speak. What you are really saying is that the "art" is gone and that the science of game tree search is not lending itself to subjective ideals of this supposed art. I would contend the art was a side-effect of recovering from or punishing less than optimal play. It was an illusion of beauty that could only arise from errors.

The science has disappeared partly in that people who call themselves 'scientists' are no longer interested in the subject and partly because the field practitioners are engaged in a sort-of obsessive hobby rather than a scientific discipline. Despite the pretense over many years, the pretense of the academic journal, the pseudo mathematics and statistics, this never was a science, it was a con-trick designed to get some money through bullshit. That money has now dried up.

Science re rating lists, for example.

1. The data is improperly collected without controls

2. There is political-personal interference in what is and what isn't allowed as data (banned programs)

3. The games played come from a play pool that is restricted to like-playing entities and therefore veers off to nowhere or at best a cul-de-sac

etc.


ELO going nowhere, means: what is the point of obsessively working on programs and then having others obsessively test them, publishing allegedly statistical scientific enumerated lists of how these programs (crippled chess entities with super lookahead) perform against each other in an endless quest for more ELO points.

What are they going to do with all these ELO points? What does it matter? Isn't it just an obsessive behaviour, pointless, filling in the void?

It is not pointless in actual fact. The reason is because the search is always going deeper. Search eventually corrects errors because search eventually discovers the end of the game.

You criticize programs playing each other, but it's like Kasparov's comment on why he engaged in post-game analysis with Karpov, and man he detests. His answer? "Who else can I discuss these things with?" Who else was able to play at Kasparov's level? Only Karpov at the time.

The same is now true of the computer. No human is in the same league with the best programs anymore. Therefore, how is your statement being objective in saying no science has happened or is happening?

it is pointless because search will NEVER get to the end of the game. You can stop the development process now, next year or in ten years time. My contention is stop now because there's nothing to be gained anymore, the target is unattainable anyway and any targets that there were, were reached long ago - so why continue the incestuous fight other than for reasons of obsessiveness?

It is not science, it's an obsessive hobby.

No, because there _is_ a goal. I want to win _every_ game I play. Whether it be tennis, pool, shooting skeet, playing chess, etc. There are exceptions in that when my kids were very young, I "threw" many games intentionally because nobody likes to lose every game. But for every game I have ever played, from high-school football, basketball, baseball, track, for every "board game" I have ever played, from chess, checkers, monopoly, life, careers, for every card game I have played, from poker, rook, blackjack, hearts, bridge, there is one common denominator. I play to win. And everyone measures their own success by measuring how often they win. So trying to take "winning" out of the game of chess removes the "game" completely. That is the _point_ of playing. A "mini-war" played on an 8x8 board.

well, I'm no psychologist and we can probably be grateful for the lack of Rolf, but there's rather obviously a psychological element here, about "desire to win", the degree of it above all else, what must perhaps be an unconcern for the 'pain' of the opponent (reflected in your own pain on losing) as opposed to say, Fernando's approach which is to have fun (I simplify).

At a purely mechanical level (ie within the rules of the game, how the pieces move, what is mate etc) obviously the goal is to win, but at a human level there have to be many many other goals. You might play a better player knowing you'll lose but gaining by learning something. You might win a lot of games but lose something much larger, you might even do as I did, lose a lot of games, but win something much larger

As a programmer, you can define the meaning of 'win' for yourself. It doesn't have to mean ELO. You don't have to accept the easy definition the consensus applies. There are many fields of battle and you don't have to allow yours to be defined for you by the lowest common denominator.

[Fernando]

I would like to correct here what seems to be my supposed approach to this matter such as I have seen appearing from the view of Bob and Chris. It is an important correction not because it matters if I am am fully understood or not, but because my approach in a sense reflects what is in the soul of most players.
I also want to win in anything I do. But, as Chris said, there are lot of meanings in the word "win" or at least they can exist. In some areas there is just one meaning: you win money or you does not. You win the battle or you lose and ever lose your life.
Nevertheless, in the field of games by definition that meaning is somewhat wider.It must be.
A game is a game because there you have a mocked fight, a mocked zero-sum situation. The essence of a game is getting entertainment. And that means not only the desire to win the game, but experiencing all the other features of being in a game.
In the case of chess, I get great pleasure when I play a move that, not knowing -as it is the case- almost nothing about opening theory or endings, I nevertheless play the correct move. It is like solving a math or logical problem. Then I get pleasure if I can see in a position a clear-cut plan that makes sense and give to my game a direction, even if lastly the plan show to be false and nothing comes of it. And I get pleasure if I see a combination, mine or in the soul of the computer and I do the right thing to avoid it or to perform it. And I get pleasure in the sheer fact of feeling how my brain is still capable of performing concentrated task with a decent degree of efficiency. AND I get pleasure if the engine hits me with an unexpected killing combination -as bloody Cstal use to doe- because I feel a kind of masochistic pleasure in seeing how what you can think about life or about chess is almost always just a piece of the full action, of the complete picture. And then I have the pleasure of learning this: what was hidden in that position and which are the limits of my chess intelligence.
And the, as you can see, I have already got lot of pleasure LOT before winning or losing the game....
Yes, if i win all that is increased, I feel I am a genius, etc, like everybody else here, but if not, anyway nobody can take from me all those previous forms of pleasure.
And so, that is the reason it has sense to play an engine capable of giving MORE of all that than a simple bean counter as Chris use to say...

My best to all
Fern

[Chris Whittington]

I would like to correct here what seems to be my supposed approach to this matter such as I have seen appearing from the view of Bob and Chris. It is an important correction not because it matters if I am am fully understood or not, but because my approach in a sense reflects what is in the soul of most players.
I also want to win in anything I do. But, as Chris said, there are lot of meanings in the word "win" or at least they can exist. In some areas there is just one meaning: you win money or you does not. You win the battle or you lose and ever lose your life.
Nevertheless, in the field of games by definition that meaning is somewhat wider.It must be.
A game is a game because there you have a mocked fight, a mocked zero-sum situation. The essence of a game is getting entertainment. And that means not only the desire to win the game, but experiencing all the other features of being in a game.
In the case of chess, I get great pleasure when I play a move that, not knowing -as it is the case- almost nothing about opening theory or endings, I nevertheless play the correct move. It is like solving a math or logical problem. Then I get pleasure if I can see in a position a clear-cut plan that makes sense and give to my game a direction, even if lastly the plan show to be false and nothing comes of it. And I get pleasure if I see a combination, mine or in the soul of the computer and I do the right thing to avoid it or to perform it. And I get pleasure in the sheer fact of feeling how my brain is still capable of performing concentrated task with a decent degree of efficiency. AND I get pleasure if the engine hits me with an unexpected killing combination -as bloody Cstal use to doe- because I feel a kind of masochistic pleasure in seeing how what you can think about life or about chess is almost always just a piece of the full action, of the complete picture. And then I have the pleasure of learning this: what was hidden in that position and which are the limits of my chess intelligence.
And the, as you can see, I have already got lot of pleasure LOT before winning or losing the game....
Yes, if i win all that is increased, I feel I am a genius, etc, like everybody else here, but if not, anyway nobody can take from me all those previous forms of pleasure.
And so, that is the reason it has sense to play an engine capable of giving MORE of all that than a simple bean counter as Chris use to say...

My best to all
Fern

well, I did qualify with "I simplify", but never imagined the degree of complexity of the correction

[hyatt]

Here's a question: How do you _KNOW_ it is unsound? What if a 40+ ply search proves that it wins, even though it just looks like it tosses the queen out a window? To be certain, you have to be able to search deep enough to see the win, loss or draw. Anything less and it is only speculation.

well, discounting obviously unsound sacrifices (such as Ed's Qxh7 in the board above your post) and combinational sacrifices that aren't really sacrifices at all because the win back is forced and easy to see, we have unclear sacrifices left which are the type I concentrate on.

the effect in a computer game of such a sac is zero compared to the effect in a human game - the opponent may well be unsettled by the sac. Even if later you can prove the sac unsound, it was sound in the sense of the game it was played in. It might well remain sound until someone finds a refutation, maybe the refutation is good, maybe someone else will unrefute it.

Where is your WIN-obsession in all this? Suddenly not terribly important compared to the fun, excitement, understanding, gain in knowledge, effect on others and so on. All lost in computer chess development concentrating on statistical win/loss/draw/ELO. Shock effect on the opponent - lost to computer chess too.

I prefer the shock effect of beating a human that expects to win, rather than the shock effect of popping out a sacrifice that ultimately causes me to lose the game. The goal of beating humans is pretty well behind us thanks to today's hardware and software. To beat the "best" chess players, "shock value" is meaningless since I've never seen a computer react to anything with emotion.

btw, your 40 ply search may well prove nothing, even if it claims a positive score, we already know that huge branches of the tree, specifically those with low material scores get pruned away - you only search your conical section, as the other machine opponents search their conical sections, Thorsten will remember the number of times CSTal would sac against some beancounter, Fritz say, where Fritz would immediately go positive, and then suddenly, several moves later, see the loss. Not only had the beancounter pruned away the sac, it pruned away the continuations from it as well - as it would, it searchs a materially biased subsection of the tree only - as does Crafty. Some sacs can remain very uncertain for a very long time, for some the jury is still out.

Answering your question again, in another way, you know quickly for some sacs, slowly for others, and perhaps "never" for others. Within all that there are games to be played as the knowledge of the soundness gets more or less slowly or never determined.

[Chris Whittington]

Here's a question: How do you _KNOW_ it is unsound? What if a 40+ ply search proves that it wins, even though it just looks like it tosses the queen out a window? To be certain, you have to be able to search deep enough to see the win, loss or draw. Anything less and it is only speculation.

well, discounting obviously unsound sacrifices (such as Ed's Qxh7 in the board above your post) and combinational sacrifices that aren't really sacrifices at all because the win back is forced and easy to see, we have unclear sacrifices left which are the type I concentrate on.

the effect in a computer game of such a sac is zero compared to the effect in a human game - the opponent may well be unsettled by the sac. Even if later you can prove the sac unsound, it was sound in the sense of the game it was played in. It might well remain sound until someone finds a refutation, maybe the refutation is good, maybe someone else will unrefute it.

Where is your WIN-obsession in all this? Suddenly not terribly important compared to the fun, excitement, understanding, gain in knowledge, effect on others and so on. All lost in computer chess development concentrating on statistical win/loss/draw/ELO. Shock effect on the opponent - lost to computer chess too.

I prefer the shock effect of beating a human that expects to win, rather than the shock effect of popping out a sacrifice that ultimately causes me to lose the game. The goal of beating humans is pretty well behind us thanks to today's hardware and software. To beat the "best" chess players, "shock value" is meaningless since I've never seen a computer react to anything with emotion.

btw, your 40 ply search may well prove nothing, even if it claims a positive score, we already know that huge branches of the tree, specifically those with low material scores get pruned away - you only search your conical section, as the other machine opponents search their conical sections, Thorsten will remember the number of times CSTal would sac against some beancounter, Fritz say, where Fritz would immediately go positive, and then suddenly, several moves later, see the loss. Not only had the beancounter pruned away the sac, it pruned away the continuations from it as well - as it would, it searchs a materially biased subsection of the tree only - as does Crafty. Some sacs can remain very uncertain for a very long time, for some the jury is still out.

Answering your question again, in another way, you know quickly for some sacs, slowly for others, and perhaps "never" for others. Within all that there are games to be played as the knowledge of the soundness gets more or less slowly or never determined.

well, ok, so computer chess development under your guidance is no Manstein, Napoleon, Hannibal or Patten, rather it's Montgomery or Paulus

[hyatt]

Actually, it is not a "person" at all. We noticed, years ago, that Cray Blitz became more Karpov-like as the speed improved. That is, even without modifying king safety, it became more cautious / solid / etc in its style of play. We never considered that to be a bad thing. We all liked Kasparov's slashing style of chess, and cheered him on when he beat Karpov eventually, but playing cautious chess never seemed like a bad or a good idea, it was more of a result. I'd label Crafty the same way today. Yes, you give it a chance and it will whip up something violent and quick, but in general against a really good human, it sort of "sits and waits". For that inevitable mistake. I've seen similar players in most every sport. From a tennis player that hugs the base line and returns everything you throw at him rather than rushing the net and ending a point with an exciting passing shot, to a football team that grinds the ball down the field "3 yards and a cloud of dust" at a time, rather than completing a flashy trick play or deep pass. We all prefer the trick plays and deep passes, but more often than not, 3 yards and a cloud of dust wins championships.

Style evolves from abilities. Not the other way around. Nothing wrong with wanting flashy, knowing the inherent risks. Nothing wrong with wanting "solid" either, knowing that an occasional risk-taker might slip one past your guard. The question revolves around "what works". And for the moment, careful testing and evolutionary changes are working. Whether that continues for a long time or not, I don't know. Our current version is about +50 better than the version we released 3-4 months ago. That's a significant trend supporting our current methodology. Burns a ton of computer time, but I have a ton of computer time to burn, so it works for me. And so far, we have discovered no asymptote providing a limit to what we can do with this approach. May well be one, but it is not yet close enough to see.

[orgfert]

well, I'm no psychologist and we can probably be grateful for the lack of Rolf, but there's rather obviously a psychological element here, about "desire to win", the degree of it above all else, what must perhaps be an unconcern for the 'pain' of the opponent (reflected in your own pain on losing) as opposed to say, Fernando's approach which is to have fun (I simplify).

At a purely mechanical level (ie within the rules of the game, how the pieces move, what is mate etc) obviously the goal is to win, but at a human level there have to be many many other goals. You might play a better player knowing you'll lose but gaining by learning something. You might win a lot of games but lose something much larger, you might even do as I did, lose a lot of games, but win something much larger

As a programmer, you can define the meaning of 'win' for yourself. It doesn't have to mean ELO. You don't have to accept the easy definition the consensus applies. There are many fields of battle and you don't have to allow yours to be defined for you by the lowest common denominator.

Your position is clear now. Robert Falcon Scott was a hero in losing while Amundsen (bean counter like) was the lesser man for winning with less drama and a colorless journal-keeping.