Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Human-Like Playing Style
- - By Zherkovv (***) Date 2007-12-06 12:28
Dear Chess (engines) Friends,

Which engine has the most human-like playing style:
Pro-Deo, Homer, Glaurung, Gandalf, Hiarcs or Rybka?

They also told me that "Pro-Deo" is good for analysing
grandmaster games?

Regards

Vlad Zherkovv
Parent - By Uly (Gold) Date 2007-12-07 00:33
I once made a ProDeo personality that played the most human like to me, however, I think that it played about 200 rating points weaker than the original (Though having a 2500 rated engine that plays like a human is nice, remember that it's basically playing 600 rating points weaker than Rybka).

Also, I don't think that's an answerable question, what seems human to someone may seem 'computerish' to someone else, so you may want to check the engines yourself to find which one seems to play more like a human to you.

And about analyzing GMs games, all engines are going to blunder at one point, or ignore the best line in a position, so you have to use the engine that has the lowest probability of doing such a thing. In this sense, nothing is better than Rybka, so far (Unless you have a 8 core machine, then you may consider getting Zappa).
Parent - - By SillyFunction (**) Date 2007-12-08 15:31 Edited 2007-12-08 15:57
For me, I feel a long-term plan or a short-term one while playing human.
But I feel a long-term trap or a short-term one while playing computer.
Now even short-term traps of engines often ruin long-term plans of human.
I predict that we will say "human style" less and less but say "human error" more and more.

And my 2 cents, the top five engines are disgusting for all grandmasters.
Engines make grandmasters feel their human error more than human style.
We see that Rybka is above any other engines in the list.
But I believe that the human who play all the top five engines would feel they are the same.
All of them can always exploit the human error.
Parent - - By Uly (Gold) Date 2007-12-09 10:52

> I predict that we will say "human style" less and less but say "human error" more and more.


To me, "human style" Vs. "computer style" is compared in positions in where several moves are playable, and the engine picks one that looks weird to a human (Like returning the knight from f3 to g1 without any seemingly reason, but the move is playable.) This has nothing to do with the strength of the moves (So, we wouldn't call a "computer move" one that a human would never play, if it proves to be the best move.)
Parent - - By Felix Kling (Gold) Date 2007-12-09 10:59
Humans know what happened before in a game, engines don't use that information.  A human will always think "oh, I played Ng1-f3 and now I want to play Nf3-g1... isn't this a lost tempo? Do I admit that I made a mistake when playing the knight back to g1?" of course humans sometimes also play this move, but they made the experience that "taking moves back" like this is not good in a number of cases. it would be possible to tell an engine that it should avoid that, but I wonder if it would benefit it's play.
Parent - - By Uly (Gold) Date 2007-12-09 11:47

> it would be possible to tell an engine that it should avoid that, but I wonder if it would benefit it's play.


In my example, Ng1 is just as good as other good moves, so avoiding it won't increase the strength a all, it would just change the engine's style to make it more human like (to the human eye.)
Parent - - By Felix Kling (Gold) Date 2007-12-09 12:24
No, i mean if you would tell the engine in general to avoid such moves, it could improve it's play since those moves are mostly not that good. Anyway, a good engine will come to the same conclusion without that rule I guess.
Parent - - By Banned for Life (Gold) Date 2007-12-09 12:29
You are overlooking the fact that sometimes doing nothing is the best thing to do.

Alan
Parent - - By Uly (Gold) Date 2007-12-09 12:55
F.Kling :

> No, i mean if you would tell the engine in general to avoid such moves, it could improve it's play since those moves are mostly not that good.


I think it could be the opposite. After all, the engine has done a lot into looking into the possible positions, and Ng1 seems the best to it, maybe by a 0.01 score, but the engine prefers the upcoming positions from there. By adding a "artificial" penalty to the move just because it seems weird to the human eye, is altering the engine's style, and it could be detrimental to the overall playing strength.

Alan :

> You are overlooking the fact that sometimes doing nothing is the best thing to do.


But that's what I hate the most about Rybka's style. Sometimes I just feel that Rybka, instead of attacking the opponent directly, just turns the board into a mine field, and then sits and waits for the opponent to touch a mine and explode. After the opponent has made this inaccuracy, Rybka suddenly becomes a monster and eats the opponent alive, but I'd like to watch Rybka becoming this monster from the beginning without needing to wait for these enemies' inaccuracies. And, if I understood Vas right, he actually agrees with me (Otherwise he wouldn't be changing Rybka's style for Rybka 3.)
Parent - - By Felix Kling (Gold) Date 2007-12-09 14:27
Yes, I said "could", I'm not sure about it :)
Parent - - By Uly (Gold) Date 2007-12-10 22:20
I'm not sure either :)

I was just saying that there's nothing bad with "computer style" or "computer moves", and that I see no reason for trying to get rid of them (Unless you want your engine to pass the Turing test ;) .)
Parent - By FWCC (***) Date 2007-12-11 01:07
Gata Kamsky is back!After knocking out Carlsen I wonder how his style would be against a Rybka.Of course to put up such money for a match would be difficult but I would like to see that either Gata or Anand vs Rybka.Are there any Kamsky games against a strong comp back in the day?
Parent - - By SillyFunction (**) Date 2007-12-11 03:13
Vytron, I mean that sooner or later engines will get near-perfect moves.
And that human will realise that many of "human plans" or "human styles" have some flaws against near-perfect moves.
Thus, we will eliminate those flaws even if they are played for a long long time as "good human style."
And engines help us do so by showing the more correct paths to survive.
We still disbelieve engines in many ways nowadays.
But as you know it, we tend to believe it more and more as time goes by.
In chess, we believe in unbeatable things, not good style.
Parent - - By Uly (Gold) Date 2007-12-12 08:29
But on a position, there are several possible moves, and on a drawn position, most of them draw.

So, how do you know what drawing move is better? It depends.

Against a stronger opponent, you want to pick the shortest path to draw (because you assume that if you don't draw he's going to beat you), and against a weaker opponent you want to pick the longest path to draw (because you assume that when the game gets longer, the chances of your opponent making a blunder increases.)

But when you don't know, or when the strength is close, it doesn't matter (And, you don't know the lengths of the paths to draw anyway), so it is a matter of style, see:

On a position, several moves draw, but a human wants to move O-O, because his style includes casting as soon as possible, and he sees no reason for doing otherwise, as all his pieces but the Rooks and Queen are developed. On the other hand, the engine thinks that Nf3-g1 is the best move, I could say that "It thinks that white is better castling queenside, and that the f4 pawn should be pushed, also, it thinks that the Knight would be of more help on h3, and that the enemy is in a 'zugswangish' position so losing a tempo by withdrawing the knight is advantageous" but that wouldn't be true. The truth is that all the engine has done was generating millions of positions, assigning them a value and comparing them, and generating even more positions from these, and comparing them, up until the ending position coming from Ng1 has a better score than anything else, and it plays it.

In the end, with perfect play, both O-O and Ng1 draw, so both of them are basically perfect, and what move was made will depend on style.
Parent - By Hetman (*****) Date 2007-12-12 18:36
I agree.
Up Topic Rybka Support & Discussion / Rybka Discussion / Human-Like Playing Style

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill