Topic Rybka Support & Discussion / Rybka Discussion / Playing strength of Rybka and GM. Comparison (3284 hits)
![[de]](/mwf/v2/flags/de.png "Germany")
I examined the games between GM and Rybka and came to the conclusion that:
1. A good GM has very good chances to draw against Rybka. This is not a matter of the color.
2. When a good GM plays defensive and drawish moves, Rybka (and all other programs) also answers with defensive and drawish moves, even when the (from the makers of Rybka announced) human
opponent has 200..300 Elo less. A few moves later, the game is trapped into a theoretical draw, without any escape chances to make the theoretical gameresult open.
3. The Contempt factor (to avoid draws against GM) is NOT an instrument to break out of this defensive and drawish play.
Gameplay still is as drawish as without such a factor, and the game is going to be trapped into an easy to hold theoretical draw.
4. I think in future development of Rybka, Vas has to find a way to teach his fish how to to avoid drawish positions and when they are once reached how to escape from them.
This could be reached with material and symmetrical imbalance play. The program should also know something obout potential of a Position. The potential of the
opening position is high (on the other hand the evaluation is -0.01 for Black).
The potential of K vs K is low (the evaluation is 0.0 for Black). So, Rybka would avoid the opening position and would prefer to reach K-K with black.
The potential of a very complex tactical position could be low, because after finding the 3..6 tactical moves the position could end into a easy to hold theoretical draw.
5. In my opinion in this behaviour a human GM has *much* more knowledge (and playing strength) than every chessprogramm .
6. My conclusion: Nowdays human GM (between 2500 and 2800 ELO ) are in generall superior to Rybka.
In My conclusion i don't take short tactical mistakes due to time-handicap or random influences
into account.
1. A good GM has very good chances to draw against Rybka. This is not a matter of the color.
2. When a good GM plays defensive and drawish moves, Rybka (and all other programs) also answers with defensive and drawish moves, even when the (from the makers of Rybka announced) human
opponent has 200..300 Elo less. A few moves later, the game is trapped into a theoretical draw, without any escape chances to make the theoretical gameresult open.
3. The Contempt factor (to avoid draws against GM) is NOT an instrument to break out of this defensive and drawish play.
Gameplay still is as drawish as without such a factor, and the game is going to be trapped into an easy to hold theoretical draw.
4. I think in future development of Rybka, Vas has to find a way to teach his fish how to to avoid drawish positions and when they are once reached how to escape from them.
This could be reached with material and symmetrical imbalance play. The program should also know something obout potential of a Position. The potential of the
opening position is high (on the other hand the evaluation is -0.01 for Black).
The potential of K vs K is low (the evaluation is 0.0 for Black). So, Rybka would avoid the opening position and would prefer to reach K-K with black.
The potential of a very complex tactical position could be low, because after finding the 3..6 tactical moves the position could end into a easy to hold theoretical draw.
5. In my opinion in this behaviour a human GM has *much* more knowledge (and playing strength) than every chessprogramm .
6. My conclusion: Nowdays human GM (between 2500 and 2800 ELO ) are in generall superior to Rybka.
In My conclusion i don't take short tactical mistakes due to time-handicap or random influences
into account.
![[us]](/mwf/v2/flags/us.png "United States")
What was your examination method and how does that method lead to these conclusions?
Good question.
Several years ago chess progammers asked GM how they choose their moves. .... Did they ever get an satisfactory answer ( a method how to find good moves) ?
I just watched the games, pridominantly life games and translated my feelings into words.
Several years ago chess progammers asked GM how they choose their moves. .... Did they ever get an satisfactory answer ( a method how to find good moves) ?
I just watched the games, pridominantly life games and translated my feelings into words.
Okay; as for your report, you are almost certainly correct in your statement that a modern GM has much more knowledge than Rybka, but very incorrect in the parenthetical "and playing strength". There is only one objective way to measure playing strength, and it's one that is agreed upon by pretty much everyone: the elo system or something very similar. In this fashion, Rybka is far superior, and it's not due to tactical abilities--good GMs don't fall into the tactical "traps" set by computer programs, and Rybka doesn't often set such things in the first place. As for which one is "superior", that depends quite on what you mean. A GM can tell you that such and such move is a "good" move because of a number of reasons that he can explain. However, he will often fail in finding the *best* move in the position, usually due to small and generally unnoticeable flaws in implementing his vastly superior knowledge. Rybka, however, doesn't have such flaws, and is able to implement it consistently at very great search depths. Thus, if you are interested in the question of "what is the best move in this position?", you would do well, more often than not, to take the advice of Rybka instead of the GM. Of course, you would do even better if you are familiar with the types of positions in which you should listen to Rybka compared with the types of positions in which you should listen to the GM.
I should also add that there are many human versus computer games nowadays in which the human is a strong IM or GM and plays very good chess without any apparent mistakes--and yet still loses. This goes along with the very small but unnoticeable mistakes that GMs make in implementing their knowledge--and this has nothing to do with tactical mistakes--these are slight positional mistakes that accumulate against engines like Rybka.
Short tactical mistakes are a part of Chess, so what you describe, the game you describe that all the GM's are superior, is not Chess.
Do you want to call it Chess with takeback, Chess with having the GM's not to play tactical mistakes, ok, but it's not Chess. In chess GM's always did, and will always do small tactical mistakes as also blunders....
Also short tactical mistakes does not occur only because of time-handicap or random influences, whatever that means. They happen because humans are incapable of not making them.....
I agree that humans understand Chess better, but since mistakes are a part of Chess, the results show that computers play Chess better....
Humans understand it better, computers play it better....
Do you want to call it Chess with takeback, Chess with having the GM's not to play tactical mistakes, ok, but it's not Chess. In chess GM's always did, and will always do small tactical mistakes as also blunders....
Also short tactical mistakes does not occur only because of time-handicap or random influences, whatever that means. They happen because humans are incapable of not making them.....
I agree that humans understand Chess better, but since mistakes are a part of Chess, the results show that computers play Chess better....
Humans understand it better, computers play it better....
![[cr]](/mwf/v2/flags/cr.png "Costa Rica")
A lot of knowledge will not really help if you generally fail in other things. Why do you believe that Top Gm's cannot win to other engines with lower ratings than Rybka under tournament time controls?
Regards,
Gambito.
Regards,
Gambito.
> Why do you believe that Top Gm's cannot win to other engines with lower ratings than Rybka under tournament time controls?
Of course. They can!
GM's have no problem to draw.
And against weaker opponents than Rybka they have no problem to win.
If you give a tactical program to a GM like Nimzo or Fritz and if you give him a 500MHZ PC, the GM would play much more succesful against Rybka.
After many games, the GM (with Nimzo help) would score more than 50% against Rybka.
With some help, of course, maybe they can. But honestly, in my opinion it doesn't have much sense to play this way.
Has not always been the chess game something merely individual when we are playing?
Regards,
Gambito.
Has not always been the chess game something merely individual when we are playing?
Regards,
Gambito.
>After many games, the GM (with Nimzo help) would score more than 50% against Rybka.
How do you know it?
>>After many games, the GM (with Nimzo help) would score more than 50% against Rybka.
>How do you know it?
Because of my intuition. GM's have much better undertanding of chess, but programms are much better in calculating (short and middle-short) tactics.
When the GM would decide not to take the initiative and if he plays defensive and closed with blocking pawn, no programm starts to grab the initiative. -> Draw by force
With enoght time to think (40 moves in 2 hours or more, better: 1 game per day) and with the help of a little nimzo-tactic-helper a GM would win against rybka in a long tournament.
That's just my three cents.
![[il]](/mwf/v2/flags/il.png "Israel")
I disagree.
I think that the main advantage of rybka relative to GM's is not tactics but better positional moves.
The question if better positional moves happen because of search or because of better evaluation is not important and avoiding tactical mistakes is not going to save GM's from loss against rybka.
Uri
I think that the main advantage of rybka relative to GM's is not tactics but better positional moves.
The question if better positional moves happen because of search or because of better evaluation is not important and avoiding tactical mistakes is not going to save GM's from loss against rybka.
Uri
maybe,
try it out. Give him enough time and give him nimzo on a slow hardware. we will see who is right.
For me, this would be a very interesting experiment.
try it out. Give him enough time and give him nimzo on a slow hardware. we will see who is right.
For me, this would be a very interesting experiment.
![[hu]](/mwf/v2/flags/hu.png "Hungary")
Actually in computer chess, beyond a certain point, tactics are much less important than positional play.
Vas
Vas
Hi,
I agree with one of your six conclusions: #4. We are planning to work on this. Already, the latest contempt is quite good, and the chances of a strong GM guiding the game into a position where Rybka doesn't have serious winning chances are quite low, but it can on occasion happen. You can find two such examples in the last Ehlvest match, games #2 and #5.
Vas
I agree with one of your six conclusions: #4. We are planning to work on this. Already, the latest contempt is quite good, and the chances of a strong GM guiding the game into a position where Rybka doesn't have serious winning chances are quite low, but it can on occasion happen. You can find two such examples in the last Ehlvest match, games #2 and #5.
Vas
Vas,
Are you saying for the latest Ehlvest match that contempt was set to something other than the default? If so, what value was used?
Thanks,
Roy
Are you saying for the latest Ehlvest match that contempt was set to something other than the default? If so, what value was used?
Thanks,
Roy
Contempt was completely redefined for this match version; instead of merely trying to avoid repetitions, it now means trying to avoid exchanges when set. The value used for the match was 35, which for technical reasons roughly corresponds to 25 on the commercial version.
Hi,
#4 ist very important. That makes chess more interesting and prevents chess from beeing too drawish.
I felt almost asleep while watching Game#2 and Game#4 (last Ehlvest-Rybka TM)
In general, i like Rybka! And Rybka is really the best! I hope you stay on improving it for many decades!
regards Dom
#4 ist very important. That makes chess more interesting and prevents chess from beeing too drawish.
I felt almost asleep while watching Game#2 and Game#4 (last Ehlvest-Rybka TM)
In general, i like Rybka! And Rybka is really the best! I hope you stay on improving it for many decades!
regards Dom
Topic Rybka Support & Discussion / Rybka Discussion / Playing strength of Rybka and GM. Comparison (3284 hits)
Powered by mwForum 2.22.1 © 1999-2010 Markus Wichitill