Engine with most human-like playing style

[fern]

Well, for me what's matter is if I get fun or not. I prefer to be beaten by an enterprising, sacrificing, speculative program than by a cold calculator that stick to one pawn advantage and drags me to a boring, yawning ending.

I play chess for fun, not for blood regards
Fernando

[ArmyBridge]

Good Question!!
Personally I think that all the programs that are among the best ones (top 10) Hiarcs, Zap, Junior Fritz of course Rybka can play some Style Human games, I have seen shredder to play the sicilian game with white extraordinarily, Zanzibar whit black play very fine too, just like GM so it depends which kind of position or opening you put the engine to play. another good example is junior that has an algorithm that allows him to award the dynamic positions and to make very human sacrifices, and Rybka positional sacrifices ,they are worthy to compare itself to that towards the great petrosian , bye the way I think Rybka , Capablanca, Karpov and Petrosian they have similar styles, Junior , Hiarcs, Fritz, Toga they belong to aggressive styles but, as Kasparov, Alekhine, Morphy etc

[Marc MP]

Ultimately what matters is did the engine win the game or not, isn't it? So to me the engine with the most human-like playing style is Rybka. Smart humans should take a look at Rybka, and see what they can do to change their playing style to be more Rybka-like.

Seems to me that instead of praising engines for being human-like, we should be praising humans for being more engine-like, if that's the trait that ultimately helps you win the game.

I read, not so long ago, a post by Vasik Rajlich saying that Rybka's positional understanding was about the same as the one of a 2000 ELO player would be.

I'm not completely sure, but I think that Vas said something slightly different: That Rybka's static evaluation (="positional understanding") is at the level of a 2000 Elo player. He also said that Rybka's search algorithm is at the level of a 1700 Elo player. But this does not necessarily mean that Rybka's (positional) play is only at a 2000 Elo level. What happens is that the high number of positions searched and evaluated makes Rybka's positional decisions much, much stronger, even when those positions were chosen relatively ineffectively and evaluated without high precision.

Another way to explain my point: The 2000 Elo estimate is Rybka's evaluation quality at depth 1. However, at depth 15, it's a completely different story.

If you look at closed positions, say from the King Indian, you'll find many positions where Rybka (as many other engine) doesn't have a clue about what to play. In many opposite wing castling position, I notice that Rybka (as many other engine again), rarely think about throwing its pawns at the opponent king; it doesn't seems to understand this concept very well. There are tons of similar examples.

If Kramnik would be allowed to consult an engine like SlowChess 2.1, (to verify the tactics), he would win a match against Rybka at a slow time control. Best correspondence chess players are clearly better than any chess software right now. If engines are stronger than humans over the board, it is because of their tactical abilities, not their positional comprehension.

I have sincere doubts about Kramnik being able to beat a pure Rybka in that scenario. First, as explained above, Rybka's positional play is at a much higher level than 2000. Secondly, results in the Freestyle events do not indicate that GMs can easily outplay pure Rybkas just by having their analysis verified tactically.

Thank you for your reply,

After a second though I absolutely agree with your suggestion that 2000 elo is for depth=1 and that with search at higher depth, the result is much higher elo-equivalent positional understanding. However we must not forget that this higher elo-equivalent positional understanding comes from the ability to combine tactical and (relatively) simple positional threats with great efficiency.

For example, if engine X (not to focus only on Rybka here) understands double pawns but not outside passed pawn there are good chances that the engine will miss an easy good positional move (from the human point of view) unless the consequences are reached within the search. That is why one shouldn't neccessarly try to imitate the engine style of play even if the engine beats the human. (That is an answer to the original poster, not to you. I was just clarifying my mind about it)

As for Kramnik & SlowChess vs Rybka that would actually be interesting to see and this was just a conjecture from me. Don't forget Freestyle events are played at a somewhat fast time control (something like 50m + a small sec increment). I was thinking at something much slower. Of course, better than Kramnik & SlowChess, would be Kramnik positional understanding and Rybka serach! I remember reading (was it from the same place as before?), that V. Rajlich said he would know how to improve Rybka by 400 elo if he were given the resources.