Will the new generation of grandmasters produce a human player that
will play at the same level (or better) as Rybka at OTB play ( e.g., "classical
time controls" or something like 40 in 2 ) ?

Even though Carlsen is relatively young and there is a big new crop of
rising talent, it's quite unlikely. The main reason is that the effective
strength of Rybka (on a "very nice" computer) must almost surely be above
2850 Elo. Of course, there is some possibility that a "chess genius"
without equal will emerge in the future and as mentioned, the very best
human players are getting stronger.

Is anyone "well-qualified" or knows anybody that has a good estimate of
Rybka's strength (at OTB time controls) ?

I think it would be entertaining ( maybe not for some human participants!) to
look at the games in a double round-robin including Rybka, one other very
strong "bot" and the very best grandmasters.

One minor point: I think that the next WC should request a super-long time control if he decides to take on Rybka. I don't understand why Kasparov didn't ask for a longer-than-usual time control against Deep Blue.

IMHO, for a human to be favored, he would need to play at about a move a
day and have access to any resources whatsoever. Even at 40 in 3 hours,
I don't think humans have much of a chance even if there are adjournments
every 6 hours (so that fatigue isn't a factor).

By "access to any resources whatsoever", are you including chess engines? Because a 2500 player would destroy Rybka if he had access to a decent engine for blunder-checking purposes.

At a day a move, if the WC could even move the pieces around freely he would easily crush Rybka. Engines suck at correspondence chess.

If the resources include a supercomputer with a copy of Rybka installed, the human should at least be even money.

second!

That's what I was thinking...

Seriously though, if you gave the WC 24 hours per move and let him move the pieces around Rybka would have zero chance.

Just wanted to point out, there is a limit to how long you can make the time control to maximize the human player's chances. For example at 1 move/year, the human has the disadvantage of possibly dying if the game stretches out too long.

Damn, and here I was thinking one move per leap year...

Maybe change "2500" to "2600" and you're likely correct.

And yes, IMHO, "bots" (without human aid!) have little chance at
"correspondence chess" where humans have access to "bots".

I'm not really interested in that for this post, but rather how would Rybka
perform at a "high-level" (say, highest category ever) round-robin
tournament? IMHO, Rybka's "performance rating" would be expected to
exceed 2850.

Look man, I don't want to be offensive, but you talk nonsense in both this and Carlsen thread. "Destroy"? Are you damn serious? a 2500 player with a help of a decent engine could draw a couple games maybe. Maybe. In a match, they would stand NO chance. It's funny how you think that chess engines, let's say Rybka, are all about calculating moves. It's not, at least not anymore! Programers were always looking (and found) ways to make bots think more "humanly", give them bigger "understanding" (or to put it in other way, write algorythms that bots would understand) of chess. And they definitely succeeded. And they will succeed even further, and that will be way quicker than development of human chess talent.

Also, I am quite positive that I had read in Chessbase something about Rybka, and I am positive that when they were talking about it's strength ELO wise, the first number was 3, not a 2.. However, I cannot find the article, so that might not be correct. Adams thought that Hydra played at around 3000 level already in 2005!

Rybka vs. Milov

This is an article detailing a match GM Vadim Milov (2705 FIDE, #28 in the world) played against Rybka 3. Time controls were 90 minutes + 30 seconds. First two games Milov had white. Second two games Rybka gave pawn + move odds. Last four games Rybka gave exchange odds. Results:

Rybka as black, regular chess: 1.5/2 (+1=1)
Rybka as black with no f7 pawn: 0.5/2 (=1-1)
Rybka as white with no a1 rook, Milov w/o his b8 knight: 1.5/4 (=3-1)
Rybka's (VERY) roughly estimated performance rating for the match: 3025

How in the world are they giving that performance rating? Winning one game out of 8 against a 2700 doesn't seem deserving a 3000 point performance rating even if he was giving odds (and the only win came from a non-handicap game!). What if he won all 8 games, would his performance be like 5000? lol

Either way, you do know of a place to get a copy of the games? It'd be interesting to see how Milov played the exchange up games. I can't imagine that that's "that" (3 thats in a row... english's just that awesome) huge an edge for black, given that white can play the opening accordingly.

The games are linked mid-article. Here they are.

As for how they estimated the performance rating, it's explained fully in the last paragraph, but essentially they used what limited databases they had of handicap games (mostly not at standard time controls, hence the roughness of the estimate), and estimated the value of the handcaps in ELO terms as 50 for white only, 450 for pawn+move, and 500 for exchange. They added these numbers to Mirov's actual rating for the base opponent strength in each game and used those numbers to calculate the performance rating for the match.

Starting the game up the exchange is ridiculously huge, I would be a favorite against any human player in the world in a slow time control IMO. Although I guess this makes sense since it's supposedly valued to be worth about 500 elo points.

Huh? Rybka is better than Rybka + Human? That's amazing.
In fact, I would even say that with a little preparation a 2000 player + engine should destroy the engine.

So you would expect to do better (if you were handicapping) giving your opponent pawn and move, than exchange (with you having white)? That seems unusual to me. The very few results of the match also seem to imply that the exchange is a "lesser" handicap, even if it is 2 material points versus 1.

Correction, I misread slightly. It was pawn+move that was 500, exchange was "only" 450. It's very close, both handicaps are worth "about 1.5 pawns overall", but pawn+move is the slightly larger handicap.

I'm still kind of critical of those huge numbers. Again going by practical results, the highest level odds match I know of was Kasparov - Chapman in 2001. Chapman was a decent player 2200-2300, and Kasparov took the match down 2.5-1.5 with a two pawn handicap! And this is after chapman prepared hardcore for 8 months prior to match including hiring some big guns on his team like John Nunn to help with preparation. And two pawns is surely much worse than two times the handicap than 1 pawn is.

According to these huge handicap numbers, supposedly Chapman should have been competitive against Kasparov with just a single pawn handicap! I think it's fair to say that's definitely not the case. Heck this also implies that a 2000 would be competitive against a strong IM given just pawn+move, or exchange and black. I really don't think that's likely.

Not necessarily. The ELO value of a handicap increases as the playing strength of the weaker player increases. When you give a 2700 player pawn+move odds, it's worth ~500, and is the equivalent of playing a 3200 straight up. Giving the same odds to a 2200-2300 would probably only be worth 300-400 or so (number pulled out of a hat, but the point is it would be less than 500), so with pawn+move Chapman still would have been an underdog to Kasparov.

Ok, yeah. That makes more sense. I was not explicitly considering the fact that he was 2700 getting the handicap. But I don't know. Somehow something still feels a little bit off saying +1 =5 -2 with the only win coming from a no handicap game (=4 -2 in handicap games) is a 3000+ performance.

On the subject of handicaps, I remember a thread about how important opening knowledge was in chess (in SMP I think).

There was a match between Rybka 2 and Ehlvest (2643) back in 2007 where Rybka had 45 mins to Ehlvest's 90 mins, Rybka had a 3 move opening book only, no endgame tablebase, black in every game, and had to run on a pathetically weak PC.

The score was 4.5 - 1.5 to Rybka (3 wins 3 draws). Not sure how many points you would rate that handicap at, but I think this is a better way to assess Rybka's playing ability since the games were still 'actual chess'.

I think the Milov match was set up because Kaufman and Rybka's other backers couldn't find any humans willing to play for a prize pool against Rybka without material odds.

btw I sort of agree that the exchange handicap seems better than pawn and move, but perhaps it's illusory.

I'm no master, but I'd prefer to have White and be down the exchange against a far weaker opponent than have Black and be down a pawn. Against a far weaker opponent I'm looking for middlegame tactics anyway; if I win, the exchange should be irrelevant.

"Performance rating" is a somewhat nebulous concept in general. Performance rating over a short match or tournament can vary wildly from "actual playing strength". Even if this is accurate, it certainly doesn't prove anything about Rybka's playing strength.

That being said, the only nebulous part is estimating the value of the handicaps. If 50, 450, and 500 are valid, Rybka actually performed BETTER in the handicap games. Playing as black, Rybka scored 1.5/2, or 75%. By definition in an Elo system, scoring 75% means you are 193 rating points better than your opponent, so if we accept the premise that Milov with white is a 2755 (50 point handicap) then Rybka's performance rating for those two games was 2948. In the handicap games, Rybka scored 2/6, or 33%, which by definition is a performance of -125 points. If we accept the premise that Milov with the handicap is 3155-3205 (450-500 point handicap), then Rybka's performance there is over 3000 (3046, to be precise). Here are full Elo percentage score to rating difference tables. Feel free to double check the math if you're still unsatisfied.

Now of course, it might just be variance and Rybka may have overperformed (a score over 3000 isn't unheard of for humans over a short stretch either. See Ivanchuk's 3537 performance rating through the first five games of MTel Sofia 2008.). Additionally, the handicaps may be inaccurately valued here, the numbers are very rough. Is Rybka actually a 3000+ caliber player, expected to score 80%+ against a world champion human? I don't know. But the performance rating here is valid, as long as you accept the assumed handicap values.

One thing you aren't taking into account here is that Kasparov, down two pawns, will try from move one to complicate the position and make it double-edged, to steer it towards a position that gives Chapman a chance to make a mistake. Kasparov is better than possibly anyone in the world at doing that, while the computer is much worse. The computer doesn't see much difference between being two pawns down in a complicated middlegame versus being two pawns down in a clearcut endgame, it evaluates them both close to -2.