Rybka Fritz and future

  

Back to forum


Thibault de Vassal    (2007-02-10)
Rybka, Fritz and future...

Computerchess is definitely an exciting challenge... The community is fast-growing, new versions of chess engines appear every day, many dream to be the next Vasik Rajlich and to produce an engine that would beat the well-known Chessbase engines and the famous Rybka.

These days, I had a look at Fruit 2.1, TogaII and Crafty source code that are available to download, and started to implement new search & evaluation functions. It's quite easy to understand why chess programming is so addictive, so much done and so much to do... finally I did not enter this mad race without an ending, probably for the same reasons Anthony Cozzie (the author of Zap! Chess Zanzibar) and many others retired.

However here are my feelings about future of chess engines, and the fight that just started between most probably Chessbase engines (Fritz, Shredder, Junior and Hiarcs) and a new era of chess engines that started with Rybka...


First, it's quite obvious to me that Rybka (now Rybka 2.3) is only another one of a long series of chess engines always stronger than each others ! .. I expect the next ones to reach 50, 100 then 200 points more (and maybe more) on the next chess engines elo rating lists, a scale that definitely can't be compared to human elo rating list ! .. Several reasons to this : (1) Chess engines are human killers at standard time controls, but chess engines are far to play perfect chess yet. (2) The way ratings are calculated.

Rybka taught us several things IMO :

- Algorithms and evaluation functions are no more enough. Now chess engines have to play chess, not only search a tree of chess positions... That's probably what Rybka brought to computerchess. Since Fruit 2.1 & Toga II source code is available, and computerchess community is constantly discussing improvements in algorithms, evaluations of positions and new ideas, to implement a chess engine becomes easier so I have no doubt that new very strong chess engines like Rybka will come.

- To become famous, a chess engine must 'also' beat his rivals. I first thought that Rybka was designed to be an engines killer only (at least before to be an analysis tool) with some tricks exploiting most engines weaknesses. No, Rybka is also a great UCI engine, simply stronger and with many options & features. Like Vasik Rajlich, who is engineer and international chess master, you'll have not only to think like an engineer to create such an engine. However I still don't think it is the best analysis tool for correspondence chess, it doesn't play really better chess and in all cases it is not enough. More, Rybka 3, 4, 5 shouldn't influence correspondence chess (maybe even human vs. machine) much... Computerchess influences computerchess first.


It's written sometimes that the strongest chess engines could reach a IM, even GM level at correspondence chess. I definitely disagree with that, at least for the moment (it will take a long time yet), but as chess engines results tend to approach correspondence chess ones (means more and more draws), I do think chess engines have much to learn from correspondence chess players way of thinking, meaning : A more psychological approach, bonus for traps detection. Evaluate moves, not only positions. A more complex search, not 'only' iterative (brute force is definitely useless). No more anti-human style, speculative moves (=weakness, ie. Deep Junior) for speculative results against strongest chess engines, draws are prefered. To avoid positions not understood by the engine. Longer games, closed games (if supported)... Opening books should look like correspondence chess GMs ones (of course according to the engine's style of play) and no more been made of FIDE GM games. A better time management... Future of computerGo may teach to computerchess about some evaluations.

A chess engine must play good moves AND try to win (which is not always the same). It seems Fruit & Rybka play solid and are waiting to exploit their opponent's weaknesses thanks to a better "chess" algorithm/knowledge. As far as I have seen, Shredder & Fritz still have the best 'eye', they see far but fuzzy. Quite the same about Fruit & Toga developped by a great engineer, Fabien Letouzey : Less chess knowledge but an improved algorithm. As for Rybka, a great chess knowledge and probably a smarter algorithm (not better, smarter !) were probably enough already. The future best chess engines will be made by good chess players...

An interesting point is it could be not so easy, maybe even nonsense, to create the best analysis tool that would also obtain the best results against other chess engines. My first prediction is Rybka won't be the top rated chess engine ever, hundreds of new ideas will appear in all parts of chess programming, slowly breaking Rybka secrets, then speed will be a factor again. Deep Fritz, Junior, Fruit or Hydra are most probably the core of the next generations of chess engines... but there's a lot of work yet :)

My two cents.


Wayne Lowrance    (2007-02-15 22:50:39)
Rybka, Fritz and future

I would like to comment. Your post is most interesting. I sort of disagree with you a little bit. But first: I do agree with you that Rybka may not be the best CC engine. I sort of like F10 for that. Rybka end game is lacking behind several other engines, including Fritz. Perhaps the best cc engine is a dark horse. named Zap, latest version. Rybka kills Zap according to posting in short games (Blitz) but recent testing evidence by reliable testers indicate that it is one heck of a engine in standard time controls. In my opinion there are several engines that have proven that Humans have lost a grip in chess play.


Wayne Lowrance    (2007-02-15 23:05:38)
more chess engine talk.

Thibault, you miss the boat on Hydra futre expectations in my opinion. Its advantage over pc engines was dedicated hardware (no necessarily speed) and ease of making program modifications. However you perhaps neglect to consider the tremendous improvement in PC performance multiple cores, processors and et all. My thought is that the pc programs already are superior to Hydra. Correct me if I am wrong, but I seem to recall that Rybka has finished ahead of it in tournament play. As far as other programs, you did not mention Zap. You best keep an eye on this one. It is very very strong and improving. Right now it is the only engine that has a chance of catching Rybka in eng-eng matches. I think it will be number two on the computer rankings. I will try to look further into Zap for a top CC engine. we see. again, my thoughts Wayne


Thibault de Vassal    (2007-02-16 00:24:16)
Hydra, multi processors

Hi Wayne.

I think you're right about Hydra and actually it's quite unfair to compare it to other engines. I don't know if such an engine using dedicated hardware still has an interesting future against multi-processors engines. Quite hard to say. And what about the Chessbrain project that could use thousands of computers - http://www.chessbrain.net

About Zap!Chess, as Anthony Cozzie retired from computer chess, I don't think we'll hear about this one during a long time and that's a pity. The race just started !


Wayne Lowrance    (2007-02-16 01:29:39)
Hydra, Multi processors

Tribault. I thought he was active. What then is the Zanzibar zap i am hearing so much about ? BestWayne


Thibault de Vassal    (2007-02-16 01:43:46)
Zap! Chess Zanzibar

Zap!Chess Zanzibar is (theorically) the very last version of the engine. Here is the announcement by Anthony Cozzie on Talkchess forum :

http://216.25.93.108/forum/viewtopic.php?topic_view=threads&p=91793&t=10956