ChessBits is an independent journal (in German) completely related to computer chess. Insider of the computer chess scene write detailed reports on the newest chess software and the latest computer chess events.

Click on the picture to visit the ChessBits pages.

MK: A very important question for our readers: when will you be able to release rebel for windows and in which form?

ES: Rebel for Windows is fully in progress. Its base is REBEL-TIGER and many REBEL-DOS functions are ported. Besides of that a new type of EOC database will see the light which we estimate will please many computer-computer lovers. I can not say much about this as we keep this project secret until release date. The search part of the Rebel chess engine is currently rewritten from scratch resulting in a much faster search as especially the last 2-3 years search was neglected because I mainly have focussed to tune Rebel's evaluation (the thing I like to do most) to perform as best as possible against strong humans (IM/GM).

I can not complain as results in the GM challenge were simply good. Also in Israel during the Israel League 1999-2000 season Rebel Century did well and ended as highest rated computer program.

So now Rebel's evaluation is tuned in a reasonable way I now focus on search. It's a part I do not like by nature as I believe chess in the end will be solved by chess knowledge and not by search but it is crystal clear that faster search (is deeper depths) is not only good for better tactics but also is responsible for better positional play which is a strong motivation for me to put my teeth for one full year in search and search only and if needed more.


MK: Much people are asking for a engine implementation with all the features as under fritz. is there something planned?

ES: In REBEL-TIGER you already can run all Winboard compatible engines. For the future of course REBEL and TIGER are the 2 standard programs as native REBEL engine but we are also talking with some other programmers to make their engine REBEL compatible.


MK: with chesstiger/chess partner you have a new line in your product family. do you be satisfied with the user acceptance and what do we have to expect in the future?

ES: There have been a few complaints on our first Windows release. The main complaint was that many typical Rebel (DOS) functions were not present yet in the Windows version. We knew this was going to happen on beforehand but it is no big deal as this happens to everybody who switches to another platform as history has proven. It simply takes time to port (re-program) and to develop new unique features. When you read the criticism of the chess press on first Windows releases it is always bad, I don't think there is a need to name examples as this criticism was directed to all. In this respect the criticism on REBEL-TIGER was pretty mild in comparison with other reviews I have read.

I don't know what causes this criticism on first versions other than that "everything is brand-new" and people have to get used to the new interface. It is also a fixed pattern that criticism fade away with newer versions.


MK: beside the gui improvements you have also to work on the engine. you were not satisfied with the comp-comp results of rebel century so you have launched the update 1.2. what are the differences, where are the improvements?

ES: A bit is already explained above. To compete in computer-computer you are obliged to spend a great deal of your time to improve search and this part has been neglected for 2-3 years because my main goal was to tune REBEL to play against humans. You don't need to have a very deep search against humans, of course it helps no doubt, but against humans the playing style is decisive and not a deep(er) search. In computer-computer it is the other way around and the program with the deeper depth usually wins.

The update Rebel Century 1.2 usually hits 1.5 to 2 plies deeper than the orginal Rebel Century 1.0 from cdrom. Unfortunately in the Israeli League updates were not allowed so Century 1.2 could not play and the slow Rebel Century 1.0 had to play its 8 tournament games. In the end the slow Century 1.0 still ended as highest rated computer and I honestly believe that the fast Century 1.2 would not have done so much better. Maybe 1/2 point more but that's about the maximum. Playing style is decisive in "Man versus Machine" and not deeper depths.


MK: A short time before you said to us that you have found further improvements for 1.2. this deep changes should be the foundation for the new version in autumn. how will this change the practical game? do you want to modify rebel to a fast searcher?

ES: NEVER is my life Rebel will become a "fast searcher" in the sense of the word as most people read it, that is "fast & dumb". Fast search yes, but I will never compromize that a fast search will make Rebel positional weaker. Fast search and chess knowledge go hand in hand, the first needs the latter, the latter can't do without the first. Of course you can speed-up search even more by removing chess knowledge but to that I will never compromize.

I don't know about autumn, that has to be seen. For the moment the engine and GUI are in full development and I have no idea where it will end until I am satisfied. Satisfied in the sense that each new version should be worthy to shoot on a cdrom as after all you are asking money for it so it has to be good.


MK: You always bet on chess-knowledge. now you are changing. why?

ES: I am not changing views. Deeper depths guarantee a) better tactics but most off all b) better quality positional moves. Actually I am catching 3 flies in one as improving search will make Rebel a better computer-computer fighter too. Now that Rebel evaluation has been tuned in a reasonable way I can afford myself to spend at least one whole year to work on search only without compromizing one bit to Rebel's playing style and chess knowledge.


MK: in comp-comp rebel is remarkable behind the tops but in games against strong humans rebel is on the top. we have just seen it in israel.what is the difference in competing gm´s?

ES: As already said, it is called "Human like Playing Style". There are several aspects here. Firstly I like to mention that in comp-comp you (very often!) can afford one or two positional mistakes and still can draw or even worse win. Try this against a GM on tournament level, one positional mistake is enough to lose the game, a GM simply will not let you escape. This rule dominates in human-comp and in comp-comp you most of the time can afford multiple (minor) positional mistakes.

Also in human-comp "strategy" is dominant. The chess program who understands "strategy" best will have the best chances. In comp-comp strategy is not dominant at all.

The last thing I consider very important against GM's is that a chess program must be able to conquer the initiative. If you replay many computer-GM games you will notice the GM almost without exception gets the initiative as I have seen so many times at the AEGON tournaments. In this respect I like to point to Rebel's last game in the GM challenge against Smyslov. It was Smyslov who got the initiative and Rebel was lucky to get away with the draw. I have been working on this very hard resulting in the "anti-GM" algorithm which simply tries to avoid the typical weak points of a chess program. It worked several times (Anand-Rebel comes to mind), it did not work against Smyslov. But a chess program must be able to conquer the initiative otherwise it is lost to defend and defend each game it plays against a GM or to win a single game because of a blunder of a GM.


MK: many users are upset because in 1.2 the autoplayer is disabled. 1.2 was designed for the battle against other programs but it can´t fight because ypu have disabled the autoplayer. nobody understand this behavior. Why have you done this?

ES: It's my job to release the STRONGEST Rebel. I have disabled the autoplayer because the "external auto232 driver" hurts Rebel's playing strength. In the meantime I have overwelming evidence for that. Just recently I also noticed that if you have the dicipline to first "power off" your PC before starting an auto232 match results suddenly become better. I also noticed that after you have exit the autoplayer and you can't launch another (random chosen) program (due to a crash) the match result is unreliable. I must have a closer look at this and if there is a solution I will certainly set the autoplayer free so everbody can use it again.

Anyway the problem will disappear by itself moving to Windows as the autoplayer works just fine in REBEL-TIGER.


MK: Chrilly Donninger (author of auto232) said it would be no problem for you to develope your own rebel-autoplayer in a short time. why didn´t you do that? it would blow away all speculations immediately.

ES: To develop an autoplayer will cost about 1-2 weeks programming. To test it 2-3 months if you want the software absolutely trouble free. I don't want to spend so much time on a piece of DOS software while things are already running without problems under Windows. Besides of that it is a matter of priorities as only 1% of the chess computer lovers use auto232.


MK: the progress in playing strength is impressingly these days. how long will it continue and where will it end? shay´s (author of Junior) opinion is that in 10 years no human will be able to win a 10 game match.

ES: I agree with the opinion that in 10 years Kasparov will lose against a normal PC chess program. Maybe it is Rebel, that would be a dream. With a well balanced mixture of chess knowledge and search Rebel nowadays hits 2550 elo on tournament time control. During the last AEGON tournament (1997) most people estimated chess programs under or around 2400. In 3 years that is a gain of at least 150 elo. I believe this will be a steady process with as last station Garry Kasparov or the then current best player in the world.

I am aware I have said the opposite in the past and that the maximum for a computer would be 2600 elo. How untrue as Rebel already has proven to be at 2550 after 30-40 tournament games. My opinion is also based on the fact that humans have weaknesses computers don't have (overlooking small things) and the fact that within 10 years computers typically will hit 15-17 plies in the middle game and I estimate that is too much to handle even when your name is Garry Kasparov.


MK: nearly all top-programmers are using the tablebases to make their programs stronger. only christophe (author of Chess Tiger) and you are not willing to implement the tb. the users cannot understand that. everyone wants that but you refuse to put it into rebel. also out of marketing reasons rebel should have use the tb from the beginning. now you are in danger that people are seeing rebel as an obsolete program because it also is running under dos and so easy but very important procedures as clip and paste with other applications are not possible.

ES: This is total nonsense. Table Bases in their current state are not more worth than 5-20 elo points and I am in agreement with my collegue Christophe Theron here. Don't forget that Table Bases are stored on hard disk and that thousands and thousands of times during search the heads of the hard disk must move to the right position in order to receive the score for the position it is ordered to find by the search. This slows down the search tremendously resulting in a loss of 2-3 plies or more very often.

Besides of that there are only few people (< 1%) who are willing to spend 2-3 giga bytes of their hard disk to install all the Table Bases. I realize that among the ChessBits readers the 1% is probably much higher. I am not anti Table Bases, on the contrarary. I predict a great future for Table Bases as soon as the most common 6 and 7 man Table Bases are available as then the gain suddenly increases to 100-150 elo. Unfortunately this is not going to happen within the next 4-5 years. But then it makes sense to invest in a big hard disk for a select group of people but for today I would say that Table Bases are a nice thing to have provided you have enough hard disk space. Nothing to be excited about.

Of course Table Bases will be programmed in Rebel and Chess Tiger that is self-undertood but for the moment Christophe and I see other promising areas which will gain more than just 5-20 elo points.


MK: you said with tb the programs are gaining maybe 5 elopoints. in this issue we have made a research and after 720 games it was evident that the main difference is about 50 points. it ís hardly possible to win as much points in a faster and more easy way. what do you say to that?

ES: Its explanation is simple. When a programmer adds Table Bases to his program he/she "removes" all existing chess knowledge which you now going to retrieve from the Table Bases as you don't like to waste valuable processor time doing things twice. As a result such testing is based on a program that absolutely knows nothing from standard endings while such testing should be based on a program that has all the chess knowledge left. Only then you will find the truth of the real gain of Table Bases.


MK: will you compete in london with rebel or tiger?

ES: Chess Tiger will participate in any case. Rebel is unclear. Much will depend on the demand of David Levy that authors must be present in London.


MK: please tell us something about the idea exchange with christophe. how is this to imagine? what have you realized and what do expect for the future to realize?

ES: One good thing Christophe convinced me about, and this interview is full of it, is to improve on search. He has given me plenty of new idea's. On the other hand I have told Christophe the main secrets of Rebel's evaluation, so be aware for the new Tiger.

When Rebel has moved to Windows we are going to work to program a "multi-engine-search". That is that both Rebel and Chess Tiger are calculating the current position and that a kind of "referee" is going to decide which move (Rebel or Tiger) is going to be played, so a kind of 2-Hirn system. Next step is of course multi-processing and then Table Bases.

I like to thank you Marcus for this interview and I hope the ChessBits readers have enjoyed it reading, for me it was fun to write it.

Ed Schroder

Deventer June 5, 2000