|
Introduction to Computer Chess, technical articles on Computer Chess, how does a chess program work, that kind of techinical stuff.
INTRODUCTION
A chess program is software, just software like any other of the hundreds of (software) programs you have on your PC. A program is a set of instructions the computer understands and executes.
If you can teach a computer the rules of bookkeeping so can you teach a computer the rules of chess, there is nothing mysterious on that.
A chess program is divided into 3 sections (modules):
The rules of Chess
- To play chess the program needs to understand the rules of chess, so this module contains all the rules, how knights move, the castling rules, the pawn promotion rules, stalemate, checkmate and so on. The module will return a list of all legal moves, for the start position the list will contain 20 moves.
- When this part is ready you already have a program that can play legally chess, just pick a (random) move from the list of all legal moves and move it to the screen, wait for the opponent response and repeat the process.
- So that's the first part, the most easiest part. All it takes is a reasonable programmer, knowing the chess rules and a couple of months spare time, you don't even need to be a chess player to make a chess program.
- However, a chess program that just plays random moves isn't going to be very successful, in fact it will lose 99.99% of its games against any ordinary novel chess player, what it needs is intelligence and that's what's the next 2 parts are all about, Search and Chess Knowledge.
The Search Algorithm
The Search Algorithm emulates the look ahead process human chess players do when it's their turn to move. Have a look at the diagram, with black to move it looks very attractive to capture the rook playing 1..Nc2xa1, however any reasonable chess player will not do that because he will find himself checkmated the next move with 2.Qf3xf7#.
That's what a human chess player does, while considering his own move(s) he lookes ahead what his oppenent might do, that's the mechanism. The great chess players of our time and those of history have one thing in common, they could look ahead many moves, sometimes up to 20 moves and foresee all the consequences.
It's exactly that where computers excel, the look ahead process, its technical name: SEARCH. Where humans, even the great players, make mistakes in their calculations (the look ahead process) computers can search flawlessly, let's move to the next diagram to demonstrate this.
In this position GM Shirov played the beautiful 1.Rxg7+!! and checkmated his opponent 14 moves later! After the game Shirov claimed that no computer could be able to find this move.
GM Shirov was wrong, a chess program with a good search algorithm can find the mate combination very easily. REBEL for instance finds the move 1.Rxg7+ in 4 seconds and announces the mate 10 seconds later, all this on a poor 850 Mhz PC.
This is certainly much faster than GM Shirov did during his game as he had to calculate the complete mate combination very carefully before playing the move.
All it needs for a chess program is to have a good search algorithm to find such beautiful tactical turns, there is no special chess knowledge needed, just knowing the legal rules and a clever search algorithm.
The Chess Knowledge
However, we are still not there. Not every position in chess is about checkmate or material gain, a chess program that plays 1.Ng1-f3 2.Nf3-g1 3.Ng1-f3 4.Nf3-g1 from the start position is candidate to lose all its games anyway despite the fact there are no direct material losses present, what the program is still missing is positional and strategic understanding of the game, that's where Chess Knowledge comes in, the heart of a chess program.
Let's consider the start position, see diagram. We know that the first module will generate for us a list of all legal moves, 20 in this case. But what is the best move from those 20? Here is how it works, for each move (of the 20) the Chess Knowledge Module is called and the module returns a number also known as the score of the position.
So all 20 moves will get a score see example below, then simply the move with the highest number is chosen as the best move and send to the screen, e2-e4 in this case, as simple as that, it's your turn to move.
e2-e4 40 d2-d4 36 Ng1-f3 30 Nb1-c3 28 c2-c4 27 e2-e3 20 d2-d3 20 c2-c3 12 g2-g3 10 b2-b3 10
a2-a3 8 a2-a4 7 b2-b4 4 h2-h3 4 f2-f3 0 f2-f4 0 Nb1-a3 -8 Ng1-h3 -8 h2-h4 -30 g2-g4 -40
These numbers, how does it work?
The Chess Knowledge Module will represent a position in a number also known as the score of the position. Let's consider how the Chess Knowledge Module will translate the position after 1.e4 into a number, the score of 40 in this case, see diagram.
The first phase will check all pieces on the board and calculate the score via a formula. For instance the value of a pawn is 100, so 100 is added to "score", the value of a knight is 300, so 300 is added to "score", making "score" 400. For the bishop the value is also 300, the rook is 500, the queen is 900, the king is irrelevant. When counting all the white pieces we get a "score" of 3900 in total.
We repeat this process for the black pieces, only that the values of the pieces are subtracted instead of added to "score". If things are done well "score" will contain the value of "0", the correct material balance of this position.
The second phase is to evaluate all the positional characteristics of the position, characteristics are: mobility of the light pieces, king safety, pawn structure, center control, pins, bishop pair, various endgame knowledge and so on, a sophisticated chess program has hundreds of characteristics. All the characteristics are (again) represented by numbers and added to "score".
For the position after 1.e4 (see diagram) mobility, pawn structure and center control look dominant, the Chess Knowledge Module will recognize those patterns and add 12 points for mobility to "score", 18 points for pawn structure and 10 points for center control, making "score" 40 in total.
The score of 40 is passed (connected) to the move 1.e4 and then the next move (of the 20) is evaluated till all 20 moves are evaluated and the move that has the highest number is considered as the best move and thus is played.
As far as I know nowadays all available chess programs work this way, others have tried different approaches such as former world champion Michael Botvinnik with his program PIONEER in the early 80's but so far the above explained skeleton still is superior.
Interview by Arvind Aaron with Ed Schröder
July 23, 1998
Interview by journalist Arvind Aaron with author of REBEL10 Ed Schröder on Ischia after the match against Vishy Anand. The
interview was published first in the sports weekly The Sportstar (India's largest sports weekly news paper) and
the chess magazine Chess Mate.
REBEL Not To Play Other Computers
By Arvind Aaron
Finally there is a specialist chess programme which works good
specifically against Humans. It is called REBEL, the Dutch
programme which shocked World No.2 Viswanathan Anand 5-3 in an
eight game match at Ischia, an island near Naples in Italy last
month.
REBEL's victory in this three day contest held July 21-23
confirms the machines' superiority in quick time chess leaving
the hope to humans in classical time controlled games only.
The programme was superior due to its human-like playing style
and the new feature "Anti-Grandmaster Style," said its bearded
48-year old programmer Ed Schroder in an exclusive interview to
Arvind Aaron.
REBEL 10 will be soon in the production line and in
the market from September-October this year.
Excerpts:
Q: Your programme REBEL won. With four blitz games of out eight,
what percentage of chance does a grandmaster have (for winning)?
A: This is a difficult question. In my opinion it depends on
which programme you are playing. One programme will do good
against humans completely different from another programme.
Speaking for my own programme, REBEL has always been on top
playing against humans. If you look at the Human versus Computer
tournament at Aegon you see in statistics from 1991-97 REBEL is
overall clearly on top playing against humans.
So, I also think that Mr. Anand has estimated REBEL in that
respect. You cannot compare REBEL and FRITZ. If you let them play
against each other, it is equal. REBEL against human is truly
stronger than REBEL against FRITZ.
Q: Why this difference?
A: Because of the playing style. If you take the REBEL, FRITZ
example, FRITZ is programmed based on speed and tactical things,
e.g., deep combinations, REBEL is of course also good in tactics
but it is positionally based. It tries to play the human style,
positional moves. That is my personal opinion. When you have a
good position tactics come by itself. So, REBEL tries to play
good positional chess like a human.
Q: How much of the victory was attributed to the powerful and
fast hardware. Would REBEL had done the same had it taken place
on a Pentium-2 400 MHz processor as was promised to Anand?
A: I think on a P-2 400 the result would have been the same. If
you go lower, then the result will be less. I think the
difference between the P-2 400 and the AMD which played here is
about 25%. It is nice to have. In Elo it is not so much.
Q: REBEL 9 had an Elo rating of 2580 which you printed on your CD
ROM. What is the estimated strength of REBEL 10?
A: The big problem is to say something about this. If you say in
computer to computer, I would say it is about 20-30 Elo points
stronger. If you use the new future of REBEL, with the anti-human
and anti-grandmaster style, then it is a major step up. Maybe
fifty of even hundred Elo points stronger.
Q: How do you define the anti-grandmaster style?
A: This is based on my experience in the Aegon tournaments
playing against grandmasters. I have played all kinds of people.
Against people rated till say 2200, REBEL always wins. Almost
without an exception.
If you go one step higher to say 2400-2450,
you see another pattern. You see that REBEL mostly wins, the
struggle is closer for the iniative. If REBEL wins the game, it
has happened because the player in that area makes a minor
positional mistake or big blunders. But in the fight for the
initiative you can see that the human player had the positional
understanding than the computer on the long term but still REBEL
mostly manages to win the games.
Then if you come to the point, it is against the grandmasters. It
is always that a grandmaster gets the initiative. It is good
strong position. Anti-grandmaster style is about having the
initiative. Don't lose it. It is important.
It will not always
play the best moves in respect to the normal REBEL. It will play
moves that will confuse the grandmaster and keep the initiative.
It is the first time for it. It has never been practised before
against humans.
If you look at all eight games, I can see it
works. Because REBEL always got chances in the games against Mr.
Anand. It wasn't locked in a position where it had no chance. But
I have seen so many times that when you play a game, after it you
say you lost to a grandmaster I had no chance at all.
In these
games, it always had chances. Also in the two tournament games,
it could play. This is anti-grandmaster. Create chances, don't
let yourself get slaughtered by a grandmaster.
Q: How do you compare this match with the one against Artur
Yusupov last year?
A: I cannot answer this because I was not present here. My co-
operator did it last year. So I do not know the atmosphere of the
games.
Q: How many people are involved in programming REBEL?
A: Me, of course and I work on the chess engine. You know Jeroen
Norman who was the operator and a chess player. He is also the
book editor. We have others who do the graphics, the database,
one for the website, one to respond to the E-mails and one for
the Windows versions. It is about seven or eight people.
Q: What is the commercial worth of this victory?
A: It is two-sided. First, it is of course promotion. The second,
as a chess addict I want to find out what is the state of art of
my programme against the top of the chess world.
Q: Are you going to sell more?
A: Yes. Of course. I have no idea. I think because of the good
result I will be able to sell ... no idea. I think what I spent
on this match I will earn it back. Apart from the money paid to
Mr. Anand to play the match which we will not disclose, our
investment was 30,000 dollars.
Q: If you had seen game four and game eight, the machine went for
material and could not assess the rook for two pieces value. Is
it easy for the programmers to eliminate this or is it habitual
for the machines to fall for this in the years ahead?
A: I can speak about the eighth game. The fourth game is not in
my head now. That was an exceptional case because REBEL had four
connected passed pawns. Normally it is worth a lot. In reality it
was worth nothing. That was the big problem for REBEL. It didn't
judge the position good. It thought four connected pawns is fine.
Normally it is. Not in this position.
This is the main
problem for chess programmes. Yesterday was such an exception. It
was kind of bad luck that it ended in this kind of position.
Q: I think this is a pattern for most computers?
A: In this game of yesterday, every computer would have lost.
Because the four connected pawns are evaluated too high. Black
can easily halt them. That was the difficult point of yesterday's
game.
Q: What was said new about REBEL in this match?
A: I think only anti-grandmaster chess.
Q: How do you hope to fare in the next world computer
championship? How did your programme do in the world PC and blitz
championships at Paris, and the overall world computer
championship in Hong Kong?
A: I am not in any mood to spend my time and money in Computer-
Computer events. Maybe in the future but for the moment no. In a
Computer-Computer match you are also commercial which means
everybody can get your programmes. I want to win against humans.
In Computer-Computer events other programmers have told me that they
have worked two months on one opening to catch me. (laughs) Of
course it works. They can open the programme and see the holes in
the REBEL opening books and shoot on that. They practise it in
tournaments. To play in Computer-Computer tournaments I have to
escape from all those book attacks and other programmers. I need
three months preparation. I need totally new lines but I don't
have that time. Besides of that, I think in Computer-Computer
chess tournaments are losing interest from people. They are not
so important any more as the days before.
Q: Why?
A: I don't know. I see it from the number of visitors in the
website. I see it in the attention it gets in the press. It is
not so attractive any more. Things like this, and REBEL playing
Yusupov, that's what people want. People believe that computers
are strong and want them to play strong grandmasters. It is a
matter of attention it gets which matters.
Q: How many man-hours have been spent on REBEL this far and how
many hours were spent on this match with Anand?
A: Incredible amount of hours. Days, months, about three-four
months on this match. Let's say four-five hours a day. On REBEL
this far is impossible to say. I need a calculator for this! Full
time is atleast eight hours a day since 1985. In the last few
years I had more people to do it and I do not spend as much time
as in the early days. I do not spend so much time on the chess
engine as I have so much in my head for other things. In the past
I only had to do the chess engine and for the other things I went
to other people. Now I have to do all the promoting stuff myself,
being in charge for the packaging, the manual and so on.
Q: Why the name "REBEL"?
A: I started programming it in 1981 as a hobby. Then there was no
literature at all. It was very difficult to write chess
programmes. Those days you had to figure out all by yourself. It
was so complicated and confusing, you programme something and it
didn't work.
I asked the programme to do something and he didn't
do it. So I said "you are a REBEL". You are rebelling
against me. These days I am the rebel and he listens to me.
Besides it is a good name. It sounds good in every country.
Q: What do you have in future for REBEL?
A: Keep on improving the engine. We decided a few years before
that we don't play Computer-Computer any more. Let's play good
chessplayers instead. I think this was a good decision. In
principle we decide to play in such an event every year. Last
year it was Mr. Yusupov, this year it was Mr. Anand and next year
some one else.
Interview by Detlev Pordzik with Ed Schröder
for Rochade Europa, 1999
a German chess magazine with 60,000 subscribers
RE: In 1982 you started with your first own program at the Dutch
Computer Championship, since 1984 you work as a full time
programmer, delivering all the above mentioned programs.
Today with an own crew of 8 specialists, from bureau over GUI
programming, engine developing to opening theory specialists -
isn't this a hard job in today's world of specialists, developing
only singular products, like databases, GUI's, ect ?
ES: Since 1984 (when I started to work for Hegener & Glaser) till now (1999) a lot of things
have been changed. In those early days I had to make one (or sometimes two) products a
year, a very convenient job as it gave me all the time in the world to improve the chess
engine. Programming the UI (user interface) was easy these days. There was just ONE
hardware, one processor, one LCD, one set on LED's all EQUAL material so no compatibility
problems.
These days we have DOS, Win 3.x, Win95, Win98. Next we have 200-300 different brand names
of Pc's. All use different materials, 400-500 different types of monitors, 1000 or more
video drivers, 300-400 printer drivers, 500 different mouse drivers, different sound cards,
different cdrom's and cdrom drivers. All are mixed and put in a mini-midi or tower model
and it is called a Pc.
If that is not complicated enough we also have laptops, different types of processors
(Intel, AMD, Cyrix...), different types of memory (Ram, Dram...). It's really one big
miracle it all seems to work, that is to say in most cases.
In the early (6502) days you developed the program on one prototype. When it was ready
the eprom was shipped for duplication and that was it. These days (since Rebel8) we sent
beta-versions to at least 25 people and we make sure that the upcoming Rebel is at least
tested on 50 different Pc's.
In the early (6502) days when the eprom was sent for duplication there were no worries. Same
hardware no compatibility problems, one could relax as the job was done. These days when we
get the cdrom's back from the duplication factory and we start all the shipments to dealers
and customers we are worried as now (although tested on 50 Pc's) the real test starts. How
is the new program doing? Is it bug-free? The first 2-3 days are crucial. If there are no
complaints we relax. We could relax after Rebel8, Rebel9 and also with Rebel10.
However Rebel7 was a real disaster as a major bug was over-looked as Rebel7 didn't run on
Pc's with some graphical adaptors. We then had to make a patched version. New dealer
replacements shipments, dealers sending the Rebel7 patch to customers, hundreds of phone
calls. At that time we decided (as first chess company) to hire beta-testers for Rebel8.
Today it's our opinion you can't release a new product without a decent beta team as a Pc
is quite a different world as a stand-alone chess computer.
RE:
If you take your time for a moment and review these years, what would you say, was the
major case among all your successes, that meant the most to you, personally ?
ES: Several. First of course the world-micro-championship in Vancouver when I got my
first world-champion title. The big surprise was certainly the Madrid event in 1992
when Rebel (then Gideon) gained the world-champion title all classes. We played on a
286 (8 Mhz) laptop with a ChessMachine card of 32 Mhz included. After each game the
ChessMachine card was taken out because the prototype was so fragile. We remembered
the sayings of our (main-frame) opponents, "have I just lost from that tiny thing?".
As third I like to mention the several first places of Rebel on the SSDF list starting
with the "Mephisto MM4 Turbokit 18 Mhz". Being on top of SSDF always gave me more
pleasure than my 2 world-titles because tournaments are just about 5-10 games and are
sometimes real lucky shots but the SSDF list is about hundreds of games and therefore
is more valuable at least in my eyes. That's why it was so sad to see the SSDF fall so
deep last year allowing unfair competition.
As fourth of course the Anand-Rebel10 event of 6 months ago. It was a real honor for me
to play against the second best player of the world with a rating close to 2800. It never
came up in my mind Rebel would be able to win this match. I will never forget the
atmosphere in the playing hall after Anand lost 3 games in just one hour. That must be
a long long time ago for Anand such a thing happened for the last time. I will also
never forget the 2 tournament games. Both were exciting. In both games Rebel10 created
chances. What more is there to wish (gain) for a chess programmer?
So if I have to pick I definitely choose the Anand-Rebel10 event. Computer - Computer
events are nice and very important but the *real* events are playing against (strong)
humans after all that's why I started in 1981 as I wanted to write a chess program
that could beat me a poor player of just 1850.
RE:
17 years are a long time, Ed - it takes quite a lot to be still on top. Once again,
looking back, beside the fact, that the data carriage changed from board computers
to PC's along with speed factors, ect - where would you locate the main difference
to today's commercial computerized, chess scene ?
ES: In the early days the most important item was the playing strength of the chess
program. This all have been changed especially the last years. On my old 486/66 machine
I still could beat Rebel but now on my PII-450 with 256 Mb I am without any chance and
must deliberately weaken Rebel not to lose all games.
Already at the time of Rebel6 we received more and more requests to add features to Rebel that would lower Rebel's playing strength. That was the world upside down. But we did as it is indeed no fun to lose all the time and then the fun playing with (and against) the computer might go away.
More changed, as (most) people couldn't win from the computer any longer people started to use a chess program in a totally different way. Instead of playing their games against the machine people want to analyze their (or grandmaster) games so a lot of analyze features have been added since then. These days people buy a chess program not only because of the playing strength (all are strong!) but decisive arguments are "user friendly", "number of features", "data" (big databases, big opening books, a big chess tree), "customer support" and so on.
Also the playing style of a chess program has become more important to people. As they analyze their games, use a chess program for their correspondence or Internet email tournaments they want to receive an intelligent (human-alike) response from their chess program. All chess program are good in tactics and are real monsters in this area and that's good for a quick blunder check of a game but what about if you need a plan for a (say) positional position? I clearly remember a remark of a customer years ago. He said, "Why buy new chess programs? They only are more better in positions they are already good in!".
That came as a real hammer. Of course this man was right and from that time on I have been focused more and more on Rebel's positional understanding and make the program play as human-alike as possible. My favorite program in this respect is Mchess. It plays very attractive human-alike chess. Close to Mchess are Rebel and Hiarcs. I consider these 3 chess programs as the best concerning the quality of returned analysis.
RE:
As far as I remember, your company was the very first to offer a registered user a special home page, where he can download quite a mass of valuable adds of different variety for the limited time of the product standing ( means : until the next release ). As I follow the growing and changes constantly - it's quite an effort. What made you decide to set up such a free page, which needs a lot of manpower - and are you satisfied with the acceptance you can figure out by hits ?
As far as I can tell we are the only chess company that offers a special download area on the Internet where Rebel10 users freely can download the latest games, opening books, chess trees, utilities and so on. It is a lot of work to maintain this download place but it's my understanding it pays off.
It has given Rebel an extra dimension and you are using the Internet for what it was created. In the meantime we have released a Rebel10.0b upgrade and it is freely downloadable for every Rebel10 customer. Next when an important (grandmaster) tournament is played we offer all the games to people. It's our understanding this is what people want and it is all so easy to distribute. Just one mouse click and you have the newest data or Rebel version.
RE:
Beside this, I think it is of common acceptance to say, that your - yet commercial - site is the one with the highest quality standard. From extra - free - tools to specialized opening books, new game collections every month - and so on + and on. Would you say, is this a very personal thing of yours to deliver all this for free - or does commercial calculating differ that much from country to country, as other Companies of your branch can't measure up in no way here :-) ?
ES: To setup and maintain such a download place of (only) high qualified stuff requires a lot of energy and man power. Since the start of the subscription area (with Rebel9) we first had to set a few things straight concerning the database updates for example a REBEL database consistency policy was developed. That is, consistent player name spelling, consistent tournament spelling, consistent abbreviations and so on.
This is entirely done by one person and the total amount of work is at least estimated at a 1000 hours a year. To answer your question, not every chess company is willing to spend so much time and man power in a free (database) service. And of course the Rebel10 subscription area is more than a database service alone.
RE:
Coming back to todays, present scene, are you satisfied with the overall engine quality of your latest child, REBEL 10 - in it's 10.b version ?
ES: Rebel10 is more stabile than Rebel9. The selective part of the engine has been improved as Rebel9 took too much risk (sometimes). It has made Rebel10 a bit slower than Rebel9 because of this but overall the changes made Rebel10 a stronger player. Also the search is rewritten from scratch with as result a more stabile Rebel. But most of the programming time was spend in the "anti-GM" option of Rebel10. As always new things cost the most time.
RE:
As internet users among our readers aren't the majority yet, please be so kind to explain the increase of the 10.b release.
ES: That is correct. People who have subscribed to Rebel10 on Internet can freely download Rebel10.0b. Besides a few bug-fixes we have added a few new features too. These are:
- Added support for the Saitek Kasparov external chess board.
- Added support for the Mephisto external chess board.
- Added an automatic game annotator while playing your games.
- Added STRONGEST SETTINGS option. This will automatically set REBEL10 on its strongest settings for maximum playing strength.
- Added SAVE ENGINE SETTINGS. On request REBEL10 will save all settings of the engine.
- Better (more precise) ELO calculation during program launch for rated games.
RE:
One of the major features of REBEL 10 is it's " Anti GM function " possibility.
In a R10 review of another German Comp. Chess Mag this function was described as "....tries to reach positions, where it can evaluate very precise + keeping up initiative......" . Ironical and not too qualified comment of the author : " Quiz - what do the other progs do.... ? " So, to end up this strange discussion : Please describe to our readers, what the aim and the special benefit of this major engine function is - and if mentionable increases could be seen in reality.
ES: "anti-GM" was born out of fear. Fear for super grandmaster Vishy Anand and the match in Italy I signed for. Having participated in all (I believe 12) AEGON man vs human tournaments I (after years) came to the following conclusion:
Rebel playing against players of 2300 elo always wins without almost any exception. Rebel is able to get the initiative, can make pressure, the human gets lost in all the complications and loses. It's almost a fixed pattern.
If Rebel plays against humans in the 2300-2500 area we see another picture. The struggle is for the initiative. If Rebel is able to get the initiative Rebel mostly wins as also here the human gets lost in all the complications and/or time pressure. If the 2300-2500 rated human gets the initiative Rebel has to defend. Defending is one of the strongest points of todays TOP chess program (not only Rebel). The top chess program are so tough, they always seems to find the best defence and are real grandmasters in narrow escapes. I have seen this so many times. The 2300-2500 human having the better position but NOT ABLE to win because of tiny positional mistakes allowing the chess program to escape from the attack. Then Rebel strikes back, gets the initiative, the human realizes he blew it and starts to make more tiny mistakes (and sometimes real blunders) and mostly the game is over very soon after that. It's a fixed pattern.
But then the real work, playing against grandmasters. Here I have seen a total other picture. In *ALL* games the grandmaster is able to win the struggle for the initiative. They *ALWAYS* get the better position. Then the above described process starts again, the human attacks and the computer defends in its usual tough style. And here we see the difference, the grandmaster hardly makes a mistake, fulfills the attack and (mostly) wins! Another fixed pattern.
So here I was, I obliged myself playing an 8 game match against world's second best player Anand with a rating of almost 2800. I considered my Rebel without any chance based on what I have seen at AEGON. So something had to be done to avoid Rebel to be slaughtered.
I decided to do something on the point where the trouble ALWAYS started. It was my opinion that the real reason Rebel (and others) lose to grandmasters because they are not able to win the struggle for the initiative. I developed a piece of software to make sure Rebel would not lose the initiative and called it "anti-GM" which is the right term for it.
If you look at the 8 games against Anand you clearly see it works. In all 8 games Rebel10 had chances. What more can a programmer wish? Nothing. The below text is from the Rebel Home Page on Internet and is from Vishy Anand himself.
[ BEGIN QUOTE ]
Vishy Anand (about game-8)
14.Qd3, this move surprised me.
14..b5
15.Qf3, I did not understand what the program want. After 15..Be7 16.Be3 0-0 17.dxc5 Bxc5 18.Bxc5 Qxc5 19.Nd7 the material gain is worthless as black has a fantastic compensation.
But then suddenly I noticed 15..Be7 16.Bg5 and my eyes became glassy as I realized that 16..Bxg5 17.Qxf7 Kd8 18.Qxg7 would have forced me to resign.
It was a shock for me that such complications suddenly can occur. You think you are playing a quiet positional game but suddenly you are in the middle of all kind of tactics.
Comment from the Rebel team
We like to thank Mr. Anand for this nice compliment. We are used to compliments but this one is certainly one of the best we have got, if not the best! It confirms our views and principals about anti-GM on which we have worked on for so many months.
[ END QUOTE ]
RE:
REBEL 10 has done extraordinary well in tournaments, lately - in which the program was operated manually. Which were the ones on major importance ?
ES:
#1. The Clodra comp-comp tournament in Germany. Here is the quote from my home page.
[ BEGIN ]
REBEL10 (in Germany) recently won the strongest computer-computer event held ever. In 14 games REBEL10 scored 10 points without losing one single game.
All participants were in agreement, that REBEL10 played the best chess and won well-earned. The tournament was played as a seven-round-swiss system tournament with three hours each side for the whole game.
Final Ranking
1. Rebel 10 10/14
2. Fritz 5.32 9,5
3. Hiarcs 7 8,5
4. Nimzo 3.5 / 99 8
5. Chessmaster 7,5
6. Zarkov 4.3 7
7. M-Chess Pro 8 7
8. Virtual Chess 7
9. Junior 5 7
10. Shredder 3 6,5
11. Chess System Tal 6,5
12. Genius 5 6,5
13. Kallisto II 6
14. Gandalf 3 5,5
15. Tasc R30 Vers.2,5 2,5
#2. Rebel10 prolonged its title in the Samanel Cup.
1 REBEL 10 400MHZ 128RAM 2555 * 1 ½ ½ 1 1 1 1 ½ 1 7.5
2 CM6000 450MHZ 128RAM 2545 0 * 1 1 ½ ½ 1 1 1 1 7.0
3 HIARCS 6.0 350MHZ 128RAM 2550 ½ 0 * ½ 1 1 1 1 1 1 7.0
4 FM JOSE ML. DOMINGUEZ 2340 ½ 0 ½ * 0 1 1 1 1 1 6.0
5 IM NELSON PINAL 2325 0 ½ 0 1 * 1 ½ ½ ½ ½ 4.5
6 NM NELSON ALVARADO 2180 0 ½ 0 0 0 * 1 1 1 1 4.5
7 IM RAMON MATEO 2470 0 0 0 0 ½ 0 * ½ 1 1 3.0
8 FM MARINO FERNANDEZ 2225 0 0 0 0 ½ 0 ½ * 1 1 3.0
9 ALEXIS MONTES DE OCA 2130 ½ 0 0 0 ½ 0 0 0 * 1 2.0
10 NM JUAN ML. JAQUEZ 2205 0 0 0 0 ½ 0 0 0 0 * 0.5
Abbreviations: FM FIDE Master
IM International Master
NM National Master
As last year REBEL10 did not lose one single game. In 9 rounds REBEL10 scored 7½ points. In the comp-comp confrontation REBEL10 beat ChessMaster 6000 and draw against Hiarcs 6.0.
#3. Rebel10 is doing well in several comp-comp tournaments of private people who publish results on the Internet.
RE:
We don't have to mention the good reasons for you, not to participate any longer in the Swedish list . Comment of a German retailer Co : " ...a prog, not participating there, simply doesn't exist......". What is your opinion on this - can this list really still be hung that high today ?
ES: A couple of points. First, I don't think that a program that exist so long and has gained so many successes needs to prove itself. On the Rebel Home Page on Internet there is a special page fully dedicated to Rebel's successes from 1991 on, and that is a shortened list mentioning only the most important high-lights. The Swedish List is simply a small part of that. The Swedish List is important but in the end it only tells you something about computer versus computer. The Swedish List doesn't give any answer how a chess program does against humans.
RE:
Nevertheless : there have been rumors lately, that you finally made up your mind, to implement the autoplayer function again - which will allow the usage in the Swedish testings. Is this true - and if, what made you finally change your mind - the respect towards your customers - or something else ?
ES: The autoplayer was removed from Rebel10 because the SSDF refused my request not to include Rebel10 in the Swedish List. This because of the unfair testing that took place last year. Because of their refusal I was forced (besides taking out the autoplayer) to put a legal notice in the license agreement that forbids them to include Rebel10 in their list. This seemed to have helped.
The (unwanted) negative side effect by taking out the auto232 software was that some loyal Rebel customers were very upset because they couldn't play their own comp-comp matches any longer. Then when I noticed the Swedish guys apparently didn't plan to violate the license agreement I promised to make the autoplayer software available again this in the good hope the Swedish guys will respect a companies express wish.
At the moment I am working to bring back the autoplayer part back in Rebel10 and include a stronger engine as well. When this is all finished we will upload this version (most probably as version Rebel10.0c) on our server free downloadable for Rebel10 customers who have subscribed.
RE:
As far as I know, the autoplayer still isn't free of strange behaviors - never seen before until last year. Or, so to say : some programs, connected to the autoplayer, show astonishing functions. One refuses to save a lost game, the other one wants a - very large opening book, which ain't content of his regular CD - if he finally gets it, he still refuses to play, as long as the book is write protected ( which is quite normal - as from CD ), another one refuses start of the game, if he can't detect endgame tables from a - totally different product. Viewing all these facts, do you think you can compete even though ?
ES: Well... the autoplayer (auto232) software is a strange fragile animal. I use it since 1994 and sometimes you see the most unusual crashes happen. Then when you exchange the very same programs to 2 other Pc's the crashes disappear. It's all very weird and unpredictable but after having played thousands of auto232 games through the years I can safely say there is no intended cheating taken place. Auto232 is a fragile protocol and crashes regularly for unknown reasons but it is my clear impression if a match crashes it crashes random. I have seen no pattern.
Concerning adding extra books, endgame table bases or whatever I have no problems as I agree with the SSDF policy that this is allowed as long it is commercial (general) available. As long as everybody can check it's all fine with me.
RE:
Keeping up the theme of the today heavily debated " pro and against " the Swedish list. What does it personally mean to you - today ? Does it still have the former value - remembering all the curiosities, that happened and still happen ?
The SSDF list lost a lot of its creditability last year. Now they seem to be on the right track again by not allowing secret autoplayers anymore.
RE:
Do you feel comfortable with the message, SSDF sends with it's listing - insisting, prog XYZ is ranked 1st place, so must be the strongest available - ect ? On the other hand, this listing shows nothing about the reality of game play against humans - for example - as implemented in Eric HALLSWORTH SS list. As we all know, it's possible to prepare a prog exactly for this usage, from specialized opening books to a brand new feature of a prog : " play vs. REBEL, GENIUS " - ect. How do you feel about this,in general ?
ES: I have mixed feelings about this. In my opinion human-comp is much more important than comp-comp after all the program is written for humans. But since the big human-comp events have disappeared (AEGON, Harvard Cup) what else is left other than to play comp-comp? So comp-comp becomes even more important.
It's my opinion this indirectly (and unwanted) hurts the progress in the quality of the chess engines as programmers are forced to concentrate on good comp-comp results which I consider a very bad development as being the (suggested) strongest comp-comp program doesn't necessarily mean you are the strongest against humans too (and AEGON has proved that) and because programmers are more or less forced to focus on good comp-comp results the quality of the returned moves by chess programs may even lower.
One (small) example, Rebel since version 9 is able to find forced checkmate combinations up to checkmates in 30 moves.
5n2/B3K3/2p2Np1/4k3/7P/3bN1P1/2Prn1P1/1q6 w - - bm e3g4;
If you import the following position into Rebel9 or 10 the program will find the "mate in 30 moves" within one second. In principal this is crazy, you don't need to search that deep against humans as the computer by far is superior in these kind of positions. It would be an improvement to limit these kind of deep searches to (say) "mate in 12 moves" against humans with as result the program will be faster (so better) in the overall performance against humans.
But the code will not be changed for reasons that other chess programs do the very same (deep tactics) so you simply need it if you compete in the comp-comp area.
RE:
REBEL has given proof of it's extraordinary game play quality against even strongest human players. In1997 a beta of R 9 beat GM YUSUPOV, in 1998 a pre release of R 10 - with the " Anti GM " setting, had an astonishing victory against VISWANATHAN ANAND, known as the strongest BLITZ player in the world at present. Even on tournament mode R 10 played excellent with a 0,5 - 1,5 score. The already mentioned author branded your advertising concerning this victory as " low classed style " - ANAND just wasn't careful enough to allow, other GM's were more professional before.....? , ect. How about this ? Is it really so, that ANAND, an experienced Top GM, didn't know, you would publish the result - no matter how it would have ended ?
ES: Of course Anand knew he was connected with a commercial company and I clearly remember Anand didn't play for free against Rebel. I have read the "low classed style" remark in Computer Schach und Spiele too and it has put a smile on my face and I am not willing to throw back with mud.
RE:
Ed, let's talk about the suggested future of computer chess, some call it a constructed hype. The astonished reader lately could read just exactly this - 1999 would bring the final change - towards this engine concept. As ROCHADE EUROPA reaches a very wide spread audience, and, so to say, the majority of the club chess playing and chess product using people, we authors here have the feeling of the sheer opposite as reality. From personal discussions, letters, emails, ect - we majorly hear, that this so loudly praised function means nearly nothing to the average user. What is your personally standing concerning this ?
ES: The engine concept is a new nice add-on for chess programs no doubt. But it is limited to a fixed minority of people who love to play multiple engines. My statistics clearly shows that most people use just one chess program. Nevertheless for the future Rebel will also allow multiple chess engines.
RE:
I would like to ask you a few questions about the daily, inside life of the REBEL crew and their work. To my surprise I never read anybody asking this - but know quite a few people, who asked ....me...if I knew or could imagine....... Ed - to give our readers a better view of this complicated job; what would you roughly suggest, how long - in hours, weeks or month does it take, to get a - R 9 to R 10 ( just viewing the engine ) ready for beta testing phase. What would you say, how many games are needed overall, how many engine concepts and changes are made - which don't turn out to be the best - until it's finally done ?
ES: It depends from year to year but in general the following is true for a typical month, 5 days engine programming, 8-10 days for manual testing the new software. Then the automatic (auto232) testing takes place for which I have 8 Pc's (4 autoplay pairs) available. These auto232 results should confirm my feelings about the changes made. If it all fits the changes to Rebel are accepted and kept for the next commercial release.
Another part of my job is to be in control for all the new features. I do this together with Rob Kemper the GUI programmer of Rebel. We both discuss the frame of the next commercial Rebel release resulting in a long list of new planned features. To the list is added all kind of useful hints we get from customers either by mail, fax or email. And then Rob can start to program.
Next I have to coordinate all kind of other things. With Jeroen Noomen about the new Rebel book, with Wybe Koopmans about the new Rebel Database, with Andy Duplain about the Windows version, with Manfred Rosenboom about the manual, the Rebel FAQ, with Jan Willem Schoonhoven who is responsible for all customer questions on Internet.
Last together with Manfred and Jan Willem I am responsible for the Rebel Home Page which is a very time consuming job sometimes and can eat a lot of my time.
RE
Last week I received an interesting question of a German Corr. Chess IM - who is naturally especially interested in high qualified opening theory. There are only a handful of these experts ww, who can write and tune an opening book, specialized to the product and good enough for common usage / training - your JEROEN NOOMEN is one of them. So : here's the question : which are the major influences, that make the decision for implements of lines to an opening book. Only latest or newest developments - or more ?
It's true, there are only few good opening specialists in the whole world and I am very lucky to have Jeroen who is completely in charge for 8 years now for the Rebel opening book. It is a real blessing for me not to have any work on this part. Jeroen has a free hand in ordering opening books of his choice and he usually orders 4 or 5 opening books a year and then starts to type them into the Rebel book and checks them with Rebel for blunders.
The nice thing about these books are that they contain all kind of grandmaster analysis of opening lines which are never practiced. This gives the Rebel book something extra. Next of course the latest opening theory is added so the Rebel book is always up-to-date. Then a whole bunch of auto232 games are played that checks the new book for possible blunders. Also there is special utility that analyses the whole new book for possible blunders too.
RE:
Thanks a lot for time and kindness for this interview.
Thank you Detlef for this interview. I enjoyed it a lot and give my regards to
your readers.
Ed Schroder
| |
|
|