Deep Blue vs Kasparov, and some thoughts on 8th WCCC
Feng-Hsiung Hsu
reposting from a different machine, with minor additions to the original
post.
The announcement was made at the closing ceremony of the 8th WCCC. IBM
has accepted an invitation by ACM to play a 6-game match at regular time
control with Gary Kasparov. The match will be played in Philadelphia as
part of ACM's 50th Anniversary celebration. I don't have the press release
with me, but if my memory is right, the match is to take place between
Feb 10 and 17, 1996.
During the past 6 months, DT-2 was powered down most of the time, so that
the new chip could be simulated on the only machine we have that has enough
memory. The new chip was searching about one node per second, roughly the
same speed as Gary Kasparov :), when running in simulation. When we play
Kasparov, it should be running at 3-5 million nodes/sec per chip. Current
plan is to use an SP2 as the host and possibly anywhere between 128-512
chess chips as the slave processors.
With that out of the way, I will say something about the 8th WCCC.
First, I would like to congratulate the Fritz team on their winning the WCCC.
The only annoying thing is that Frederic will be intolerable when I talk to him
next time:). I first met members of the Fritz team back in '93. Very nice
chaps. Never did meet the new guy who prepared Fritz's opening book for this
event. From what happened in the last two games, and from talking with GM
Robert Byrne, I would have to say "a job well done". Fritz played a tricky
transposition in the game against Deep Blue Prototype (DT-2). Normal book line
is to play Bg7 before f4. Our game database did contain the transposition,
and gave the verdict that o-o or o-o-o is no good in the position, and g3 or
c3 is called for. The automatically generated book, unfortunately, cut off
right after the early f4, and our luck ran out. Normal book line is Bg7, o-o,
f4 and then g3, by the way. This loss is probably good for us in the long run.
Book preparation will be taken far more seriously from now on. One additional
side note. Instead of c4? allowing Qh5, the immediate g3 appears to hold--
might still be lost against the likes of Kasparov, but I doubt Fritz can push
it through. The machine did play g3, but the phone was disconnected at the
worst possible time, and when it was restarted, it did not have enough time
to rediscover the move. I have not really seen the *Socrates-Fritz game, but
Grandmaster Robert Byrne has done some analysis in the past on the ancient
line that Fritz played and is quite familiar with the line. His opinion
is that unless white is prepared for the line, white is likely to get into
deep trouble as *Socrates did.
Hm, book preparation will be taken VERY seriously from now on:).
Next, I would like to make some random comments about the competition itself.
There are surprising number of draws in this event. Before the last round,
there was an ICCA/players' meeting. Players, or rather programmers, were not
happy with the chancy nature of the format, and were asking that future events
use a 7-round swiss format instead of the 5-round swiss format that has been
used in the past few years. I would much prefer a double round-robin format
myself, but it is probably logistically impossible. The tournament location,
Hong Kong, is fairly interesting. I was born in Taiwan, about 2-hour flight
from Hong Kong, but this was my first trip to Hong Kong. Even though it was
only May, weather was quite hot and humid. The temperature was not really too
bad. High was around 32 C, or 89 F, but the relative humidity was hovering
around 90%. Food is excellent, but somewhat oilier than I would like. One
pleasant surprise is that smoking is not as popular as most places outside of
North America. Signs are in both English and Chinese. The Chinese written
language is roughly the same as in Taiwan. The spoken dialect, Cantonese,
however, is mostly incomprehensible to me. Once in a while, I did hear
Mandarin being spoken. There is fast growing trade with Mainland China, and
Hong Kong will be reverting to Chinese rule in 1997. We managed to do some
sightseeing, but not enough to do the place justice. The playing site, the
Chinese University of Hong Kong, is on a train route that connects directly
into China. Some of the players took the train into China the day after the
closing ceremony. We had the plane to catch, and had to pass. The University
is on a hill, and the playing hall has a spectacular view looking out from the
window. The Guest House, where the closing ceremony is held, had an even
better view. Almost like looking at a Chinese Painting. There were some
mishaps with the internet connection from the University. The three main
universities in Hong Kong use a common carrier for internet connection. Before
the event, the remote teams tried out the connections remotely and the
connections appeared ok. As it turned out, the connections into the Hong Kong
universities appear fine, but getting out of the universities is almost
impossible, at least during the tournament. IBM Hong Kong, as chance would
have it, announced internet service for Hong Kong area the day before the
tournament. The remote teams got to learn how to install OS2/Warp on PCs as
a result:).
Finally, some comments on the "Deep Disappointment". Yes, we are disappointed,
but not "Deeply". Given the format of the tournament, our chance to win it
was around 50%, while any other team's chance was at around 4%, with the
possible exception of *Socrates. Going into the last round, our winning chance
was up to around 90% (we pretty much clobbered Fritz in test games). As
Murphy's law would have it, Fritz grabbed the 10% chance. In hindsight, there
were a bunch of things that we could have done to avoid the disaster, but
hindsight is 20-20. The tournament also highlighted some of the deficiencies
of the current hardware. The current hardware does not detect repetition, and
effectively we are losing 4 plies as far as repetition is concerned.
Repetition detection is not only important to hold draws, but also serves as
an early warning system. In the game against WChess, the move Ra8? by DT-2
forcing rook trade was positionally incorrect, but it was also tactically bad
as it allowed at least a repetition draw. The o-o?? move in the game against
Fritz also allowed at least a repetition draw and would have been avoided
if the hardware were capable of detecting repetition. In the past,
we had not been burned by this deficiency, and it is a testimony to how far
the other programs have advanced that we did get burned by it. We will see
how far we have or will have advanced with respect to Kasparov next February.
Guido Stepken
:
: ..........................Going into the last round, our winning chance
: was up to around 90% (we pretty much clobbered Fritz in test games). As
: Murphy's law would have it, Fritz grabbed the 10% chance. In hindsight, there
: were a bunch of things that we could have done to avoid the disaster, but
: hindsight is 20-20. The tournament also highlighted some of the deficiencies
: of the current hardware. The current hardware does not detect repetition, and
: effectively we are losing 4 plies as far as repetition is concerned.
: Repetition detection is not only important to hold draws, but also serves as
: an early warning system. In the game against WChess, the move Ra8? by DT-2
: forcing rook trade was positionally incorrect, but it was also tactically bad
: as it allowed at least a repetition draw.
Hello !
For 5 years I admire your work in devellop chess computers, very impressing !
Nice to hear, that your program does not detect repetition. I think, it may be
possible to hold a draw, forced by repetitions for many programs !?
10% chance for FRITZ is really a lot, but I really must say: Next time
computer tournaments should become longer for the first 10 candidates to get
rid of random wins.
B.t.w., I am really interested in the overhead you do have by using highly
parallel systems and I am sure there are many more persons ?!
greetings, Guido
Robert Hyatt
>: was up to around 90% (we pretty much clobbered Fritz in test games). As
>: Murphy's law would have it, Fritz grabbed the 10% chance. In hindsight, there
>: were a bunch of things that we could have done to avoid the disaster, but
>: hindsight is 20-20. The tournament also highlighted some of the deficiencies
>: of the current hardware. The current hardware does not detect repetition, and
>: effectively we are losing 4 plies as far as repetition is concerned.
>: Repetition detection is not only important to hold draws, but also serves as
>: an early warning system. In the game against WChess, the move Ra8? by DT-2
>: forcing rook trade was positionally incorrect, but it was also tactically bad
>: as it allowed at least a repetition draw.
>
>
>For 5 years I admire your work in devellop chess computers, very impressing !
>Nice to hear, that your program does not detect repetition. I think, it may be
>possible to hold a draw, forced by repetitions for many programs !?
>10% chance for FRITZ is really a lot, but I really must say: Next time
>computer tournaments should become longer for the first 10 candidates to get
>rid of random wins.
>B.t.w., I am really interested in the overhead you do have by using highly
>parallel systems and I am sure there are many more persons ?!
>
>greetings, Guido
>
He didn't say *exactly* that... he said that the hardware chess
processors don't detect repetition. however, the machine that
handles the "bottom" of the tree (used to be an IBM RS workstation,
and it was searching about 6-7 plies deep and then handing these
positions off to the special-purpose hardware. This part of the
tree *could* detect repetitions. However, if the repetition didn't
occur until ply=10, they would not see it since the hardware didn't
have access to the software move stack.
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170