Computer will defeat W.C. in Year X?
Matthew W. L. Phelps
In article <1993Mar3.1...@math.toronto.edu>, hs...@math.toronto.edu (Paul Hsieh) writes...
>In article <i7mTZB...@tweekco.uucp> j...@tweekco.uucp (Jay Whitley) writes:
[etc.]
>>Same QUESTION to rec.games.chess readers:
>> "In what year do you think a chess program will be able
>> to defeat the human World Champion?"
>>
>Looking into my crystal ball ... Seriously though, recent
>improvements in computer chess programs seem to indicate that
>it just might happen. The year 1996 or so sounds right to me
>and I am willing to bet that it will NOT be Deep Thought/Blue,
>but rather something like Mephisto or Schreoder's machine,
>once they decide that they've got to use some better hardware.
>I have never seen Deep Thought play any particularly stunning
>chess. The thing is just a machine and you can tell that
>from the way it plays. Mephisto on the other hand, now
>there's a machine that plays more than just tactics. Hey!
>maybe Kasparov's Gambit running on a 100 Mhz Pentium will beat
>Kasparov (or Short or Anand if they everbecome champion)!
How about Kasparov's Gambit (or MChess Pro, or whatever) running on a
multi-200Mhz Alpha chip DEC PC running Windows NT? It'll happen sooner than you
think.
My guess for World Champion class performance by a computer is early 1994. I
doubt a computer will ever be the "official" World Champion (whatever that
is/may be) because of organizational issues. i.e. There's no computer IMs or
GMs because they aren't allowed to have FIDE titles.
>Now
>wouldn't that be something? Smells of a Frankenstien's
>Monster analogy.
Will we then become the rioting townsfolk with pitchforks and torches? ;-)
>
>Paul (qed)
Matthew W. L. Phelps mph...@amama.enet.dec.com
Digital Equipment Corp. "My stuff, not DEC's"
Transaction Processing Systems Engineering, Littleton, MA, USA
Jay Whitley
posts and the Deep Blue Copenhagen presentation reminded me
of Chapter 1 in David Levy and Monty Newborns' _HOW COMPUTERS
PLAY CHESS_ (1991, W.H.Freeman and Co.). Computer programs'
performance ratings from 1967-1985 were extrapolated by Levy
back in 1986 to guess when a computer might achieve World
Champion level. 1994 was the projected year via the graph.
The book also listed estimates from 43 computer science
experts and strong human players attending the 1989 World
Computer Chess Championship in Edmonton. The mean of the
finite responses to the question listed below was the year
2005. It looks to me like the median response was the year
2000. Some of the estimates included: Feng-hsiung Hsu (1994),
Larry Kaufmann (1995), Robert Hyatt (2000) <ANY CHANGE BOB?>,
David Levy (2014), David Cahlander (NEVER).
Steven J. Edwards
> experts and strong human players attending the 1989 World
> Computer Chess Championship in Edmonton. The mean of the
> finite responses to the question listed below was the year
> 2005. It looks to me like the median response was the year
> 2000. Some of the estimates included: Feng-hsiung Hsu (1994),
> Larry Kaufmann (1995), Robert Hyatt (2000) <ANY CHANGE BOB?>,
> David Levy (2014), David Cahlander (NEVER).
>
> Same QUESTION to rec.games.chess readers:
> "In what year do you think a chess program will be able
> to defeat the human World Champion?"
In 1985 I made a US$50 wager that at least one chessplaying program
would attain USCF Senior Master (2400) status before the end of the
year 1990. I won the bet as Berliner's Hitech passed 2400 about a
year before the deadline.
Instead of collecting my winnings, I made a double-or-nothing wager:
if a program attains world champion status (either by rating or by
defeating the human champion in a match) by the end of 1995, I win
US$100.
[The above opinions expressed are my own; not necessarily held by others.]
== Steven J. Edwards Bull HN Information Systems Inc. ==
== (508) 294-3484 300 Concord Road MS 820A ==
== s...@xylos.ma30.bull.com Billerica, MA 01821 USA ==
"That Government which Governs the Least, Governs Best." -- Thomas Jefferson
Paul Hsieh
improvements in computer chess programs seem to indicate that
it just might happen. The year 1996 or so sounds right to me
and I am willing to bet that it will NOT be Deep Thought/Blue,
but rather something like Mephisto or Schreoder's machine,
once they decide that they've got to use some better hardware.
I have never seen Deep Thought play any particularly stunning
chess. The thing is just a machine and you can tell that
from the way it plays. Mephisto on the other hand, now
there's a machine that plays more than just tactics. Hey!
maybe Kasparov's Gambit running on a 100 Mhz Pentium will beat
wouldn't that be something? Smells of a Frankenstien's
Monster analogy.
Paul (qed)
Robert Hyatt
>posts and the Deep Blue Copenhagen presentation reminded me
>of Chapter 1 in David Levy and Monty Newborns' _HOW COMPUTERS
>PLAY CHESS_ (1991, W.H.Freeman and Co.). Computer programs'
>performance ratings from 1967-1985 were extrapolated by Levy
>back in 1986 to guess when a computer might achieve World
>Champion level. 1994 was the projected year via the graph.
>
>The book also listed estimates from 43 computer science
>experts and strong human players attending the 1989 World
>Computer Chess Championship in Edmonton. The mean of the
>finite responses to the question listed below was the year
>2005. It looks to me like the median response was the year
>2000. Some of the estimates included: Feng-hsiung Hsu (1994),
>Larry Kaufmann (1995), Robert Hyatt (2000) <ANY CHANGE BOB?>,
>David Levy (2014), David Cahlander (NEVER).
>
Yes. I still believe it will happen, but after studying some of the
difficulties for years, I believe the solution is farther away than
most of us felt a couple of years ago.
My reasoning is based around a simple evaluation problem that I have been
puzzling over for a couple of years, the isolated passed pawn. In some
games an IPP wins outright, in other games it is lost. The difficulty
comes in deciding whether it can be held, held and advanced, or must be
lost.... and this is not an easy thing to do. What does it cost the
opponent to relocate pieces to attack/hold/win it? what does it cost
the program to defend/advance it? As a human, I can pretty well make
the decision, but the program has one hell of a time since it can't
possibly search deep enough to create the passed pawn, relocate all the
pieces for both sides optimally, and then see what's what. For a good
example, look at our last round game against *tech in the last ACM
tournament two weeks ago. We were playing black with a really solid
position, and created an isolated, passed C-pawn when we had the
chance. The program got so tied up holding that pawn that it was lucky
to emerge with a draw although it got quite excited when it noticed the
passer. In other similar games we have created such a pawn and then
won with it. The distinction between the two cases is more obvious to
me than to the computer.
This is but one example where we are trying to create and apply general
purpose chess knowledge to situations where it does not apply. A funny
case we saw was where a program had a pretty large bonus for having rooks
on the 7th rank. The problem was that there were no pawns on the 7th,
and the rook was needed elsewhere. It kept it on the 7th until it was
too late. Good rule, wrong circumstance.
The tactics are going to become nearly overwhelming if Deep Blue can actually
reach 14 plies (I remain skeptical but hope they pull it off). However, it
they don't tactically crush their opponent, they aren't going to do it with
such general purpose knowledge. It will get them past masters, and maybe
even IM's, but the Kasparovs and Karpovs of the world are better than that;
IF they can survive the tactics. I believe that they currently can. Computers
are beating humans tactically even though their positions are often full of
weaknesses. In fact, humans often overextend trying to attack *all* of the
computers weaknesses and fail to a tactical coup. Haven't seen Kasparov
fall to such tactics yet, and I don't expect him to. When the knowledge
gets good enough, the program searches are already where they need to
be. I now suspect that 2010-2020 might be the decade where it might
happen...
--
!Robert Hyatt Computer and Information Sciences !
!hy...@cis.uab.edu University of Alabama at Birmingham !
Richard Title
>... Mephisto on the other hand, now
>there's a machine that plays more than just tactics. Hey!
>maybe Kasparov's Gambit running on a 100 Mhz Pentium will beat
I doubt it. 100 Mz Pentium is not that big a deal. Let's do some
calculations: It's 2x the clock speed of a 50 Mz 486. Throw in
another factor of 2 for pipelining, that's 4x. 4x speedup is not even
1 ply. Studies have shown that 1 ply is worth about 200 rating points.
So, start with a 2400 program and speed it up 4x, that makes 2600 at
best. Good, but still not in Kasparov's league.
Now, if you were to parallelize one of the top PC programs and put it
on, say, a 1000-processor CM5, that would be something...
Rich
dwe...@uoft02.utoledo.edu
> Same QUESTION to rec.games.chess readers:
> "In what year do you think a chess program will be able
> to defeat the human World Champion?"
>
please clarify "defeat"
do you mean "defeat in a match" or "just a single game"
game: 1996
match: 2020
don wedding
Paul Rubin
do you mean "defeat in a match" or "just a single game"
game: 1996
match: 2020
don wedding
Maybe not--
game: 1996 (sounds like as good a guess as any)
match: never (once a game has been lost and it's obvious
the machines keep getting stronger, why would
a reigning WC ever agree to play a match
against one?)
Colin Allen
> but rather something like Mephisto or Schreoder's machine,
Does anyone know some reference to published descriptions of these
systems? Thanks.
--
Colin Allen INTERNET (& NeXTMail) : co...@snaefell.tamu.edu
Assistant Professor | BITNET : cfa0802@tamvenus
Philosophy, Texas A&M University | FAX : (409) 845-0458
College Station, TX 77843-4237 | VOICE : (409) 845-5660
Roy Eassa
If you mean defeat the world champ IN A SET MATCH, I'd guess 2003. If you
mean in a single game, I'd guess 1993.
Roy Eassa
>Paul (qed)
I don't think a 100 MHz Pentium is, realtively speaking, a particularly
fast processor (compared to a super mainframe or a multi-processor
environment). Also, I doubt that Gambit will be that strong. I'll bet
it's clearly weaker than Lang's Chess Genius and Hirsch's MChess Pro.
Jay Whitley
>> "In what year do you think a chess program will be able
>> to defeat the human World Champion?"
>do you mean "defeat in a match" or "just a single game"?
I was afraid of this potential question. I listed Levy's
question exactly as he had it in his book. My interpretation
from reading that chapter was they were basically talking
about the point at which a computer would become World
Champion and would have a higher ELO performance rating than
the human WC. (Others might interpret it differently.)
So, I guess we're talking about A MATCH...i.e. when the
computer is demonstrably SUPERIOR to the human champion.
Since Robert Hyatt took part in the 1989 survey of experts,
perhaps he could give his interpretation of the `game or
match' question. (Today, Bob already kindly posted a year
2010-2020 answer to this rec.games.chess question.)
Otherwise, let's presume MATCH (i.e. when a computer becomes
World Champion.)
ABHAY
>> "In what year do you think a chess program will be able
>> to defeat the human World Champion?"
>
The first game will be lost because of a mistake/oversight/error/blunder
and you can not set a time frame for that.
If you are talking @ a series, well it is a different proposition
all together. A series would be won by a better player/machine.
I guess it would be 2008. And why would W.C. chapmpion play a m/c? For
money of course, lots & lots of it.
Abhay Thorat
Robert Hyatt
I believe (assume?) that we were talking about a match to "become" the
world champion. I would suspect that Deep Blue, Hitech, Cray Blitz,
Mephisto, Socrates II, and a host of others would expect to win at
least one game in a 20 game match with Kasparov as he is no more than
200-300 rating points above the best of this group. When the question
was posed, we were talking about "when will a computer become the world
champion, if such a thing were possible?" In that context, a match was
implied. The language "become the world champion" was not used because
it is currently not possible for such a thing to happen as FIDE would
not sanction such a match... note the world checker championship "match"
between Tinsley and Chinook occurred unofficially after Tinsley broke
with the ACF/WCF and played the match against their wishes.
Bob
Byungwook YU
the highest rating among other notable programs such as Zarkov, Fritz, MChess,
etc. Its rating was 2460 on 486-33. What was amazing was that it was all
achieved using only 200kb of code! I was wondering if anybody out there
ever heard about it, or if anybody can verify it.
Roy Eassa
The main .exe program is under 50k (I may have pklite'd it to this small
size). There are other files, and I don't know which one(s) other than
the .exe constitute "the program".
It ain't so new, either. It's the venerable Mephisto program ported from
the 68000 series to the 80x86 series. Been around in one form or another
for around a decade.