| |
Main
Date: 08 Dec 2006 16:52:12
From: johnny T
Subject: A missed future???
|
I don't know if anyone noticed, but Fritz did an amazing thing in Game
6. It came up with a novelty, and a different way of approaching a
setup, than had been considered by anyone "human" to date.
The moves were "wrong", they were mocked by experts live (giggling
happened), and ultimately proved correct.
This should give everyone the chills.
The worry is that lack of interest will mean that humans will not be
willing to learn from the machines.
Nobody pays attention to what happens in machine vs machine matches.
And nobody who uses these machines will uncover these new ways of
playing a position. They just blindly record the score.
What is unique about a machine/human match like this, is that we care,
people care, grandmasters care about what was played. About how a
position is managed by the players.
In game 6, this wasn't a blunder by a human. This wasn't an obvious
thing that an expert would do. The earth shook a little. A machine
had a different way of looking at a problem, and came up with a
different, yet correct answer. THAT IS FREAKING AMAZING. Not that
Kramnik lost. That is not nearly as amazing. It was the WAY that he lost.
We should have more events like this. The machines have more to teach
us. If we care.
|
|
| |
Date: 11 Dec 2006 19:01:41
From:
Subject: Re: A missed future???
|
johnny T wrote:
> [email protected] wrote:
>
>
> What is the ELO value of hardware? How has it improved over time.
These websites will help you quantify that.
http://www.sedatchess.com/hardware.html
http://www.jens.tauchclub-krems.at/diverses/Schach/fritz9_benchks.html
A few things you ought to consider.
1. 64-bit chess engines add about 70% speed over 32-bit.
2. Going from 2 to 4 to 8 processors doesn't translate into comparable
speed improvements. Different multi-processor chess engines scale up
differently, but none of them scales 1:1. Typically the more
processors, the less efficiently that power increments. So you might
get 4x the speed of a single-thread machine with an octo. It varies.
3. Each absolute doubling of speed (in real terms) is worth 60-75 ELO
points. Or rather, it has historically. As we get into ever-stronger
hardware, software becomes critical, because search pruning becomes
increasingly more important than absolute processing speed. So my rule
of thumb may not hold much longer.
4. Of course, if you can afford to overclock and you know what you're
doing, there's a quick way to add another 10-20% in speed. (Some
people even do better, but the cooling methods get increasingly more
extreme and expensive, not really worth it when processors keep
improving so quickly.)
All the above assumes no other factors influencing results: opening
book, tablebases, etc.
|
| |
Date: 08 Dec 2006 18:42:16
From:
Subject: Re: A missed future???
|
While I don't share your astonishment (because I take it for granted
that machines are way better than humans at chess, far more innovative,
and are the future of chess whether anyone admits it or not), I agree
with your basic premise.
A few points. First, impressive as DF10 was against Kramnik, all the
rating lists will soon indicate that Rybka 2.2 mp 64-bit is at least
100 ELO points stronger on comparable hardware. (Preliminary results
indicate this; I think DF10 will make the first list this weekend.) It
plays at over 3000 ELO.
Computers already play totally counterintuitive and inexplicable moves
that lead inexorably to mate. People have started on 7-man tablebases
that have produced positions that result in mate in over 200 moves
(apart from the 50-move rule). I defy any GM to see the logic in the
moves that force such long-distance mates.
Finally, on innovation, it's very simple to create brand-new theory.
Lately I've been seeing some serious possibilities in 1.h3, which looks
like a silly patzer move, but in reality is just as sound as 1.e4. One
move is as miraculous as the next; brilliant combinations happen all
the time between machines; they just aren't noticed. Someday they will
be, because the self-evident superiority of computers cannot be denied
or ignored much longer.
|
| | |
Date: 11 Dec 2006 17:55:52
From: johnny T
Subject: Re: A missed future???
|
[email protected] wrote:
> A few points. First, impressive as DF10 was against Kramnik, all the
> rating lists will soon indicate that Rybka 2.2 mp 64-bit is at least
> 100 ELO points stronger on comparable hardware. (Preliminary results
> indicate this; I think DF10 will make the first list this weekend.) It
> plays at over 3000 ELO.
Actually, This is a figure that I had been interested in, and not sure
if anyone has come up with it...
What is the ELO value of hardware? How has it improved over time.
The hardware can be representative, and we can remeasure after
significant events.
But I am curious how bad ass my Fritz 8 is on my MacBook 2 gig CD
processor is. Especially compared to how it was, when I first bought
the program to run on my Pentium 233. What would the ELO difference be?
|
| | |
Date: 11 Dec 2006 10:01:19
From: David Richerby
Subject: Re: A missed future???
|
<[email protected] > wrote:
> A few points. First, impressive as DF10 was against Kramnik, all
> the rating lists will soon indicate that Rybka 2.2 mp 64-bit is at
> least 100 ELO points stronger on comparable hardware. (Preliminary
> results indicate this; I think DF10 will make the first list this
> weekend.)
But these ratings are only from computer-computer games. While it's
likely that Rybka's ability to beat other computers will translate
into an ability to beat GMs, it's not guaranteed.
> It plays at over 3000 ELO.
Note that this rating is not directly comparable with FIDE's rating
list. Saying `Rybka is 3000 Elo' is like saying that the size of
something is `150, measured with a tape measure'.
> Finally, on innovation, it's very simple to create brand-new theory.
> Lately I've been seeing some serious possibilities in 1.h3, which
> looks like a silly patzer move, but in reality is just as sound as
> 1.e4.
Well, intuitively, playing 1.h3 shouldn't be significantly worse than
playing Black and playing Black doesn't appear to be a forced loss
right now. (It might be slightly worse as it weakens the kingside a
little.)
Dave.
--
David Richerby Transparent Atom Bomb (TM): it's like
www.chiark.greenend.org.uk/~davidr/ a weapon of mass destruction but you
can see right through it!
|
| | |
Date: 08 Dec 2006 22:46:21
From: Ange1o DePa1ma
Subject: Re: A missed future???
|
<[email protected] > wrote in message
news:[email protected]...
> While I don't share your astonishment (because I take it for granted
> that machines are way better than humans at chess, far more innovative,
> and are the future of chess whether anyone admits it or not), I agree
> with your basic premise.
>
> A few points. First, impressive as DF10 was against Kramnik, all the
> rating lists will soon indicate that Rybka 2.2 mp 64-bit is at least
> 100 ELO points stronger on comparable hardware. (Preliminary results
> indicate this; I think DF10 will make the first list this weekend.) It
> plays at over 3000 ELO.
>
> Computers already play totally counterintuitive and inexplicable moves
> that lead inexorably to mate. People have started on 7-man tablebases
> that have produced positions that result in mate in over 200 moves
> (apart from the 50-move rule). I defy any GM to see the logic in the
> moves that force such long-distance mates.
>
> Finally, on innovation, it's very simple to create brand-new theory.
> Lately I've been seeing some serious possibilities in 1.h3, which looks
> like a silly patzer move, but in reality is just as sound as 1.e4. One
> move is as miraculous as the next; brilliant combinations happen all
> the time between machines; they just aren't noticed. Someday they will
> be, because the self-evident superiority of computers cannot be denied
> or ignored much longer.
I agree with you and Johnny. From time to time there are discussions in the
chess newsgroups about computer play. Most posters believe the nonsense they
hear about the inferiority of 'computer moves' and how computers can't play
creatively, etc. Hogwash. They're stronger than anyone at this point.
You're right, but I'm not sure the Kramnik match was a good barometer. Vlad
overlooked a mate in one, and in the last game had nothing to lose so went
all-out. Take away the one-mover and this match is a draw.
|
| | | |
Date: 10 Dec 2006 05:40:48
From: DorothyFan1
Subject: Re: A missed future???
|
In article <[email protected] >,
[email protected] says...
> <[email protected]> wrote in message
> news:[email protected]...
> > While I don't share your astonishment (because I take it for granted
> > that machines are way better than humans at chess, far more innovative,
> > and are the future of chess whether anyone admits it or not), I agree
> > with your basic premise.
> >
> > A few points. First, impressive as DF10 was against Kramnik, all the
> > rating lists will soon indicate that Rybka 2.2 mp 64-bit is at least
> > 100 ELO points stronger on comparable hardware. (Preliminary results
> > indicate this; I think DF10 will make the first list this weekend.) It
> > plays at over 3000 ELO.
> >
> > Computers already play totally counterintuitive and inexplicable moves
> > that lead inexorably to mate. People have started on 7-man tablebases
> > that have produced positions that result in mate in over 200 moves
> > (apart from the 50-move rule). I defy any GM to see the logic in the
> > moves that force such long-distance mates.
> >
> > Finally, on innovation, it's very simple to create brand-new theory.
> > Lately I've been seeing some serious possibilities in 1.h3, which looks
> > like a silly patzer move, but in reality is just as sound as 1.e4. One
> > move is as miraculous as the next; brilliant combinations happen all
> > the time between machines; they just aren't noticed. Someday they will
> > be, because the self-evident superiority of computers cannot be denied
> > or ignored much longer.
>
> I agree with you and Johnny. From time to time there are discussions in the
> chess newsgroups about computer play. Most posters believe the nonsense they
> hear about the inferiority of 'computer moves' and how computers can't play
> creatively, etc. Hogwash. They're stronger than anyone at this point.
>
> You're right, but I'm not sure the Kramnik match was a good barometer. Vlad
> overlooked a mate in one, and in the last game had nothing to lose so went
> all-out. Take away the one-mover and this match is a draw.
>
>
>
I agree that this wasn't the proper barometer to showing the difference
between the human against computer in chess. It artificially insulated
Kramnik from losing badly because he could rely on book theory. Take
away that comfort zone by throwing the World Champion to the wolves with
a Fischer Random match against Rybka and THEN we can judge just how good
or bad humans are against the computer. Let me give you a hint...GM
Radjabov tried to be really "cute" against the computer using Fischer
Random rules by simply "rearranging" the pieces to get a familiar
structure he knew. GM Radjabov was promptly crushed in less than 20
moves.
Someone asked former World Champion Gary Kasparov if he'd play a
computer using Fischer Random rules. He said no because the computer
would simply outcalculate him tactically. Huh? I thought it was commonly
said that human "intuition" made up for the advantage computers have
with their calculating ability.
|
| | | | |
Date: 10 Dec 2006 11:37:01
From: Tony M
Subject: Re: A missed future???
|
On Sun, 10 Dec 2006 05:40:48 GMT, DorothyFan1
<[email protected] > wrote:
>
>Someone asked former World Champion Gary Kasparov if he'd play a
>computer using Fischer Random rules. He said no because the computer
>would simply outcalculate him tactically. Huh? I thought it was commonly
>said that human "intuition" made up for the advantage computers have
>with their calculating ability.
That human intution consists mostly of pattern recognition and
memorization of openings and themes. Scramble the starting position,
and a large amount of those patterns and themes are suddenly
invalidated. No more charging the h pawn towards the enemy king in
order to open up the file for the unmoved rook behind it, because now
the rooks start on b1 and c1. No more comfort from playing your
favorite Sicilian variation. The loss of that sort of stuff doesn't
faze the computer, because it always calculates anew, in the same
fashion, no matter what the position. Those fleshbags that call
themselves humans don't stand a chance.
Tony
|
| | | | | |
Date: 11 Dec 2006 10:04:04
From: David Richerby
Subject: Re: A missed future???
|
Tony M <[email protected] > wrote:
> DorothyFan1 <[email protected]> wrote:
>> Someone asked former World Champion Gary Kasparov if he'd play a
>> computer using Fischer Random rules. [...]
>
> That human intution consists mostly of pattern recognition and
> memorization of openings and themes. Scramble the starting
> position, and a large amount of those patterns and themes are
> suddenly invalidated. No more charging the h pawn towards the enemy
> king in order to open up the file for the unmoved rook behind it,
> because now the rooks start on b1 and c1.
Actually, the rooks can't start on b1 and c1 because, in FRC, the king
has start to be between the two rooks. (And the bishops must be on
different coloured squares.) But the main thrust of your point still
stands.
Dave.
--
David Richerby Moistened Lotion (TM): it's like a
www.chiark.greenend.org.uk/~davidr/ soothing hand lotion but it's moist!
|
|
