Regardless of the context of the statement, what Dave said about computer
chess is completely incorrect.
Therefore, his conclusion (whether correct or not) does not logically follow
from this.
I disagree even with his second paragraph, where he says:
By contrast, there are no obvious evaluation metrics that work for Go, and
the
search tree is impossibly large. Advances in computer speed have not, and
arguably never will make the same simple, brute force methods that work for
Chess suitable for Go.
There appears to be all kinds of obvious metrics that work for Go. They
are all flawed just like evaluation in chess is all a flawed approximation
that goes badly wrong in many situations, but they do in fact work. Local
search is one obvious method that works very well and has been used for
years. Patterns is another.
His second sentence is flawed too, it's a sound bite that has been repeated
for decades but the fact of the matter is that chess has not used brute
force for decades because there are far better methods. Nobody writes a
brute force chess program these days. So I think Dave has not really kept
up with what is going on with computer chess.
By the way, I cringe whenever someone says, "I tried that and it doesn't
work." What is that supposed to mean? To say something works or
something does work requires one to define what their objective or goal is.
If a mousetrap does not catch mice, it doesn't work. So when Dave
says brute force "does not work" it's a meaningless statement. To me,
in games programming any idea should be considered a success if it improves
the play of the program, and by this metric it's already been demonstrated
that even brute force "works" in GO, you can clearly make a program play
much stronger using brute force alpha beta than not doing any search
whatsoever. However brute force is by far inferior to other techniques
in chess and I'm sure it's that way in GO too.
On Wed, Jun 2, 2010 at 12:12 PM, steve uurtamo <[email protected]> wrote:
>
> This is WAY MORE than just alpha/beta and counting pieces.
>>
>
> Keep in mind what the original question was about -- why is one so much
> further along strength-wise than the other, or what it is that makes the two
> games different from a machine-attacking point of view.
>
> Sure, chess programs will continue to be developed, but the time when
> non-grandmasters could have a reasonable chance of beating even an
> open-source program from a few years ago has long since passed. It's
> arguable that the time when grandmasters could have a reasonable chance has
> long since passed as well, but that's not required for this point. The point
> is that go programs are not at this same level, and that the chess approach
> does not work on a go board with go stones and the rules of go, and that as
> far as the current approaches are concerned, go appears to be a "harder"
> game.
>
> Nobody doubts that chess improvements happen every day, but they certainly
> aren't required to continually crush humans at chess.
>
> s.
>
> _______________________________________________
> Computer-go mailing list
> [email protected]
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
