: [computer-go] Is MC-UCT really scalable against humans?

[compgo123 at aol.com]

Just for  reference. There other evaluation methods, such as the end game theory, that can give
acurate scoring.

DL

-----Original Message-----
From: ivan dubois <dubois_ivan at yahoo.fr>
To: computer-go <computer-go at computer-go.org>
Sent: Tue, 22 Jan 2008 4:14 pm
Subject: Re : [computer-go] Is MC-UCT really scalable against humans?



Hi Don,
I agree that the very best players can know, if they are given enough time, 
hether they are winning or not. But here we are talking about professionnal 
layers. But you know, even a 9d Pro, probably cannot be certain of the outcome 
t the yose stage, if he plays a blitz game. Still, if he thinks he is leading, 
e will always play "normal" moves, defending his territories instead of playing 
trange and apparently useless consoliding moves like a MC player would do. Yet 
f he thinks he is losing, he will probably try unsound invasions and try 
opeless things exactly like a MC would do, when the outcome of the game matters 
or him.
or example, i am 4d on KGS, and when there is some trade to be made beetwen 
oints and safety, this is always a tough decision for me.
ecause i dont know how many points i can concede exactly, and i dont know how 
isky some line of play realy is.
 have lost a lot of games because i made the wrong choice in such situations. 
o i think you are over estimating the "evaluation" ability of human players. 
his is the reason why human generaly dont play like MC, because they do not 
ave the same informations. MC are good at knowing what the situation is, humans 
re good at playing moves that tend to maximise territory most of the time. 
umans and MC just do not have the same strengh. Somehow, i would say that MC is 
ore intelligent, because it relies on understanding the situation rather than 
atterns and reflexes. 

---- Message d'origine ----
e : Don Dailey <drdailey at cox.net>
 : computer-go <computer-go at computer-go.org>
nvoyé le : Mardi, 22 Janvier 2008, 22h22mn 42s
bjet : Re: Re : [computer-go] Is MC-UCT really scalable against humans?
I agree that probably most players play as you say.    But it's
ifficult for me to believe that the very best players don't know if the
re wining or not.    
I think if I had the skill of a professional,  I would make it my
usiness to know, ESPECIALLY  if it were really close.    
I'm talking about the cases where a UCT player thinks it's winning with
igh probability.  If a UCT player knows this,  I'm sure a really
trong professional knows.      It's a different story when several
roups are in question and a UCT players is "skillfully" assessing the
dds.    But I don't think we are talking about that case.    UCT
oesn't start playing too bizarre until a win is in the bag for one of
he players.  
I've seen this for myself.    The score gets about 95% but there are
till many ways to lose if you play stupid,  but the MC player knows
arefully takes care of business - sealing this and that off so that
here is no chance.  
Many times I've seen it throw away 1 or 2 points by not claiming an
xtra point right next to the point that it does claim.    In some of
hose cases I counted it off myself and to my amazement it just didn't
eed that point and it was probably a random choice.
But in those cases there was no chance of tricking a human.    
evertheless, the human sometimes is under the impression the program
ade a series of "really bad" moves and STILL WON,  making the human
eel frustrated.    But really it was not like that at all.
- Don

ivan dubois wrote:
 Hello Don,

 I think you are mostly right, but there is something you seem not to realise. 
 MC players are very good at "scoring". They know when they will win for sure 
y 0.5. However, human players, even strong players, generaly never have such 
ccurate counting skills. So if they are ahead by 0.5, there is no way they can 
nticipate it at as a sure win, because it would require amazing couting skills. 
ctually, a lot of strong players never count during the course of a game, 
specialy during fast games.
 So often, the most rational way of playing is to be conservative with the 
oints you have, and just always play the natural, conservative defending moves. 
t may be too hard for human to reason about "probability of winning", it is 
asier to reason about points you have or not.
 Of course, i think the ultra rational way based on accurate couting by MC 
rograms is a great strength they have. Its a specific strength of MC that 
uman, unfortunately for them, do not have.


     _____________________________________________________________________________ 
Ne gardez plus qu'une seule adresse mail ! Copiez vos mails vers Yahoo! Mail 
ttp://mail.yahoo.fr
______________________________________________
omputer-go mailing list
omputer-go at computer-go.org
ttp://www.computer-go.org/mailman/listinfo/computer-go/


________________________________________________________________________
More new features than ever.  Check out the new AOL Mail ! - http://webmail.aol.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://computer-go.org/pipermail/computer-go/attachments/20080122/5a216510/attachment.htm