Re: [computer-go] Search = Bad!

[Mark Boon <tesuji@xxxxxxxxx>]

Don Dailey wrote:

> > Evaluation function & pruning is a pretty inefficient way of getting
> > somewhere. It's like a guy lost in the jungle without a compass or GPS. He
> > is trying to find an open spot in the jungle a mile away by looking at the
> > trees around him. When the trees look slightly "open-spottish-like", he
> > takes that direction.
> > Well that is silly. He'd better had taken a bearing beforehand and use a
> > compass!
> >    
>  
> > ...
> > Thanks in advance for your replies.
> >    
> Hi Frank, good to hear from you.  Yes, this is an old idea.  Do you
> have a new slant on it?
>
> Personally, I don't see any way to ignore search.  Right now all the
> good programs use search (which is certainly not a very good
> recommendation by itself, since even the "good" programs are pretty
> weak.)  But then you have to realize that good humans also use search
> (called reading) and you probably could not play a good game of go
> without it.
>
> Your analogy of the jungle sounds like what is called "hill climbing"
> in computer science, specifically AI techniques.  The idea being that
> if want to get to the highest possible spot in a vast terrain, but can
> only see 10 feet away, you will head for any spot higher than you are,
> even if it's just an ant hill.  And if you are already standing on an
> anthill and nothing higher is within 10 feet, you will be stuck in a
> "local maxima", with no motivation for exploring "lower" ground even
> though lower ground is the right path to eventually climbing the
> mountain.
>  
Maybe it makes sense to make a distinction between search and lookahead? 
I think a Go playing program can do without search, but not without 
lookahead. The difference between the two is that lookahead establishes 
a goal (the open spot in the forest) and tries to find the fastest way 
to get there. Search just tries to move randomly and see if it got 
closer to its goal. Search works for chess because the pieces act as 
landmarks and make it easy to see if you get closer to your goal. In Go, 
stones are like trees in a forest, they all look alike in any direction 
you move. (Although I don't want to take this analogy too seriously ;-))

What perplexes me though is Frank's sudden change of mind. Not so long 
ago patterns with search was the holy grail to solving Go. Although it 
never hurts to look at things from a different perspective once in a 
while of course.


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/