Re: brute force and knowledge

[johnc@xxxxxxxxx (John Clarke)]

Hi

> 1) I said that go will never be solved by brute force : of course this is
> speculative, and I was referring to the high branching factor of go, nothing
> more. But I think erveryone would agree that if brute force is to be used in  
> go, then the evaluation fonction has to be *very* smart (may be using
> "internal regularities")

No, that's where I absolutely do not agree.  The entire point of brute force
is to apply some simple operation many times times - possibly in a small range
of variation.  But the aim is lots of simple things over and over.  Think
of the Vietnam war:  lots of guys in black pyjamas vs. a (relatively) small
number of high-tech bombers, helicopter gunships etc.  The Vietcong were
the ones employing brute force, the US were the ones trying to be clever.

> 2) how could you say that, in the end, human brain uses brute force ? Yes, it
> is
> highly parallel. But that's necessary, not enough, for the brain to be like a
> massive parallel computer. It might be so, but it might not : maybe there is
> some "global" complex structures and behaviour  that arise when you put
> together
> an enough number of basic devices (such as neurons), but maybe also there is a
> "soul" that is NOT localized in the brain and that need only a brain to
> express
> itself ?

Until Intel starts selling bottles of (souls / pixie dust / elvish magic) I'll
settle for Turing computable.

> 3) This topic is also higly philosophical. I have one opinion, you may have an
> other one, but for me the essence of go is in the question : why are human go
> experts so unable to express their knowledge ? 

Because it's mostly subconscious pattern matching.  The conscious part of your
mind is there to simulate a Turing (general purpose) computer which checks
the subconscious part's suggested moves.

<Stuff about board sizes snipped>

It's grossly unfair to compare a PC to a child since the child has so vastly
more processing capability.  Around 2020 PC's will reach human levels of
processing power.  The fact that alpha-beta works well on 7x7 boards is
a huge encouragement to the "simple, dumb and lots of it" school of
thought.

> 5) Furthermore, strong human players have all their own philosophy of what go
> is, and they know that their opponent has its too, and they play accordingly.
> The stones position on the board is not the only "input " for humans !

Every 9 dan pro always talks about playing against god - making the best move
you can, and ignoring what traps etc. you think you may be able to pull off.
Just make the best move.  Always.

They may not always achieve it, but that's what they aim at.  I've found it
a good discipline too and I'm only 4k.

> I do not think that computer go is hopeless, but I think we have a lot of
> progress to do in AI, philosophy and even maths before we are able to build a 
>
> 5 dan computer go player... Unless we find a breackthrough idea that use 
somehow
> the mathematical properties of go.

I think it only needs a lot of good engineering, no magic breakthroughs.

Regards
John Clarke
johnc@xxxxxxxxxxxxxxxxx