Re: computer-go: A problem with understanding lookahead

["Mark Boon" <tesuji@xxxxxxxxx>]

Well Vincent, I may not be a strong Chess player, but I'm not really a
beginner either. It certainly looks like I know more about Chess than you
know about Go.

Given an average Chess position, if one side or the other suddenly loses a
piece (I mean a 'piece', not just a pawn) that will make a big difference. A
game deciding loss on every level but the very lowest ones. Positions where
you can lose a piece but the positional advantage balances the loss out are
relatively very rare compared to the positions where it would mean losing
the game. I challenge your statement that controlling the center is more
important than material. Give me your queen and I'd be happy to let you
control the center. Not to mention losing your king (mate plus one ply).

Evaluation in Go is not just calculating some influence from stones and
determine areas (territory), even though those operations are already way
more costly than the simple piece count in Chess. If you'd do that, you
would have to read the position up to a point way beyond where even very
weak players would have started passing. (Probably at least 50 more ply.)

I'm aware of course that a Chess program that only counts pieces is not very
strong, but still relatively strong compared to nowadays Go programs. Add a
few simple rules of thumb, like center control and pawn structures and it
gets you a long way towards an expert player. There's nothing so simple in
Go. Just to know which stones are dead or alive, even very simple cases, you
need to make a very sophisticated program to figure that information out.
And that doesn't have so much to do with board-size. If it were, it should
be easy to make a very strong 9x9 program, but it isn't. It's hard because
evaluation in Go is very, very difficult.

    Mark

----- Original Message -----
From: Vincent Diepeveen <diep@xxxxxxxxxxxxxxxxx>
To: <computer-go@xxxxxxxxxxxxxxxxx>
Sent: Friday, January 19, 2001 4:16 PM
Subject: Re: computer-go: A problem with understanding lookahead


> At 03:50 AM 1/19/01 +0100, you wrote:
> >
> >----- Original Message -----
> >From: Vincent Diepeveen <diep@xxxxxxxxxxxxxxxxx>
> >To: <computer-go@xxxxxxxxxxxxxxxxx>
> >Sent: Friday, January 19, 2001 1:24 AM
> >Subject: Re: computer-go: A problem with understanding lookahead
> >
> >
> >> At 11:18 PM 1/18/01 +0100, you wrote:
> >> >
> >> >----- Original Message -----
> >> >From: Don Dailey <drd@xxxxxxxxxxxxxxxxx>
> >> >To: <computer-go@xxxxxxxxxxxxxxxxx>
> >> >Sent: Thursday, January 18, 2001 10:02 PM
> >> >Subject: Re: computer-go: A problem with understanding lookahead
> >> >
> >> >
> >> >>
> >> >> You obviously are not familiar  with the computer chess world,
because
> >> >> "counting  material" is a  horrible evaluation  function and  does
not
> >> >> lead to "sophisticated positional judgement" in modern programs
> >> >
> >> >The qualification of 'horrible' is rather relative of course. There's
no
> >Go
> >> >program that has an evaluation that comes even close to the precision
of
> >> >counting material in Chess, and I don't think there will be for quite
> >some
> >> >time.
> >>
> >> I think this is not true Mark, of course you need to burn
> >> more system time for evaluation as the board is bigger,
> >> but i think the only difference is the branching factor.
> >>
> >
> >Well, we disagree on this point then. I think relatively speaking
currently
> >all go programs make evaluation errors on a regular basis that misjudge
the
> >position by a chess-piece or worse. I don't think this discrepancy is
> >covered by the deeper search simple-minded chess-programs do.
> >
> >    Mark
>
> The basic problem is that you look to chess as a beginner,
> for me GO is a very simple game. Just throw a few stones on the
> board calculate some area and group influences and that's it.
> O yeah, of course take care a group doesn't get hung!
>
> So the basic evaluation though burning more system time is
> not much difference from GO and chess.
>
> In chess a material only search will lose from any chessplayer
> who is playing in a chessclub.
>
> Material is an important factor, but more important is to
> develop near the center. Even a piece less in beginnersgames
> is not importantthen.
>
> Basically we talk about a real strong approach of go for your
> GO program versus a beginners approach for chess, whereas i
> have the same for go.
>
> My vision is that making a go program is dead simple, but that
> the only problem is the branching factor!
>
> You don't even need to take into account king safety
> for example, the most important and hardest to calculate
> factor of a chessprogram. In my program it can be worth
> 30 pawns of compensation!
>
> Some programs nowadays lose games to other programs and to
> mankind, BECAUSE they sacrafice a pawn for nothing,
> whereas in GO sacraficing a few stones is no big deal
> as you have plenty!
>
> Greetings,
> Vincent
>