Re: computer-go: A problem with understanding lookahead

[Vincent Diepeveen <diep@xxxxxxxxx>]

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. 

If one would use your evaluation and
search 12 ply in GO with of course extensions added for
some forced lines.

The first few chessprograms were also very easily to fool
by feeding them material for obvious positional and strategical
shots.

However real soon the selective tactical search depths of chessprograms
were like 12 to 15 ply.

Now if a go program would get in about all critical tactical 
lines such search depths and of course the forced lines even
deeper, then we can talk about a fair compare.

What you need to compare with your go program is a chessprogram
searching 2 ply or something, i think that's like a 1980
chessprogram (and sure not chess), which positionally is getting
beaten blindfolded by me.

In fact the last world championship a participant joined from
France who ONLY did a 12 ply tactical search without positional
evaluation at all.

During supper i beated this program blindfolded.

The reason why chessprograms with a 12 ply search play so strong is
because the center is in most games that get played so important
in chess. And with a 12 ply search you can very easily see the
tactics in that center.

Still todays top programs know very little from non-center things.

We also must take into account that the real strength of chessprograms
gets a lot overrated. Yes they can sometimes beat the strongest chessplayers,
but they can also lose matches from a big amateur player. However
as commercially you cannot win anything by organizing amateur matches,
those matches hardly get organized.

We only see blitz games a lot on the internet against chessprograms
which by the way in blitz get 8 ply or something, some even more.

A depth where we already are completely ignorant of what happens
if all go programs get positionally 8 plies and more.

I would suspect that if at 3 minutes for the whole game, which is
the most popular blitz level at the internet, if at that level
go programs get 8 plies *at least* that we would talk
about a major problem for the GO players too.

I'm losing nearly every game in blitz against chessprograms at 3 0
level.

However at 2 hours a game level i have still 100% against chessprograms,
and for now i don't see a big change there.

I'm just an amateur player, who is already happy if he
gets international master within 2 years of time. 
I have no idea what dan-level this represents, most likely
1 class lower as the professional dan levels, as i beat
*sometimes* professional chessplayers, but usually 
lose from them.

Of course no commercial company is going to challenge me nor
many other amateurs for a slow match. Suppose they lose!

But a 3 0 (3 minutes for whole game and no time added)
game against a professional go program getting each line 8 ply
positional or more i would actually bet that when in go about
the same energy is put as in computerchess, as i consider
most go programs are still searching with their extensions
at quite pathetic amateur levels (because they even post that
searching deeper is not working, so that means they have HUGE
BUGS in their program), that the go
players face big problems too.

Most likely exactly the same problems as chessplayers: you forfeit
in a won position against the program.

Vincent

> Mark