Re: Turing test - Deep Blue

[Vincent Diepeveen <diep@xxxxxxxxx>]

At 12:15 PM 5/27/98 -0700, you wrote:
>Vincent wrote:
>> but did you count the number of ? moves of kasparov and all the
>> GM's calling it 'weak gm level'?
>
>I am not good enough at chess to judge DB or Kasparov. But 
>assuming weak GM level for DB rates it about 2400 ELO.
>That would be about 5dan (amateur) at Go!!!
>The best Go programs are 8-10kyu (real strength) at best.

But it makes it clear that the IBM team did not follow the right road.
Instead of making it faster and faster, they should have done other things.

Not that i criticize its programmers. I criticize the IBM propagandamachine
which brings it as the 8th 9th and 10th world miracle, which is not true.

>I think even if DB is not better than the best human players
>it is at least close to them. Anybody who plays Go seriously
>should beat the best programs after 1-2 years ...

It is not near close to it.
It plays awfull chess.
Do not believe the IBM fairy tail. They don't even wanna play
with that cpu, although its 0.6 micron means it's very very cheap
to reproduce and attach to a PC. 

>> Giving up 1 move in chess is sometimes possible. Giving up 2 moves is in
the 
>> short term losing. especially if you also move the pieces to a worse place,
>> which happens.
>we cannot compare 'moves' in Go and Chess. They are way too different.
>
>>Here a chess game from my own program against crafty.
>>...
>>So giving up many many stones is easy.
>
>I do not get your point. Giving up stone/moves is only easy, if you
>play somebody much weaker than yourself.

This is 6x hardware difference, and yet giving up bunches of stones doesn't
win for the program which is way way faster. In blitz computer
speed usually means you're dead meat if both programs are reasonably well.

So the point is clearly proven now: just getting faster is not the right
way to become invincible. It doesn't make you weaker of course,
so if you can get that speed, then try to get it.

>> Most questions asked here and a lot of discussions in this computer-go
group
>> are kind of naive from my viewpoint, as these discussions have been done 
>> years ago in computer chess.
>
>well, at computer-chess you can solve many problems with a deep
>(almost full-width) search, that is completely unreasonable for 
>computer-go, even if you put in millions of hardware dollars.
>
>> So with a smart selective way of searching ...
>> For example, if i allow my program to search for a night, then it gets
>> selectively a search depth which can be compared to a brute force search
>> depth of about 18-20 ply, extending some local tactical variations up to
>> 56 ply

>this is exactly the point! You claim to have found a way to search
>18 ply in chess, good. But in Go 18 ply is a) much less valuable
>b) much harder to do.
>And what enables you to do a deep search in chess (tree pruning) is
>also harder in Go, since there are so many strategical, long term
>effects.

I only want to point out that there has been made a lot of progression,
and you clearly do not get my point.

>> Would huge main search depth lead to closing that gap in the same way?
>no, not with 18 ply. 
>A brute-force 18-ply go program would not play at weak GM level,
>but you couldn't run such a program anyway.
>(200 billon/trillion/etc positions/sec is pathetic if you want to 
>search 200^18 ~ 10^41 :-) 
>Christopoh

Few years ago in chess search depths over 15 ply seemed impossible,
but right now i get way more at analysis level (and between
10 and 13 ply search at tournament level) and deep blue gets 15 
in tournament level. In the past branching factor was said would never
get lower than 5, as the average number of moves is 30, and square root
of never get lower than 5 with transposition tables. Yet todays chess
programs have branching factors at huge depths between 2 and 3.

Now my question was: suppose you would get such a depth with go,
how strong do you think it would play, yet you do not answer my question,
but instead with the knowledge of a child judge about something
you do not know a thing about, and even doubt it without answerring
the first question.
  
>

Vincent Diepeveen