Re: Turing test - Deep Blue

[Vincent Diepeveen <diep@xxxxxxxxx>]

At 03:11 PM 5/26/98 -0400, you wrote:

>-David
> 
>>         I just want to comment that Vincent had one major clue that we
>> don't have when playing the Chess programs--Time scale.   Computers
>> often don't make obvious moves "immediately", and also make other kinds
>> of time blunders.  I especially noted this in the Kasparov--Deep Blue
>> match when the commentators were all commenting about the fact that 
>> Deep blue was taking minutes to do an absolutely obvious and 
>> critical move.

>I'm not sure how reliable a clue this is. Human go experts also often
>don't make the obvious move fast - like computers, they may be thinking
>not just about the next move, but the move after that :)

Yes he's also wrong for computerchess. Strong programs take back pieces
in fact quite quickly, that is generally easy to make.

Considering that there are 361 squares at a go board to occupy, and
considering that there are in chess 15 pieces to capture before you win
by mating the opponent, then taking back a queen which is forced
is the same like capturing a group of 90 stones, which threatens to make
a second eye.
( 361 * 10/(1*10+2*5+4*3+8*1) = 361 * 10/40 = roughly 361/4 = 90 )

That Deep Blue didn't do this categorizes its level. Kasparov's level was
below mine in those 6 games, except for game I. 

I'm amateur. Grandmasters who lose a game usually make about 2
obvious weak moves a game. Kasparov made in that match average 5 bad moves
a game. See for example analysis of GM Seirawan in ICCA.

Kasparov never explained why he forced that 2.5 - 2.5 still one game to go,
he only insulted the DB team from 'cheating'. Probably hiding his own level.

Enfin, i think this match isn't representive for how computers play.

some differences and reasons why IBM isn't representive for the 
computer chess community:

  - strong chess programs commercial versions do not crash when playing
    a game, deep blue crashed several times, but was allowed to play on.
  - strong chess programs don't have that big gap between knowledge and
    calculation power. Deep Blue's moves clearly show that it makes faults
    commercial programs already 5 years ago did not even make (bxg6? two
    times), yet we
    can be quite sure that it saw tactical even more than strong pc programs,
    which are already tactical strong enough compared to their positional
    and strategical play.
  - time division. DB searched exactly 3 minutes unless it got fail low.
    This is weird. That's the most primitive time division i know of except
    for a time division which doesn't have fail low research.
  - commercial chess programs are reasonably tested to very well tested.
    DB only played few games before playing Kasparov.
  - It is not possible to conduct more tests with DB, after IBM shares got up
    22% or something. They sold the huge computers it used, although Deep Blue
    is using hardware processors, which can only play chess and are not 
    general purpose.
  - it was a short match, just 6 games, 3 times with white, 3 times with
black.
    chessprograms who are playing on the internet or for the
    so called 'swedish list' play thousands of games, so their attitude 
    towards learning and book was different than IBM team.
  - DB played against one opponent, namely Kasparov. So they it didn't play
    a bunch of people.
  - big problem in deep blue project was of course parallelism and cpu design
    and how to integrate that, so communication between processors and
       chessprocessors and communication between several cpu.

    The cpu's were bye the way 0.6 micron technology, which is outdated and
    terrible cheap, the PR for the match costs were way way more. I couldn't
    even estimate what their propaganda department got. So that gives the
    last difference, PR was more important than the event itselve, which
    were just another 6 games against an opponent who this time clearly didn't
    want to win with big difference, and what happened the last game, we
    can only guess. Did Kasparov underestimate DB, did he not realize
    where computers play strong (tactics), or was it one big setup?

    Enfin, Kasparov gave the computer 9 stones ahead and yes the computer won.
    How surprisingly. At the internet chessprograms already win that kind
    of positions since many years.

    I think this is not the forum to discuss this too much further. 

Enfin, this all will give you the feeling that this is of no use for
the computer go world, unless you want to know how to not do it...
  
Greetings,
Vincent