RE: Turing test

[Vincent Diepeveen <diep@xxxxxxxxx>]

At 11:19 AM 5/26/98 -0700, you wrote:
>
>	[Scott Dossey]  Vincent Diepeveen had the following to say about
>the Turing Test: 
>> One game is not representing. We need more games.
>> Also when playing a weak human, especially beginners, then you cannot
>> conclude a thing from it, because in contradiction to experienced
>> players
>> beginners simply lack lots of knowledge, and therefore could be just
>> as
>> 'intelligent' as the program. Also beginners play very inconsistent. 
>> 
>> I've passed this experiment already myself in computer chess.
>> At the internet chess server i play regurarly people (guests) who
>> don't
>> carry a name. I picked all computers out of it within 3 quick games.
>> Some of the programs i could even name.
>> 
>	[Scott Dossey]  
>	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.

Yes this is true too, most programs carry a (C), so when they match you
you already know you're playing a program. 

I'm now talking about the fact that on this chess server (telnet
chess.lm.com), where thousands log on, some of them some log on 
anonymeously and as i'm not one of the worst players and also in my
finger carrying who i am (programmer of Diep bla bla) they
challenge me and enter the moves BY HAND. so i'm not talking about
the unattended computers, which play their openingsmoves within a second,
which of course alreadygives you at once an indication,
but i'm talking about the what i call 'cheaters'. People not wanting
to know me they're using a computer.

Now first game it goes like this: i of course don't doubt them,
and play like i'm playing a human. This means we get tactics at the
board, and as we all know in chess computers are tactically stronger
than amateur players in blitz, and then this game i'll see that this 
opponent is tactical superior to me, and i usually lose that game
('computer shock' as described by Robert Hyatt). I already have then
categorized my opponent as being tactical powerful, and some moves, which i
think myself
positional weak i wonder why my opponent made them.

Now it's a known fact that Europe is the strongest chess continent on
earth (especially Russia). So most positional knowledge is over there.
Now this means that there are very very talented and strong players
over the earth living at all continents which play positional at a low
level, but tactical are way better than positional.

So i still don't know whether i'm playing a program unless game was very
long and there was plenty of room for the program to make positional errors
(meaning that the most logical 'human' moves are not logical from tactical
viewpoint or from the little knowledge viewpoint of the computer).

Enfin i lose the first game usually quickly, and might excuse to my
opponent for my poor level saying: "sorry i'm still getting awake".

Asuming i lost it quickly i still don't have enough info.
Game II then becomes more interesting, because i KNOW i'm playing
against a tactical very strong opponent.

So i tactically carefully watch my moves. This usually means that
i'm playing myself a little bit less positional and more tactical.
So my positional level suffers under my wish to play tactical stronger
than the first game. This undoubtly leads to an openingsposition and
position after that which is not quickly winnable by the opponent.
It might have of course advantage (or from GM viewpoint: the win might
be strategically easy), but it'll never win it quickly. 

This means that the opponent MUST make choices. Now if i'm playing one
of the weaker programs then i see directly that i'm playing a program and
not a human. When playing against one of the strongest programs on this
planets (so we only have 2 handful of programs left then), then it might
play this game very well so the number of 'computer moves' (semi-best 
moves which computer prefers where humans prefer other,
or simply positional horrible moves which are not
very much made by strong humans but might be made by programs) might
be still limited. 

However assuming i didn't play like an idiot, then it's impossible that
my opponent played perfectly, so if it's a computer then there are a number
of bad moves made by it which i can analyze, and as i was very involved
in the game i already did that.

So if i play a computer then i already suspect that after few games.
What then happens is that i am gonna test it. All programs aren't intelligent,
and one of the things that happens to idiots is that they don't have an
overview of the whole picture, if we see overview as the root of all
decisions that you make.

Enfin, certain type of positions (closed positions) are simply next 30
years way too hard for chessprograms, and the good thing for mankind is
that you can always get a closed position from the openings position.

One of the main reasons why closed positions are so hard is because you have
so many different choices in these positions. Bad moves aren't always
pruned away because there is usually no tactical threat for it. Also
heuristics
for playing closed positions are different from the normal open or half open
positions.

Enfin, now i see within 15 moves whether i play a program instead of human.

It's interesting to know about those closed positions that younger players
are also quite horrible in them, which supports the opinion that knowledge
is one of the most important things when playing board games.

The second conclusion is that i am LEARNING from the games that i'm playing
a program.

It's not like: here is a dataset, conclude from the dataset whether it's
a program. 

No i can in fact INTERROGATE my opponent by playing the 3d game an
anti-computer opening. 

In fact knowing that i'm playing an anonymous person, and knowing that i 
always ask my opponent's name, i usually already know the second game 
for 99% sure whether i play a computer. The sureness is very much depending
on the type of positions i played. If i lose within 20 moves, then everyone
could have won like that. Just one tactical strong move doesn't 
say whether it's obvious. I KNOW where computer play horrible.
I KNOW where they might play brilliant. 

>From a human you don't KNOW where they play strong or weak, unless you know
them. From a stronger human opponent you EXPECT to posses the same
capabilities, but this might not be the case.

The reason why computers are so easily recognizable in chess is mainly
because they don't learn, they don't fear, and what's in their range they
always perform without huge failures, and all of this gets done with the
knowledge and logic of a young child.

More interesting than this subject is next subject. Now that computers
at certain terrains in chess (tactics) are stronger than most 
humans are, how to catch that human who cheats; he is playing himselve,
but is what we call 'blunderchecking' his moves. So he's checking whether
his moves don't lose obvious or directly. So in fact he is an android.
How to catch the android?

I mean i wouldn't know how to catch them. I have seen and played certain
opponents from which in the end i suspected them to do this.

Note that for 99% of the games these androids aren't a thread.
to blundercheck your game you need to do some stuff in a different window
with your program. This eats up time. It appeared at the Paris world champion
blitz for computers that i'm one of the faster operators in this world.

I needed 1 - 2 minutes a game just to enter the moves op the 
opponent at the keyboard and then move them at the board. 

At internet when cheating you need to do more than that. You cannot perform
like a zombie, because then you're a true computer then which gets easily 
catched, when playing the wrong persons. also you need to keep track
of another window. This means you lose simply some more seconds a move.
Then you need to read the output of the computer and to see whether you like
it. This eats up seconds. I would guess at least 6 seconds a move for the
quick androids. Then you still need to move it with the mouse at the board.
Now assume 60 moves at least a game. This eats 6 minutes of the androids
own time just for administration.

Most games played at the internet by the class A-level and above (so
the interesting opponents for androids) are between 1 minute and 5 minutes 
for the whole game a person.

So again time is favouring mankind!


Vincent Diepeveen









Now after that game if i play a program then i 

>	-Scott   
>
>