Re: Turing test

[Frederic Boutet <boutet@xxxxxxxxxxxxxxxx>]

At 16:31 20/05/1998 +0900, vous avez écrit:
>Here is a little puzzle for you all:
>a game record between a person and a computer program. Now which is which,
>and why? Please try to form an opinion by looking only at the moves played,
>even if you happen to know the game already. If you were the human playing
>or if you were writing the program, what would you do differently?

Just as Turing, I am often surprised by machines. 
I've been really surprised by the puzzle: I thought that it would be
easy to determine who was the computer. And it was really not the case for me!
So I read again "Computing Machinery and Intelligence", written by Turing
and published by the journal "Mind" in 1950.

A very interesting argument of Turing is the answer he gives to 
those who think that it is not possible to build a set of rules which 
would describe all a human should do in normal circumstances. 
He rewrites this idea by:
if every man had a set of rules upon which he could organize his life, 
he/she would not be superior to any machine. Since these rules do not exist, 
man cannot be a machine.
Turing answered that he did not know any circumstances in which scientists had
been able to say: "stop, we have searched enough; such rules describing
a human behavior do not exist".
This was true in 1950. Is it still true now ? I think the answer is yes.
Biologists only tend to isolate genetic diseases in ADN but just cannot say that
it is impossible to predict the general rules that govern a man or a
woman upon his/her genetic identity (a recent movie shows a world, "Gattaca", 
where it is possible; great! makes us think a lot!). People would have stopped 
researches in genetics otherwise.

Finally, Turing says that he wrote a little program which, given a number
composed by 16 digits, returns another number in two seconds. Then, he
challenges everyone to predict the answer of the program, with an input that 
has never been tested before!

Of course, this is not sufficient to say that a go-program is "intelligent".
Even if you cannot predict its behavior, you can adapt your strategy. 
But to be forced to play just like against another human is already 
a great advance!

So the fact that the match between "white is the computer" and "black is
the computer" is close is probably a sign that we try to find systematic
behaviors that would caracterize more a computer than a human. But this is
very comon: we you play a great number of games with a friend, and you realize 
that he/she plays in a systematic way, you will consider yourself stronger. I 
don't think it is conceptually typical to computers.

What is still more typical of computer is passive learning, that is learning
by waiting for someone to increase the size of its joseki database. If 
go-programs could include patterns simply by observing proper games,
on the net for example, they could become stronger more rapidely. The
problem on the net is to FIND proper games :-)
This makes me asking: "what is a proper game ?". How can a program comment
a game and say it is proper or not ? How can a program recognize one of
its own games ?

fred b