Judging programs from one example and random play

["Magnus Persson" <magnus.persson@xxxxxxxxxx>]

> Darren Cook wrote:
> 
> Kojima 9-dan did a commentary of the Many Faces/Young Leaf game and said he
> was very impressed with the way it ('it' being young leaf I think) made a
> group live. Something about he didn't think computers could make such good
> moves.

Just some randoms thoughts that are not so relevant, but I could not 
stop writing after i started:

It is probably too easy to make too positive conclusions 
from just one case in each game. I think one should expect any 
computer program  to make at least one strong move  in each game.

As an example I once made a program that played completely randomly, 
except that it did not fill in "solid eyes". I discovered that if you 
gave that program a free handicap about 100-120 stones it actually 
became a quite addictive variant of go. The nice thing with a random 
opponent is that you can never trick it or learn some systematic 
weakness. And after playing some game I got the feeling that at least 
once in every game, the random player played at that spot where I 
would play myself if I were black - almost as the computer had read 
my mind. 

But this is not strange after all since if the computer plays perhaps 
100 moves in a game like that, it would be very strange if not one 
or two moves would be really good.

I do not think that Kojima was wrong in his commentary since it is 
much harder to get a sequence of moves right just by coincidense , 
but I just wanted to point out this proverb:

"Even an  idiot can  play at least one strong move in each game..."

I do not know if it is useful but it could also be interpreted like 
this:

"Never expect a weak player to not find a winning move!"

The risc that a player slightly weaker than you find those moves are 
very high compared to idiots and  random player, and you cannot play 
moves that leaves serious weaknesses behind  that many times in a 
game and get away with it. Sometimes when I lose against "weaker" 
players it seems like I have violated that rule consistently 
throughout the whole game. 

Another interesting thought about a random player (about 100 kyu) is 
that it is possible to make programs that play even worse, even if 
you disallow "filling in of solid eyes". This means that 
random play cannot be considered the weakest possible, it really 
takes a little (but not much)  go-intelligence to play really bad! 
Also, a random player can in theory beat any professional, since 
nothing stops it from playing only good moves. But the probability of 
that to happen in a game is perhaps even lower than the probability 
of spontaneous creation of life in a soup of random chemicals. 
Current programs are perhaps so deterministic that a strong player 
can always force them into playing bad moves, and do not have a 
chance at all even in theory... or perhaps some programs do add a 
small amount of randomness in hope of a surprising win sometime?

- Magnus Persson 
--
Magnus Persson
Department of psychology, Uppsala University
Box 1225, SE-751 42, Sweden
Tel 018-471 2141 (work), 018-460264 (home)
magnus.persson@xxxxxxxxxxxxxxxxx
http://www.docs.uu.se/~magnuspe