Re: Go program strength

[Elmer Elevator <bobmer@xxxxxxxxxxx>]

Kees van de Merwe wrote:

> What I remember is that Kasparov complained that the programmer's changed
> the algorithms of Deep Blue in order to change the playing style of Deep
> Blue itself. This is completely different way of upgrading the strength of
> your program, than the "tweaking" that is discussed in this mailinglist. But
> changing the playing style is a effective method, because now the human
> can't use his/her knowledge of the program to exploit its weakenesses. I
> agree with Kasparov that this is unfair, because this is something a human
> will never do, simply because he can't.
>
> just my $0.02
>
> Kees van de Merwe

Well, in chess a player CAN do this. There are many famous master games in which
a player, well known for a particular group of openings and defenses, and very
familiar with his opponent's handling of these openings and defenses, surprised
his opponent by venturing into unfamiliar early combinations (which, of course,
the player had previously tried to analyze).

A player would do this not simply for psychological surprise, but because he/she
had gauged that, in the likely openings he/she and the foe were likely to
select, their strengths and experiences were very comparable, and thus the
outcomes in these familiar strategies were comparable and uncertain.

It's risky, but players can indeed change strategy, and those who do have the
advantage of having prepared for it; the foe has to ad lib and improvise his/her
response as he/her goes along. The foe will not get a chance to step back and
analyze the novel territory in depth until a mid-game postponement, if the game
lasts that long; and by that time the damage may already be irreparable.

Your post might possibly illuminate a fundamental shortcoming of computer Go
programs -- however good they are to date, each of them is designed around a
single or a limited and related repertoire of strategies. Experienced Go masters
probably have more intellectual leeway to jiggle strategies right and left on
the fly if they intuitively detect that their foe (carbon or silicon) is
somewhat rigid and dogmatic, thus plunging the foe into unfamiliar waters. Lord
knows how easy it is to play a  move that bewilders me.

Finally, back to Fair and Unfair ... a tendency to say "Well, that's not fair"
can be countered with the argument that the loser is looking for excuses. How
often does the winner say that the loser did something unfair? How often does a
winning jockey lodge a complaint against a losing jockey?

Bob Merkin