Re: [computer-go] future KGS Computer Go Tournaments - two sections?

[William Harold Newman <william.newman@xxxxxxxxxxx>]

On Tue, May 10, 2005 at 10:04:16PM -0400, Compgo123@xxxxxxxxxxxxxxxxx wrote:
> In a message dated 5/10/05 3:21:25 AM Pacific Daylight Time, 
> tesujisoftware@xxxxxxxxxxxxxxxxx writes:
> 
> > The question of how to deal with derivative work has always been
> > difficult.

It's really quite a nasty problem even when human competitiveness and
misbehavior aren't present to complicate it. Determining "how similar"
two programs are is unlikely to be tractable when just deciding
whether two programs are "the same" is intractable. In languages which
support "functional programming" styles, where it's easy to whip up
function objects on the fly, it's natural to want to know whether two
functions are logically equivalent (not just whether they happen to be
represented by exactly the same object stored at 0x772f64a0 in main
memory); too bad, then, that that question contains the halting
problem.
 
> One solution is to charge $30 for entrance fee. I wish 1 million GnuGo 
> derivatives  would participate.

This part seems elegant and stable, and I think it is a useful
fallback to keep in mind for when and if people get very serious about
a tournament to determine which is best, as opposed to a sportsmanlike
social exercise. But I think to make it truly stable, you end up
having to let the pot involved get arbitrarily large (and, probably,
giving the money back to the winner of the tournament), and once
you're running a tournament with thousands or tens of thousands of
dollars involved, various other issues become much more deadly serious
and thus generally more difficult. It's probably worth trying 
informal cooperation first...
 
> One the other hand no matter how many copies of the same program
> participate, it doesn't really change the odds of wining. Unless
> your program has the exactly the same playing strength as GnuGo. In
> this case you may as well join the crowd and use a copy of GnuGo to
> particite, beacsue it makes no difference.

As David Fotland and others have pointed out, this part is wrong
unless "your program" can beat GnuGo essentially 100% of the time,
which is not the usual situation. 

Furthermore, even if "your program" had sufficiently close to 100% win
ratio -- perhaps because it's The Turk Mark 2006 with an insei inside?
-- you could still have problems unless each game costs you
essentially nothing. Otherwise, having to play 463 duplicates of one
opponent, instead of playing that opponent just once, could be a
significant problem. For example, if you had to pay your insei a
salary:-) or, more likely, you had the SlugGo problem of tying up so
much computer hardware that having to do it for an arbitrarily long
time was not convenient. One reason that I think the entrance fee idea
is elegant is that it seems to cover this problem: just don't enter
the tournament with any system whose operating costs exceed $30 per
game.

-- 
William Harold Newman <william.newman@xxxxxxxxxxxxxxxxx>
PGP key fingerprint 85 CE 1C BA 79 8D 51 8C  B9 25 FB EE E0 C3 E5 7C
Ubi saeva indignatio ulterius cor lacerare nequit. -- Jonathan Swift's epitaph
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/