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

[Nick Wedd <nick@xxxxxxxxxxxxx>]

In message <Pine.LNX.4.61.0505111121150.26140@xxxxxxxxxxxxxxxxx>, Michal 
Bazynski <bazik@xxxxxxxxxxxxxxxxx> writes
> On Wed, 11 May 2005, Vlad Dumitrescu wrote:
>
> > If the answer is "to crown the best playing program", my feeling is that any
> > program should be allowed to enter, in any amount. I assume an
> > "all-play-all" tournament, so that no competitor is pushed down in a lesser
> > league just because there are many entrants.
> hmm i do not understand where the problem understanding why it's wrong 
> to have two same/similar programs comes from... to continue David's 
> example:
>
> assume MFoG and GNU can beat every other program out there 100% of the 
> time.
> assume MFoG can beat each GNU version 70% of the time.
> let's have 2 GNU's and one MFoG play in a tournament.
>
> what is MFoG's chance of winning? 70? nope, it has come down to 50%. 3 
> copies of GNU's and its down to 35%. and all that time MFoG is clearly 
> best - 70% winning chance or better against every other opponent.
Yes, it's clearly best to you, and to all readers of this mailing list. 
But it is no longer clearly best to the marketing guy in the Japanese 
company that David Fotland is negotiating with.

Ok, I don't even know if David is negotiating with any Japanese company. 
Let's consider a hypothetical, but entirely plausible, case.

<hypothetical>
Suppose I have written a good Go-playing engine.  I started doing this 
for fun, but my engine is good enough that I sell it commercially, I and 
my family now depend on it for our living.  Using the published results 
of past computer Go Tournament results, I have persuaded a Japanese 
company to pay me for it, put their own Japanese-language UI on it, and 
sell it in Japan.  The marketing guys in this company can't program and 
can't play Go, but they can read results tables;  and if they see that 
my program has lost to a rival they will start thinking, particularly if 
they can get that rival for free.

Now, I believe that my program is a couple of stones stronger than GNU 
Go.  I am happy to enter it for a tournament where it will be playing 
against GNU Go.  I am even happy to risk the occasional loss.  But if I 
am going to have to play multiple copies of GNU Go, it will be more than 
an occasional loss, and it is my livelihood that is threatened.  I am 
not going to enter a tournament under those conditions.
</hypothetical>

If a company executive, or indeed an ordinary member of the public who 
is deciding what program to acquire, reads a results table that begins
    1st    GNU Go version 3.8
    2nd    Many Faces of Go
they may not read any further.  Irrational of them, perhaps;  but this 
is what is likely to happen.

I would like commercial programs to enter these KGS tournaments.  But 
they are unlikely to do so unless they have their fair chance of 
winning.  This is what I am trying to arrange.

I am planning on having a "formal" division and a "casual" division. The 
formal division will only allow one copy of any one program, and will 
require entrants to state their real names, so that we know who to 
revile if they are found cheating.  The casual division will have few 
restrictions, it is entirely for fun.

Programs that qualify for the formal division will be permitted to enter 
both divisions (though as they will be played simultaneously, this may 
present problems for them).  I expect that commercial programs will only 
enter the formal division.  I hope that people deciding what program to 
buy will look at the results of the formal division, and ignore those of 
the casual division.

> what would make it fair is if MFoG could enter as many copies as 
> GNU-based folks can.
This is what I am proposing.  What do you think that number should be? I 
think the obvious answer is "1".

Nick

> (i am assuming with those calculations that the GNU version that gets 
> lucky against MFoG beats other GNU-s, so the real numbers are not as 
> bad (assuming different GNU's can't arrange the wins between them, but 
> that is a totally different topic))
--
Nick Wedd    nick@xxxxxxxxxxxxxxxxx
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/