RE: computer-go: prevent cheating

[David Fotland <fotland@xxxxxxxxxxxxxxx>]

1) Very few will enter a competition where they have to make source
public.  There is too much money at stake.

2) The Ing competition, with the $1M prize, does pay partial plane
fare for the top programs.  So there alreaady is a competition
like you describe.

3) Whoever wants to sponsor the competition can make whatever rules
they like, so it doesn't matter much what we discuss here unless one
of us plans to make a contest :)  If you want to attract entrants,
make the rules reasonable.

David

At 11:27 AM 10/20/99 -0400, Thomas Johnson wrote:
>I've been following this thread about having a competition.  There's a couple
>points noone's bringing up yet.
>
>1) If someone's go program is capable of standing it's own against a
>professional dan player, I think that's a suitably big step where we all
>would like to know how that person did it.  To this end, the requirements
>shouldn't be framed around using standard platforms or what not.  Before you
>can receive any substantial award, you should have to make public your source
>code (under suitable copyrighting of course) and produce a white paper
>describing how it works.  Afterall aren't we all here to share knowledge and
>make go fun for everyone and all that?  I can guarantee you that if you
write a
>program that beats a professional dan player and release the source code for
>it, if it's a hoax, some one will figure it out real quick.  If you don't
want
>to have to eventually share your source code, don't enter the competition.
>
>2) If the big award is $1 million, and you're capable of raising a trust fund
>on the order of $2 million, why is it so important to save on airplane
tickets
>for the chief competitors once a year?  You can play placement competitions
>throughout the year on IGS or whatever which don't receive awards, the best
>"x" players plus a few admin people then get free transport to an annual
>convention which showcases the programs.  If you have, say, 4 players
plus, say,
>4 volunteer/low-pay admin., that'd run you probably less than $20,000 a year
>(incl. convention room) if the competition is only 1 day and you look for
deals
>on rooms and all.  That's only 1% of your $2 million trust fund.  You could
>probably even splurge to pay for a pro dan to show up and play the champion
>computer player without problems.  Just a plain money market account
earning 3%
>a year will give you enough to do that.
>
>At this point, the individuals get to use platforms of their choice
>which are linked on an internal network to a go-server with no external
>access.  Obviously we would want the machines suitably checked so we're
certain
>there isn't any radio communication or some other such stuff.  (And we'd also
>want to make sure we didn't have a 286 playing against a power pc, for
>example.)  If it's confirmed that only the box is making the moves, it
doesn't
>matter how it decides to make them.  Why would we care that both programs use
>the same random seed or whatever?  As long as the box is completely isolated
>from everything except the go server, you can be fairly certain that there's
>some kind of program running on it that knows how to play go.  It's at that
>point that you weed out the fraud.  And again, the winner has to supply
source
>code.  There are plenty of people out here who would put it to the test. You
>could even offer a small reward for proving someone's program is fraudulent. 
>Kind of like spawning off bounty hunters.
>
>The only question then is how to raise the $2 million and who to entrust
it to?
>
>Anyways, any thoughts on those points?
>
>-Tom
>
>---
>Thomas P. Johnson
>Strategic Forecasting                       ph:  (607) 266-7636
>Box 1034 Langmuir Lab                       fax: (607) 257-4279
>95 Brown Rd.                  tjohnson@xxxxxxxxxxxxxxxxx
>Ithaca, NY 14850            http://www.strategicforecasting.com
>
>