computer-go: Authenticating the identity of a remote go-playing computer program

[John Aspinall <jgaspinall@xxxxxxxxxxxxx>]

I too believe it is next to impossible to guarantee the identity of a
remote program.  In messaging systems, the environment where you compose
the message is "open", and only the transport of the message must be
closed against eavesdropping, spoofing, etc.  This is very different
from trying to close a boundary around a program that we (the tournament
organizers, let's say) can't even examine.  Insisting that the program
run within a closed sandbox such as the Java VM is not a complete
solution, unless we can control the sandbox.

Instead, let's do some social engineering.  Let's devise a tournament
format that
 - uses the net to its maximal advantage, but verifies (in the closed
environment of an isolated machine) the final contestants;
 - discourages cheating, rather than rewarding it.

Consider a 3-tier tournament:
 Tier 1 : Over the net, open to anyone, no verification, honor system. 
Winners proceed to...
 Tier 2 : Over the net, with verification by observers.  Observers
trained by e.g. James Randi in skills for detecting cheating. 
(Possibly) entrants post a monetary bond, refundable on the positive
report of the observer.  Note that a cheating contestant in tier 1, gets
no tangible reward for winning in tier 1.  Winners proceed to ...
 Tier 3 : Central location.  Clean (i.e. RF-free) machine room.  Strict
protocol for operators, etc.

Now this is a lot of infrastructure, possibly too heavyweight for some. 
I throw it out as a possibility for a "serious enough" tournament.