[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

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



David Fotland writes:
> 2) The trace is way too easy to fake.  For example, the program could 
> calculate the
> top 20 moves, and have the 7 dan choose between them.  The move chosen by the
> 7 dan has a valid trace of the logic the program used to create it.  Any moves
> that the program thinks are better than the 7 dan move can be excluded from 
> the trace,
> so that it looks like the computer correctly identified the best 
> move.  This 7-dan assisted
> program won't play at the 7 dan level, since often none of its moves
> will be the best move, but it will play much better than the unassisted 
> program.

Any trace generated by the program can contain secret encoded
information. Only if the program used by the referees to verify the
game record cannot exploit the encoded data would a trace be a
solution. In practice this is never the situation.

>From the suggestions thus far only the most trivial seems secure, and
doable in practice. Namely, if every program is deterministic, depend
only upon the previous moves and the initial state then (with some
further extra conditions) can we have a protocol which detects
cheaters with large probability. 

I do not know any way to loosen the requirements which would still
guarantee that cheating is not possible.

-- 
Mika Kojo