[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





   I tend to agree with the opinions already expressed here, that there is no
   cryptographical way to make sure the participants are actually programs. I
   think the most obvious solution is the one quoted in the original question:
   Make sure that
     > the stakes are not high enough to motivate anyone to cheat. 


Hi Heikki,

Read my reply to Dave Dyer, because I changed my mind about this after
having giving it some more thought.  Earlier I  would have agreed with
you.  My new opinion hinges on the  concept of viewing  each move of a
given program  as  a weak   signature and   viewing  many moves  taken
together as  a very  strong signature.  I   think this is  a perfectly
valid notion,   however there is  lots of  room  for disagreement.  In
practice, I think you will find that no  2 programs are very likely to
play the same exact   set of moves  in a  game.  To  me this is   very
similar to  saying   md5 might  produce a  few   similar bits   on two
different keys, but extremely unlikely to produce 128 bits exactly the
same.


Don
 




   Requiring the participating programs to be submittend to the organizers will
   help a lot, but as long as we allow any randomness in the play, we can not
   demand that the programs will later reproduce every one of the moves. Thus
   there is still room for a human supervisor to sneak some advice to the
   program (through a file, socket, or what ever), perhaps to propose one or
   two good moves - or even better, veto a few really big blunders.

   Of course these considerations apply to "physical" tournaments as well. It
   may help to require impartial operators for all programs, but that opens up
   a new can of worms, especially with tricky interfaces and responsibility for
   errors.

   Of course, if programs want, they can show "good faith" by explaining how
   they reached a certain conclusion, and by a controlled use of randomness.
   Both features sound very useful for anyone developing programs, so requiring
   them should not be that much of an extra hazzle. But then again, we never
   require such display from human players, who also could be cheating by
   having a professional walk by and shake his head...


   - Heikki


   -- 
   Heikki Levanto     LSD Levanto Software Development   heikki@xxxxxxxxxxxxxxxxx
		  "In Murphy we Turst"