[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




   From: Dave Dyer <ddyer@xxxxxxxxxxxxxxxxx>

   "program" is a very difficult thing to nail down.  I can provide
   a trivial "program" with a checksum that will never change, but which
   can do anything I want by dynamically loading other DLLs, executing 
   other programs, using other data files, or opening other communication
   channels.  

   Packaging a complete set of bits, so it would be dropped into
   a freshly installed windoze machine, is really hard.  

Everthing is hard with  windows.  But no matter what  OS is used there
is potential  difficulty   nailing  down  the concept of   a  program.
Instead  of calling  it  a program,  we  should call  it  a "computing
system" because  there  are dependencies involving  hardware, software
and general machine state (like DLL's and such.)

Having said  that, however,  I have not  found it  impossible to write
software   that runs deterministically   on any computer that supports
some version  of Unix.  My chess  program  runs on Suns, PCs,  SGI and
Alpha boxes and plays exactly the  same on each platform under various
version of  unix, as long as  the hash tables are set   to be the same
size.  I did have to go to a certain amount  of trouble to ensure this
behavior, but in  almost every case where  they varied, I discovered a
BUG in my  program and  indeed correcting  this behavior forced  me to
find a lot of bugs.  You would be surprised.

I am in  complete agreement  with you, that  trying to  implement this
would probably force a  lot of particpants out  of the contest.   I am
probably not quite as skeptical as you and  don't yet consider it "out
of the question", but I am definitely skeptical!

Don


   I agree
   with the comment that this requirement would make contests a lot 
   simpler to judge by eliminating most of the participants.

   Practically speaking, I think going beyond official observers
   as authenticators is out of the question.