[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




> 4. The problem really is b).  As many posters have pointed out, there would
> seem to be absolutely no way to verify that a particular move was in fact
> produced by the program in question, as opposed to being input by the famous
> 7-dan human sitting next to the computer.

I  don't know how   many pointed this   out, but it's basically wrong.
With a properly constructed program I can  ALWAYS tell, with no doubt,
that a  program "would have" played  a particular move.  What  I can't
tell is WHO  else might have also  played the move.   It  may turn out
that many  moves would  also have been   played by  masters  (or other
programs) but that doesn't prove your  program couldn't have played it
too.

> There has been useful discussion on how this could be done, and I don't
> think anyone is willing to say that it is theoretically impossible to do
> this, but as even the people proposing this themselves have noted, it is not
> really a practical approach.


I would like to comment on the  issue of how practical it is or isn't.
In some of these discussion  terms like "impossible" have been bandied
about.   Even  saying "theoretically  possible"  makes  it sound  like
something that could only happen in someones science fiction fantasy.

But none of these terms are  really accurate.  More accurate is to say
it's  simply  "impractical."   It  doesn't  require any  kind  of  new
technology  or  clever  programming  or special  tricks.   It  doesn't
require special knowledge, any of  the programmers could do it if they
wanted to.

The reason it's  "impractical" is at least twofold.   First of all, it
would  require program  authors to  modify their  programs,  and some,
perhaps  even   most  of   the  programs  would   require  substantial
modifications.  That's  a LOT  to ask of  every Go programmer  just to
play in  internet tournaments.  Secondly, the  procedure for verifying
each game requires too much effort, it's work intensive you might say.
This would be such a big  hassle it would probably serve to discourage
tournaments, not encourage them.

I looked  at the  stuff you suggested  and it's quite  interesting.  I
would have  to think about  this some more,  but I think what  you are
proposing, reduces to pretty much the same ideas I presented, the main
difference being your stuff is more colorful and interesting.

Consider, you are basically suggesting  a type of "signature" based on
program output,  the same as  I am.  In  my suggestion, the  output is
just a  single move,  (or all  the moves taken  together) and  in your
suggestion the output is a  move and various state associated with the
production of  the move.  The problem and  the associated difficulties
in the implementation are not different, except that in your scheme it
requires even more effort on the programmer.  The tradeoff is that you
are rewarded with more confidence of authorship in your scheme.

It could be that I didn't  digest every point, so feel free to correct
any  misconceptions.  I  can't say  that I  completely  followed every
point you were making, some of it got fairly abstract.



Don