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

[Heikki Levanto <heikki@xxxxxx>]

Steve Korfhage <korfhage@xxxxxxxxxxxxxxxxx> wrote:


> Couldn't this be done by using a device driver which controls and monitors 
> hardware in the computer? (mouse, keyboard, and other ports of 
> communication)  Maybe this requires a standard hardware configuration, 
> where the driver won't even load if certain hardware (like network cards 
> and keyboards) are present.  

This is a whole new can of worms, and hardly practical. 

To begin with, this would require only a limited few operating systems to be
allowed. Now it may not sound like there would be all too many to worry
about, but for example with Linux, it is no problem to optimize the
configuration of your kernel for your application. Also, who will take on
the job of verifying the [custom?] drivers loaded into my system, or the
whole OS kernel, for that sake? This could be as big a job as writing the
competing program!

if one would want to go this way, it might be possible to insist that all
programs participate on given hardware, where all connections to the machine
would be limited (or monitored) during the competition. Let 'em install
what they want, but once the program starts running, do not allow *any*
input from *any* humans - only (say) a modem protocol for playing the game -
against a central game-control machine. 

This would disable the most obvious cheats of the type where a [presumably
stronger] human would feed information to the program.  It would not
helpanything against the [more likely?] cheats where someone would pass
another program as his own creation.


Maybe the time will come when computer go tournaments will have to be split
into different categories. Like in car racing, there are classes where the
cars have to be sold at a fixed low price after the race, to discourage too
bold use of advanced technology, and to keep the focus in driving skills.
And there are classes with weight limits and other technical regulations.
And there are classes where everything goes...

Personally I would be happy with a tournament form where the full source
code would have to be disclosed just before the tournament, so that all
would be able to verify that the programs do not cheat, and to learn from
each other. This would allow the competitors to learn all the tricks, and
copy them for the next years tournament, so any clever idea would only last
one year. But in the long run, programs willing to participate in this sort
of tournament system would be developing faster, and presumably get better
than hidden and closed systems. But this is my personal preference, and I
admit it may not fit well in the world of commercial programs competing of
market shares. That is why I am assisting GnuGo instead of trying to write
my own commercial product...



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