On Tue, Sep 7, 2010 at 12:09 PM, Dave Dyer <[email protected]> wrote:

>
> I take the observation that human players lose at first, then rapidly learn
> to win against the best programs to be evidence that the programs have
> systematic weaknesses.
>

All players have weaknesses of course.

The issue is that humans have a strength that no program (yet) possesses (at
least to any reasonable degree) and that is the ability to adapt.     If
there is a systematic weakness this is definitely it.

Humans are good at hiding their weakness, avoiding their opponents strengths
and exploiting the weaknesses of their opponents.   With humans that
probably tends to average out to an extent.

It's difficult to see how the much weaker classical programs differ in this
repsect.  Is there something about this that you are not telling us?     Was
the old many-faces not subject to this problem?


> The mono-culture that is developing, with all programs using variations on
> monte carlo, is just driving the herd (of programs) up a blind alley.
>

There are different ways to program GO but they will all have these same
basic weaknesses.    For the moment MC just has them less.    MC has a feel
for the game that I believe no other method has and more closely resembles
how humans play.   This is just my opinion of course.    By all means we
should explore other ways.

By the way,   I have seen this exact phenomenon in computer chess,  you
would buy a computer with an advertised ELO rating about the same as your
own,   and after a few games you would feel "contempt" for the program.
The feeling was exacerbated by the fact that if a computer was the same
basic strength, it probably played weaker in general but got a lot of free
games based on blunders you would make.   So after you learned how to play a
program it really felt like it would 200 ELO weaker or more (even though it
really wasn't.)

I think you will find that this ability to learn to beat a computer only
applies if you are relatively close to the strength of the program.   The
advice in the old days of computer chess was to buy a machine that was at
least 200 ELO stronger than you!

I have noticed over the past few years that you seem to have a real contempt
for MCTS programs and have often made disparaging comments that implies that
you know of a better way to proceed.    But you have never (to my knowledge)
layed out what way that is.




>
> _______________________________________________
> Computer-go mailing list
> [email protected]
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to